PARAMETER TUNING OF AN INDOOR POSITIONING SYSTEM AND THE APPLICATION FOR LAMENESS CASE STUDIES IN GROUP-HOUSED SOWS Aantal woorden: 26.226 Kevin Syryn Stamnummer: 01270040 Promotor: Prof. dr. ir. Dirk Fremaut Tutors : Ing. Katrijn Ingels, Dr. ir. Jarissa Maselyne, M.Sc. Shaojie Zhuang Masterproef voorgelegd voor het behalen van de graad master in de richting Master of Science in de biowetenschappen: land- en tuinbouwkunde Academiejaar: 2017 - 2018
Transcript
PARAMETER TUNING OF AN INDOOR
POSITIONING SYSTEM AND THE
APPLICATION FOR LAMENESS CASE
STUDIES IN GROUP-HOUSED SOWS
Aantal woorden: 26.226
Kevin Syryn Stamnummer: 01270040
Promotor: Prof. dr. ir. Dirk Fremaut
Tutors : Ing. Katrijn Ingels, Dr. ir. Jarissa Maselyne, M.Sc. Shaojie Zhuang
Masterproef voorgelegd voor het behalen van de graad master in de richting Master of Science in de
biowetenschappen: land- en tuinbouwkunde
Academiejaar: 2017 - 2018
PARAMETER TUNING OF AN INDOOR
POSITIONING SYSTEM AND THE
APPLICATION FOR LAMENESS CASE
STUDIES IN GROUP-HOUSED SOWS
Aantal woorden: 26.226
Kevin Syryn Stamnummer: 01270040
Promotor: Prof. dr. ir. Dirk Fremaut
Tutors: Ing. Katrijn Ingels, Dr. Ir. Jarissa Maselyne, M.Sc. Shaojie Zhuang
Masterproef voorgelegd voor het behalen van de graad master in de richting Master of Science in de
biowetenschappen: land- en tuinbouwkunde
Academiejaar: 2017 - 2018
I
Copyright protection and confidentiality
The author and the promoter give the permission to use this thesis for consultation and to copy
parts of it for personal use. Every other use is subject to the copyright laws, more specifically
the source must be extensively specified when using the results from this thesis.
De auteur en de promotor geven de toelating deze scriptie voor consultatie beschikbaar te
stellen en delen van de scriptie te kopiëren voor persoonlijk gebruik. Elk ander gebruik valt
onder de beperkingen van het auteursrecht, in het bijzonder met betrekking tot de verplichting
de bron uitdrukkelijk te vermelden bij het aanhalen van resultaten uit deze scriptie.
Datum:
Handtekening:
Promotor: Prof. dr. ir. Dirk Fremaut Auteur: Kevin Syryn
II
Acknowledgements
I would like to thank some people who cooperated with the thesis.
First of all, a special thanks goes to the institution for agricultural, fisheries and research (ILVO)
for providing me the necessary tools needed for this study. The ILVO let me use their gestation
barn equipped with a prototype indoor positioning system for research purposes.
Furthermore, I would like to thank my supervisor Prof. dr. ir. Dirk Fremaut and co-promotor ir.
Katrijn Ingels for assisting me with the thesis at the university.
More acknowledgements go to the co-promotors at the ILVO, namely dr. ir. Jarissa Maselyne
and especially M.Sc Shaojie Zhuang for structural and statistical support which I needed to
accomplish the thesis. Thanks goes to Ir. Olga Szczodry for helping me preparing the sows
needed for the practical research.
And last but not least I would like to thank all other people who were directly or indirectly
involved with the thesis.
III
Abstract
Nederlandse versie
De studie onderzocht de integratie van een indoor positioneringssysteem voor praktische
toepassingen in een drachtstal met in groep gehuisde zeugen. Het systeem is gebaseerd op
tijd tot aankomst berekeningen met ultra-wideband signalen. Door stijgende aandacht voor
diervriendelijkheid in de varkenshouderij, meer specifiek kreupelheid, werd het
positioneringsysteem gebruikt om de niet-kreupele en kreupele zeugen met elkaar te
vergelijken. De sterkte van het integratie was gebaseerd op enkele nauwkeurigheidstesten,
waaruit vervolgens een positie gerelateerde algoritme werd ontwikkeld. De algoritme werd
gebruikt in combinatie met een bilaterale filter, om ruis in de data (ongewilde en abnormale
waarden) te verminderen. De filter was afhankelijk van drie parameters, namelijk de
vensterbreedte en afvlakparameters σd (dimensie parameter) en σr (omvang parameter). De
gebruikte instelling was respectievelijk 5 , 1,5 en 0,8. De parameters werden visueel
geanalyseerd op basis van het onderling vergelijken van de gefilterde gewandelde trajecten
van de zeugen. Voor de kreupelheidstest werden drie zeugen getest. Elk herhaalde het zelfde
parcours drie maal (vooruitgaande beweging in het centrale gangpad van de stal). De zeugen
werden vergeleken op basis van de snelheid en de afwijking van de X-coördinaten. De
gemiddelde snelheid van de niet-kreupele, voor-kreupele en achter-kreupele zeug was
respectievelijk 1,205 m/s , 0,862 m/s en 0,734 m/s en toont een significant verschil tussen
kreupel en niet-kreupel aan. De afwijkingen van de gewandelde trajecten vertonen significant
verschil met een stijgend aantal herhalingen. Dit suggereert dat, afhankelijk van de
kreupelheidstoestand, de zeug beïnvloedt zou zijn door moeheid, pijn en vermoeilijkt wandelen
Time based methods are also known as distance based methods. Three terms are used in
time based positioning systems, each referring to a way of position determination: lateration,
trilateration and multilateration. These methods determine the position by measuring its
distance from multiple reference points. The prefixes “tri-“ and “multi-“ describe respectively
three and more than three fixed points are necessary to determine a position. Time of arrival
(ToA), time difference of arrival (TDoA) and round-trip time of flight (RtoF) are examples of
propagation-time systems (Farid et al, 2013; Zhang et al, 2010).
Time of Arrival
ToA or Time of Flight (ToF) is a technique based on the time required for a radio signal to
travel from a transmitter (mostly a mobile device) to several receivers (base stations). The
distance between the receiver and the transmitter is calculated based on the transmission time
and the corresponding speed of the signal. Next, as shown in figure 8, for each receiver a circle
is constructed, with centre the station and radius (𝑅1−3) the calculated distance, and the
transmitter is localized via the intersecting point or region of these circles. The principle of ToA
is shown in figure 8. Note that this intersecting point can vary due to several errors (Farid et
al, 2013; Lanzisera et al, 2006; Yassin et al, 2017).
Synchronized clocks between transmitter and receiver are essential for an optimal working
system. As will be explained further, ultra-wide band (UWB) uses ToA and is less susceptible
to multi-path effects. The ToA usage with UWB signals makes ToA one of the most accurate
techniques for indoor usage (Gu et al, 2009). However, maximising the accuracy means a
precise time synchronization and therefore more human efforts. Beside, an additional server
is needed for time delay calculation, which increases the overall complexity and costs (Dwiyasa
& Lim, 2016; Farid et al, 2013; Hightower, 2001).
Figure 8: ToA and ToF principle (Farid et al, 2013) Figure 7: AoA-based positioning (Farid et al, 2013)
8
Time difference of arrival
Time Difference of Arrival (TDoA), the position determination is calculated by transmitting a
signal (from a mobile device) to several receivers (base stations). Next, the difference between
time of arrival shows the distance from which the positon can be calculated. TDoA uses
multilateration to determine the position. The difference of arrival time produces a hyperbolic
curve (figure 10), the intersection of multiple curves shows a possible location of the transmitter
(Farid et al, 2013; Nuaimi & Kamel, 2011; Zhang et al, 2010). Whereas ToA uses absolute
time measurements, TDoA is based on relative time measurements, illustrated in figure 9.
Thus, time synchronization is only needed between the receivers (Farid et al, 2013). Peak
efficiency of TDoA systems can be found for large and relatively open spaces, thus this system
is mostly found in outside environments. NLOS and multipath effects interfere with the
transmitted signal which affect the time of flight of the signal (Bouet & Santos, 2008;
CiscoSystems, 2008).
Round-trip time of flight
Round-Trip of Flight (RToF) or Round Trip Time (RTT) determine the position by measuring
the time-of-flight of a signal pulse travelling from the transmitter to the base station and back,
or vice versa, which is illustrated in figure 11. The
measuring principle is comparable to ToA technique, but
whereas ToA needs two local clocks to calculate the
delay, RToF only needs one. This advantage solves the
problem of synchronisation. Nonetheless, some
processing time is needed for greater distances, which
results in a less accurate position determination
because the position could already have changed
during the processing time (Dwiyasa & Lim, 2016; Farid
et al, 2013; Liu et al, 2007).
Figure 10: TDoA based on relative time measurements (Farid et al, 2013) Figure 9: Position determination via hyperbolic
curves (Rohde&Schwarz)
Figure 11: Two way measurement of RToF (P = position; A – C = base stations; R1-3 = radiuses (Liu et al, 2007; Tomasi & Manduchi, 1998)
9
2.2.3.3. Power based
Received Signal Strength (RSS) or Received Signal Strength indicator (RSSI), also known as
signal attenuation or signal property-based system, is based on the strength or power of a
signal. RSS compute the distance to the position by measuring, at a base station, the signal
strength or attenuation of a transmitted signal. Next, theoretical and/or empirical path-loss
models translate the difference between the transmitted and received signal strengths into a
estimation range (Dardari et al, 2015; Dwiyasa & Lim, 2016; Farid et al, 2013; Liu et al, 2007).
Visual representation of the principle is found in figure 12. Nowadays, RSS can be measured
by mostly all wireless devices, which is the main advantage compared to the other methods.
No need for time synchronization makes this
advantage more desirable in location
determination techniques (Dardari et al, 2015;
Dwiyasa & Lim, 2016). Furthermore, RSS has
nearly no impact on power consumption,
sensor size and cost (Barsocchi et al, 2012).
However, the disadvantage of RSS is the
multipath effect, which changes the amplitude
of the signal and therefore results in a lower
accuracy (Dwiyasa & Lim, 2016).
2.2.4. Fingerprinting
Fingerprinting principle differs from the above mentioned techniques. Where the previous
methods determine the distances between the mobile device and base station and afterwards
triangulate the position, fingerprinting locates the mobile device by comparing the obtained
RSSI values to a previous determined radio map (Honkavirta et al, 2009). Fingerprinting
consists of two main stages: (1) an offline stage, in which a database of RSS is built, and (2)
a localization or online stage, in which algorithms search for the best match in the database
(Sánchez-Rodríguez et al, 2015; Tuta & Juric, 2016).
Figure 13 describes each step of the fingerprinting method. In the offline (training (Chen et al,
2013; Werner, 2014) or profiling (Chowdhury et al, 2016)) stage, RSS vectors are collected at
predetermined grid points (or access points). Each vector contains the coordinates from all
base stations in the area. Thereafter, a radio map database is constructed with these vectors
(Swangmuang & Krishnamurthy, 2008). During the online (estimation (Chowdhury et al, 2016)
or positioning (Chen et al, 2013)) stage, a sample vector is obtained and compared with the
radio map and therefore the previous obtained RSS vectors. The closest match to the sample
of RSS is used as estimation of the user’s position (Swangmuang & Krishnamurthy, 2008). To
match the online with the offline database, systems such as probabilistic methods, data mining
techniques, Kalman filters etc. are used (Sánchez-Rodríguez et al, 2015).
Figure 12: RSS positioning (P = position; 𝐿𝑆1−3 = measured path-loss; A – C = base stations)
10
Fingerprinting looks like a promising technique, however, there are also some drawbacks. A
minimum time is needed to collect and determine the location at each position. This could
affect the accuracy because the position could already have changed by the time the system
is done calculating the previous position. Even changes of the environment in time, caused by
movement and varying number of people, around the measured location effects the position
estimation (Farid et al, 2013; Kaemarungsi & Krishnamurthy, 2012; Shang et al, 2015). Beside
the time-disadvantage, is the necessity of site survey higher and positively correlated with the
resolution of the radio-map. This means to obtain a higher accuracy, more cells or vectors
have to be surveyed (Hernández et al, 2017).
Figure 13: Fingerprinting procedure with (a) training phase, multiple access points (AP) communicate with the mobile user (MU) which is located at a specific and known reference point (RP). The communicated data is stored in the form of X and Y coordinates and accompanying RSS received from each AP. The training is repeated for each RP; and (b) positioning phase, the MU registers several RSSs. The RSSs is analysed using a specific designed algorithm and used the stored data from the training phase to determine the position of the MU (Li et al)
11
2.3. Positioning systems
In this section, a survey of the positioning systems is given. In general, these systems can be
divided into two categories: systems which (1) uses the absolute position, were the estimating
procedure uses external devices such as beacons, landmarks, etc. Systems using absolute
positioning are GPS, ultrasound, Bluetooth, wireless local area network (WLAN) etc.; or (2)
using the relative positon, which does not need infrastructure. The used sensors, such as
encoder system and inertial navigation system (INS), are only installed in mobile devices (Kim
et al, 2017).
Whereas section 2.2 discussed the different used techniques, here, systems based on these
techniques are put forward. Each system will be classified in a general group, based on
comparable specifications. Next a description as well as some pros and cons are given.
Note: as shown in figure 14, there is a fundamental difference between transmitting and
receiving devices. Depending on the used signals, the mobile device can send, then called a
transmitter or beacon; or receive signals, then called a receiver. The same principle goes for
the stationary device.
Figure 14: Transmitter and receiver principle (Metratec, 2010)
2.3.1. Global positioning systems
For many years, global positioning systems (GPS) or global navigation satellite systems
(GNSS) (Lu et al, 2016) are used for multiple purposes, from tracking to navigating people in
an outdoor environment. The system has a worldwide coverage and can be used by a number
of users at the same time while providing user privacy. This can be ensured by the way the
system is designed, where passive one-way signals are transmitted by Earth-orbiting satellites
and determination of the position happens at the receivers (Richard, 1990). Thereafter, the
accuracy can vary from several meters to only a few millimetres. However, a higher accuracy
is correlated with a bigger acquisition and maintenance cost (Grimes, 2008; LaMarca, 2008;
Werner, 2014).
12
2.3.2. Infrared based systems
The system is based on transmitting a signal which uses the spectral region of infrared (IR)
light. The IR-technology is used for proximity-based location techniques (Chowdhury et al,
2016; Kuriakose et al, 2014). IR estimates an absolute position (Gu et al, 2009) and is used in
wired and wireless devices, which communicate between transmitter and receiver via two
techniques: (1) line-of-sight path, where the main disadvantage of the system is the interruption
of the light source (Farid et al, 2013; Liu et al, 2007) or (2) diffusion, which relies on the
reflections from a large diffusive reflector, such as a ceiling (Fernando et al, 2003).
The usage of IR-systems for indoor position determination is rather limited due to several
disadvantages (Zhang et al, 2010). For both systems, a dispersion, caused by reflectors,
fluorescent light and/or sunlight, results in an impediment between the transmitter and receiver
and therefore an inaccurate position determination (Bahl & Padmanabhan, 2000). Expensive
hardware and maintenance costs make the system expensive. Lack of security and privacy
make the user vulnerable. However, the mobile devices are small, light-weighted and easy to
carry, which make IR-system more advantageous for portable use (Farid et al, 2013).
Examples of infrared systems are: Active badge (Want et al, 1992), Firefly (Cybernet System
Corporation; Ann Arbor, Michigan; USA), OPTOTRAK (Northern Digital Inc.; Waterloo,
Figure 15: Generic block diagram of passive (left) versus active (right) RFID systems ( red arrow = data transfer; green arrow = energy transfer)
15
Several procedures define the overall efficiency of the RFID system. Fine-tuning of the training
phase, resolution, accuracy and hardware components are essential. The resolution of RFID
depends on three factors: (1) scan rate (2) maximum range of the system and (3) configuration
of the network hardware. A lower scan rate reduces the accuracy of the system. The
robustness of hardware depends on the vulnerability to noise and therefore stability of the
signal strength reading (Manapure et al, 2004).
In comparison to other positioning systems, RFID is more desirable due to its small size and
relatively low cost (Jinhong et al, 2011). As followed, more systems are being developed based
on RFID principle: LANDMARC (Ni et al, 2003), SpotON (Hightower et al, 2000). However,
frequently changing environments, difficult integration into other systems and the use of
proximity positioning technique make the RFID technology less desirable (Abdat et al, 2010;
Jinhong et al, 2011).
Bluetooth
Bluetooth is a wireless personal area network (WPAN), which in comparison to WLAN covers
a smaller area (Fahmy, 2016). The signals operate in the 2,4 GHz ISM band (Farid et al, 2013)
and is classified as IEEE 802.15.1(Benini et al, 2006). Position estimations via RSSI
(Kyamakya et al, 2003; Vorst et al, 2008) and proximity (Kuriakose et al, 2014) uses Bluetooth
signals. Bluetooth positioning system can be used as master/slave principle, called the
infrastructural mode and shown in figure 16, where the mobile devices only communicate with
the base station. However, the mobile devices can also exchange location information,
illustrated in figure 17. Both systems form a piconet, which is a network that links devices using
Bluetooth technology, based on the medium access protocol (MAC) (Kotanen et al, 2003;
Thongthammachart & Olesen, 2003).
Figure 16: Bluetooth infrastructural mode network (Thongthammachart & Olesen, 2003)
Figure 17: Bluetooth ad-hoc mode network (Thongthammachart & Olesen, 2003)
16
The system has an range between 10 - 100m with a position error of about 2 - 5 m (Gu et al,
2009; King et al, 2009; Vorst et al, 2008; Yassin et al, 2017), and depends of a balance
between range, accuracy and power consumption (Hallberg et al, 2003). Due to high security,
low cost, and small size Bluetooth is much more useful for indoor positioning applications but
disadvantages of the system are the higher power consumption and localization latency (10-
30s) because for each location finding, the system runs the device localization procedure,
which is time consuming. Therefore, Bluetooth is unsuitable for real-time (Farid et al, 2013)
and more used for small scale (Jinhong et al, 2011) positioning. Examples of used Bluetooth
(hybrid) systems nowadays are Topaz (TADLYS, 2004) and BLPA (Kotanen et al, 2003).
Ultra-wideband
Ultra-wideband (UWB), IEEE 82.15.4, is a short-range, high-bandwidth, fault-tolerant and
multipath resistance communication system (Farid et al, 2013; Zhang et al, 2010). It has
relative bandwidth larger than 20% or absolute bandwidth of more than 500MHz (Alarifi et al,
2016). Depending on the capacity, preferences and constructer, the used frequency wherein
UWB works is between 3,2-10,6 GHz, and can be divided into low-band (3,2-4,7 GHz) and
high-band (5,9-10,2 GHz). These frequencies are mostly regulated by European commission
(EC) and U.S. Federal Communications Commission (FCC) regulations, which imposes
certain power emissions limits. An overview of the used frequencies is shown in figure 18.
UWB is mostly been used for indoor applications, but regulations permit short-range outdoor
usage (Alarifi et al, 2016; Sahinoglu et al, 2008; Sahinoglu et al, 2011). An UWB setup consists
of a radio wave generator and several receivers. The needed base stations are expensive, but
mobile devices or UWB beacons (and RFID-tags) are relatively cheap compared to mobile
devices of other systems (Farid et al, 2013; Werner, 2014), and have a lower power
consumption. As already described in the definition, UWB has a good multipath resistance.
Therefore, a higher accuracy (20-30 cm) compared to the other systems can be obtained (Farid
et al, 2013). Better capabilities in NLOS conditions are achieved compared to WLAN systems
(Galler et al, 2006). Other advantage of the large bandwidth is high-speed data communication,
resulting in a faster position estimation for real-time usage (Sahinoglu et al, 2008). Mostly
TOA, but also AOA, TDOA and RSS uses this UWB accuracy for position estimation (Alarifi et
al, 2016; Dardari et al, 2015; Jayabharathy et al, 2014). Examples of UWB based technologies
are: (Sapphire)DART (ZEBRA, 2017), Ubisense (Ubisense, 2017) and PulseON350
(timedomain, 2012)
Figure 18: Different bandwidths used for positioning signals ( a: global positioning system (GPS) [1,56-1,61 GHz], b: personal communication system (PCS) [1,85-1,99 GHz], c: microwave, Bluetooth [2,4-2,48 GHz], d: WLAN [5,725-5,825 GHz] and e: UWB [3,1-10,6 GHz]) (Sahinoglu et al, 2011)
17
2.3.4. Ultrasound based systems
Ultrasound technology is based on the natural phenomenon of bats. The technique operates
in a low frequency (22-26 kHz) bandwidth compared to other systems. Ultrasound can be
detected up to 17m away, thus short ranged. The accuracy varies between 3cm-1m (Alarifi et
al, 2016; Farid et al, 2013; Segers et al, 2015).
The position is estimated by the base station, which receives an ultrasound signal from the
emitter tag (Farid et al, 2013). However, different ultrasound signals can be emitted. The first
and most basic technique, is sending a ultrasonic pulse at a given frequency. Disadvantage of
this technique is the in-band noise, which affect the accuracy. Another approach is using a
frequency with orthogonal sequences, which have better reliability against noise. This
technique is called code division multiple access (CDMA). For the moment, two major systems
are developed: (1) direct sequence spread spectrum approach (DSSS) and (2) frequency
hopping spread spectrum (FHSS) (Segers et al, 2015). TOA (figure 20) and TDOA (figure 19)
use ultrasound waves to locate the user’s position.
Figure 20: TOA approach with ultrasound waves
Ultrasound systems can only be used for LOS applications. Furthermore, due to interference
from reflected ultrasound signals, the systems suffers from multi-path effects, which influences
the accuracy. The used tags are relatively cheap (Farid et al, 2013; Jinhong et al, 2011; Zhang
et al, 2010).
Examples of ultrasound based systems are: Active
Bat (figure 21) (Ward et al, 2002), (Priyantha et al,
2000) and iLoc, Sonitor (Sonitor, 2015) Note: most
of this systems are in combination with
radiofrequency signals and thus increase the
accuracy (Segers et al, 2015).
Figure 19: TDOA approach with ultrasound and RF waves
Figure 21: The Active Bat system (Gu et al, 2009)
18
2.4. Performance metrics
Before comparing different IPSs, it speaks for itself that some key factors have to be described.
By means of performance criteria, differences can be shown between the systems. Note that
a disadvantage from one particular system in general will be problem for a correct display of
the user’s position (LaMarca, 2008). Further in the study, these elements will be evaluated for
each positioning system.
2.4.1. Accuracy
Accuracy is the most significant and therefore the most often-cited metric in the determination
of a position. The metric refers to the correctness of a system’s location estimation and is
expressed in percentages. However, the term “accuracy” is in the literature in general a fusion
of two other terms, namely accuracy and precision. Both terms, illustrated in figure 22, vary
between papers. Gu et al (2009) describes accuracy as the average error distance expressed,
whereas Hightower & Borrielo (2003) refers to the closeness degree between the truth and the
measurement. Next, both authors define precision respectively as the success of position
estimations with respect to predefined accuracy, and the repeatability of the measurements
(Gu et al, 2009; Hightower & Borriello, 2001). For simplicity, further in the paper, both terms
will be combined and be called “accuracy”.
Depending on whether the estimated location data contains errors (Hightower, 2001), a
percentage can be given. These errors are mostly expressed as a “median error” (ME), which
means 50% of the estimations are at least that accurate. A “normally distributed error”
expresses the error at 1 or 2 sigma, meaning respectively 66% and 95% of the estimates are
at least that accurate (LaMarca, 2008).
Figure 22: Accuracy vs. precision for a one-dimensional positioning system( a and b both present estimated positions of the true position (middle); Precision indicates the deviation of the location estimation from the same location, whereas Accuracy indicates the deviation from the true position (Werner, 2014)
19
The accuracy depends on many factors: (1) the structure of the environment, as already
mentioned in paragraph 2.1, obstacles can lower the accuracy; (2) technology, this aspect
varies across used positioning systems, algorithms or filters to deal with the measurements as
well as the geometric layout of the nodes (Sharp et al, 2009); (3) time needed to determine the
signal and (4) overall errors. In general, the higher the accuracy, the better the representation
of the position and the better the system. Note that even the same positioning system can
show significantly varying accuracy (Shang et al, 2015). Besides, the purpose of the IPS
determines the level of accuracy needed for that specific application (LaMarca, 2008;
Mohapatra & Suma, 2005). For example: for knowing the exact position of the user, a more
accurate system has to be used in comparison with knowing in which room the user is located
(Gu et al, 2009).
2.4.2. Reliability and robustness
Both terms refer to the systems components as well as transmitted signals. The indoor
positioning components sometimes have to work in a harsh environment, for example the
design and construction of the housing of the system’s hardware has to be able to withstand
water, corrosive gasses (NH3-gas in agricultural environments), outer and inner temperature,
light or outdoor pressure. The robustness is a critical aspect for a long-term durability of the
system (Liu et al, 2007). IPS must be well designed to adapt intrinsic and extrinsic errors, like
loss of signals and hardware malfunction respectively. These problems should have a low
impact on the location information and the system maintains its functionality (Fahmy, 2016; Gu
et al, 2009).
2.4.3. Responsiveness
The responsiveness represents how fast the location estimation of a moving target can be
updated. In other words, responsiveness describes the delay of an IPS, which contains the
delay of measuring, calculating the position of the estimated target and forwarding the
position’s information to the requesting parts (Farid et al, 2013; Gu et al, 2009). The timing or
delay of the system’s infrastructure can introduce discrepancies between the actual and
reported positon during the positioning procedure. These delays are mostly caused by two
reasons: the indoor environment is changing dynamically or the target is moving too quickly
(Gu et al, 2009). Though for most of the systems delay is seen as an error which effects the
accuracy, some IPS’s work on the principle of time delay (Gezici et al, 2005).
2.4.4. Scalability
Scalability refers to the performance of the system when the area of interest changes. For
example, the performance of the position determination is inversely proportional to the distance
between the transmitter and receiver. Further, scalability can address more specific structures.
Geographic scaling describes the coverage of an area, whereas density scaling represents the
number of units positioned per unit geographic space or area per time period (Yassin et al,
2017).
20
2.4.5. Cost
The cost can depend on many factors like: time, space, weight, energy and intensity or
accuracy. Each factor depends on the user’s preferences. Time refers to installation and
maintenance duration. Network-based systems are nowadays more interesting because of the
smaller infrastructure and therefore more cost-effective. Mobile capabilities are closely
connected with space and weight. Energy and accuracy relate to the application of the IPS (Gu
et al, 2009; Liu et al, 2007). As already quoted within accuracy-paragraph, there will mostly be
a trade-off between accuracy and cost (Haute et al, 2014).
2.5. Applications for positioning systems in animal welfare
As positioning systems nowadays become more important in everyday usages, so does the
integration in animal welfare. Over the last years, the size of an average farm has been growing
systematically (FODeconomie, 2014). Thus, the farmer’s available time per animal reduces,
which could lead to more health problems (Willgert et al, 2014). In order to keep track of what’s
happening on the farm, integration of electronical systems help the daily activities of the farmer.
The usage of positioning or locomotion detection is a new step to ease the workload. Because
the development is still in its initial phase, not that many applications are used for animal
purposes.
In indoor environments, determination of the position is mostly needed to know the behaviour
of the animal. Next, depending on the usage of the application, the farmer can anticipate and
take further actions. To further explain this point of view, this paragraph describes some used
applications for animal welfare.
Improvement of location system resulting in hybrid systems
Dandan and Lin (2016) used passive RFID tags ,which were implanted on the animal, and
UHF (134 kHz) signals. The system was designed to track multiple mice in lab environment,
this in combination with trajectory identification via a vision system (CCD camera). The purpose
of the experiment was to show the combination of the two systems. They concluded that the
errors of identities swaps using only visual tracking can be reduced by combining the
positioning systems. Furthermore, an higher accuracy and reduced computing complexity
were also concluded in the tests. However, by increasing the number of mice, the accuracy of
the combined system reduces (accuracy of 97% with two mice to 76% with four mice) (Dandan
& Lin, 2016).
21
Location monitoring
Knowing where the animal is located, is a first step towards integration of IPS’s for animal
related purposes. The user only needs the IPS to locate the animal(s) in real-time. For outdoor
environment, GPS is already been used for several applications (Menzies et al, 2016).
Kim et al. (2010) integrated the RFID system in zoo’s. Zookeepers uses the location system in
combination with GPS to know the real-time position and (health-)information of the animal.
This cage management system is also used in case the animal escapes. Furthermore, the
information could be an extra touristic attraction, where they can be provided with the health
status (Kim et al, 2010).
Behaviour monitoring
The next step in using IPS is monitoring the behaviour of animals. The behaviour can explain
how animals interact with environmental structures, with each other and with themselves. Note
that the tags, which are mostly attached with a collar around the neck, should not interfere with
the behaviour of the animal.
Cornou (2010) classified a sows’ activity using acceleration patterns. The sow wore a neck
collar which contained a digital accelerometer. The collected data was transmitted via
bluetooth signals to dongles on the ceiling, and saved on a central computer. The acceleration
was recorded four times per second during 20 days. Afterwards, by using a multi-process
kalman filter (MPKF) and predefined parameters, the frequency of five different activities are
registered.
22
3. Lameness
Lameness can be described as “a clinical appearance of locomotion disorders” (KilBride et al,
2009) or as “an impaired movement or deviation from normal gait” (Smith, 1988) that
depending on the severity and type of the disorder, results in discomfort, pain and immobility
Lameness is an important aspect for animal welfare in pig farming, especially when animal
welfare is becoming more and more important for the consumer (Heinonen et al, 2013). For
group-housed sows, lameness is about 80% caused by foot problems (van der Meulen et al,
1990) and is also the third greatest reason for culling sows, which results in a greater economic
loss for the farmer. Culling sows due to lameness is estimated between 8%-25% (Anil et al,
2009a; Pluym et al, 2011; Schenk et al, 2010; Tiranti & Morrison, 2006).
Lameness, detected in group-housed sows, (1) has anegative effect on the health status of
the sows, for example neurological disorders, trauma, metabolic and infectious diseases and
mechanical-structural problems; and (2) is associated with poor reproductive performance,
which results in a decreased litter size, poor farrowing performance and decreased sow
longevity (Anil et al, 2009a; Engblom et al, 2008; Grandjot, 2007). Stress, pain or both, caused
by lameness, have a negative effect on the immune system. Therefore, other health
complications could appear (Barnett et al, 1987; Kroneman et al, 1993).
3.1. Risk factors inducing lameness
Lameness can be caused by many factors. Causes differ between non-infectious hazards,
such as osteochondrosis, degeneration of bone, limb malformation and infectious disorders
such as skin lesions and joint arthritis (Cador et al, 2014). Knowing which of this aspects are
causing lameness in the group-housed barn gets the farmer one step closer to reduce and
preventing lameness-related problems in the barn.
Most problems with lame animals are results of bad floor and/or housing conditions. Hard,
concrete floors are not well for sow comfort (Tuyttens, 2005), furthermore, floor characteristics
e.g. hardness and slip-sensitivity are often associated with lameness (Calderon Diaz & Boyle,
2014). The usage of some systems can provide extra comfort for the sow while staying in
gestation barn. Installing rubber mats provide comfort and could reduce lameness (Bos et al,
2016), however, several studies have shown that no difference for lameness related problems
(Elmore et al, 2010; Smith, 1988). In other words, extra comfort is not (always) proportional to
less lame sows, but could insure a better recovery due to a softer surface of the rubber mats.
Slatted floors in comparison to solid floors were associated with a higher risk of body lesions
(Bonde et al, 2004b). An electronic sow feeder (ESF) increases the likelihood of having all
kinds of claw lesions (Anil et al, 2007). Above given housing influences can increase when
sows are group-housed, mostly due to mutual interaction and aggression (Schenk et al, 2010).
23
If the animal suffers from injuries at the legs, hip or claw, bacteria could infect these wounds.
Next, these organisms will, via these lesions, get into the blood stream and gain entry to the
joints. Examples of these bacteria are Streptococci species, Staphylococci species and
Escherichia coli (Smith, 1988). Providing a clean environment and rearing conditions can
decrease the bacterial presence in the barn (Cador et al, 2014).
Nutritional components of the sows diet will interfere with the animal in the form of rapid growth
rate (i.e. osteomalacia), heavy muscling and nutritional status. Fast growth rate could result in
more growing pains and weaker bone structure, mostly due to a decrease of bone
mineralisation and bone development (Cera & Mahan, 1988). Some studies link a lack of biotin
(also known as Vitamin H) or trace minerals (Cu, Zn and Mn) to more claw lesions (Anil et al,
2009b; Smith, 1988), others related low feed intake per day (gilts: lower than 3.1kg ) or per
year (sow: 1200 kg) to an increased risk of leg disorders (Cador et al, 2014).
Genetics is an aspect which is less lameness-induced and but could ensure fewer lame
descendants. Research has shown that some breeds are more vulnerable for weak legs or
increasing susceptibility to injuries. Selectively breeding and choosing for a more robust boar,
whose know for a better stature, for production of new sows could result in a less lameness-
sensitive animal (Estienne et al, 2006; Schenk et al, 2010).
3.2. Visual and non-visual consequences of lameness
Each cause of lameness results in several visual and non-visual effects. These effects will
affect the behaviour and health status. It is up to the farmer to recognize these lameness-
related consequences in order to prevent further suffering of the animal. Furthermore, fast
response of the farmer can increase longevity and decreases involuntary culling, which
decrease costs and financial losses (Jensen et al, 2010).
In most of the cases, the animal suffers from stiffness, has an abnormal gait and/or stance, is
unwilling to move, has an arched-back with the hind legs tucked under the abdomen and sways
the hindquarters more (Jørgensen, 2000; Stavrakakis et al, 2015). Some abnormal stance
characteristics are shown in figure 23. Mutual differences depend on the number of joints
affected. The abnormal gait/stance is due to the redistribution of the weight of the animal on
the limb(s), resulting in less weight on the painful limb(s). Indirect influences on the gait can be
caused through prolonged standing times due to less comfortable lying (KilBride et al, 2008;
Main et al, 2000; Sun et al, 2011).
Some forms of lameness show lesions (e.g. overgrowths, white line, junction between sole and
heel, hip/shoulder) all over the sows limbs, predominantly claw-specific injuries (Cador et al,
2014; Pluym et al, 2011). Some studies showed more (severe) lesions on the forelimbs than
on the hind limbs (Anil et al, 2007; Tiranti & Morrison, 2006). Sever forms reflect in swollen
joints and could contain serosanguinous fluid. In some of the cases erosion of the sole, toe,
heel or whole claw appear. The dog sitting position of the sow is assumed as severe lame,
difficulty in rising and maintaining stability (Smith, 1988).
24
Beside the change of visual appearance also the behaviour of the sow will change. A
decreased activity is found due to increasing pain and stress. Following from this is a lower
number of feeder visits, which can results in hunger and thirst (Heinonen et al, 2013). The
animal is also restricted in covering distances due to piling up pain, and may result in reduced
feed intake (Bos et al, 2015). Activity in the form of walking speed differs between researches
and variations are mostly based on the way of collecting the walking speed data. Grégoire et
al. (2013) collected the data using sows via kinematic posture detection and found a walking
speed of 0,836 m/s for lame sows and 0,948 m/s for sound sows, but the study did not specify
the type (front/hind) of lameness. Another study (also based on kinematic detection) used gilts
(with an average weight of 140 kg) for lameness research. They specified in lameness affecting
the front limb and hind limb, with a walking speed of respectively 0,890 m/s and 0,837 m/s.
The non-lame reference gilts had a walking speed of 0,942 m/s (Stavrakakis et al, 2015). In
general, a reduction of general activity, social interaction and exploration can be observed
within group-housed gestating sows (Millman, 2007; Weary et al, 2009).
General health can be influenced by lameness. Alterations of the immune system make the
sow susceptible for secondary diseases. Associated to sickness is a loss of body condition
(Bonde et al, 2004a). Less activity and more sitting increased the dirtiness of the sow and
makes the animal more vulnerable to infections (BARA et al, 1993; Oravainen et al, 2007).
Normal stance Standing under Normal pastern Weak pastern
Sickled Straight Buckle
dd
a) b)
c)
Figure 23: Leg weakness traits ( a: hind limb stance; b: leg stance; c) front limb alignment)(Van Steenbergen, 1989)
25
4. Material and methods
4.1. Introduction
Preventing or at least reducing welfare problems and accompanying financial losses, a close,
early and accurate lameness detection and treatment are essential. Despite the high
prevalence of lameness nowadays, diagnosis can be unreliable. This is mostly because of lack
of criterion-referenced standard or different lameness-scoring of the caretakers (Anil et al,
2009a). Furthermore, lameness is evaluated by visual scoring of the gait, but this is a
subjective technique and has low reproducibility (Pluym et al, 2013).
In order to have a more objective lameness score and reduce the time spent manually scoring
the animals, researchers are developing several techniques. To eliminate the disadvantages
of visual, pressure or posture lameness scoring, we introduce indoor positioning system (IPS)
to detect and eventually score lameness of sows in a group-housed environment. The real-
time position of the sow is recorded via the IPS and send to a central computer, where the
sow’s path is constructed. Next, by analysing the path, several variables can be measured
which will tell more about the lameness status of the sow.
The aims of this study were (1) to help tuning an accurate IPS which can register a sow’s
trajectory; (2) to help designing an collar which will be worn by a sow; (3) to evaluate smoothing
parameters in order to obtain a de-noised path which represent a walked trajectory of a sow;
(4) to derive and test the repeatability of different walking variables that are relevant for
lameness research; and (5) to investigate whether these variables differ significantly between
lame and sound sows;
4.2. Animals and housing
The barn is located at Scheldeweg 68 9090 Melle (Belgium). The building is property of the
Institute for Agricultural, Fisheries and Food Research (ILVO) and is part of the animal science
unit. The group-housing barn itself is divided into two compartments by a (hard plastic) wall in
the middle. As shown in figure 24, each compartment consist two pens. All four pens have the
same dimension and are symmetrical to each other. Two types of flooring can be found: (1)
closed floor, divided with 5-6 cm thick concrete wall, and (2) slatted floor. Depending on other
studies done at the same time, the floor could be covered with rubber maths. Each pen also
contains a metal feeding station (brand: Nedap), located in the corner of each pen; a water
dispenser; a heater; a normal and electronic brush and a hanging ball used as a toy for the
animals. The dimensions and a 3D view of one compartment is shown in figure 25.
Figure 45: Ideal settings with σd and accompanying σr based on the autocorrelation function (sound ww = 7 ( ) , sound ww = 5 ( ) , front lame ww = 7 ( ) and front lame ww = 5( ))
0
0,5
1
1,5
2
2,5
3
3,5
4
0 2 4 6 8 10 12
σr
σd
43
The paths belonging with the settings (in figure 45) of the sound sow with ww = 7 settings are
showed in figure 46. Only the ideal settings for the sound and front lame sow datasets were
tested. Further progress regarding the autocorrelation function technique was stopped due to
several circumstances. Firstly, as figure 45 illustrates, using different datasets results in
different parameter settings for the online application. So, an ideal setting which can be used
for both lameness statuses is hard to find. Using a specific parameter setting for one dataset
will result in an over- or under-filtering of the other dataset. Secondly and more likely the more
decisive reason: Figure 46 shows for each ACF-based parameter setting the accompanying
de-noised path. Visual analyses of the de-noised path shows for each parameter setting an
over-filtered de-noised path. At first sight, no significant difference between the settings is
visible. However, as can be seen and will be quoted in the next paragraph, the de-noised paths
indicate significant over-filtering. This deletes or reduces essential information, which
transform the de-noised path to a vague representation of the original path.
5
5,2
5,4
5,6
5,8
6
6,2
6,4
6,6
6,8
7
19 20 21 22 23 24 25 26
X (
m)
Y (m)
11 1,5 5,7 1,6 4,4 1,7 3,8 1,8 3,5 1,9 3.2 2
3 2,1 2,9 2,2 2,8 2,3 2,7 2,4 2,6 2,5 original
Figure 46: original versus de-noised path (ideal settings according to the autocorrelation function) (number left: σd, number right: σr, window width = 7)
44
5.2.1.2. Based on visually comparing
The investigation starts with a training path, chosen from the sound sow dataset. Figure 47
shows the curved path, consisting of 21 data points, two sharp turns and some noise. Next,
four test paths (shown in figure 48) are chosen based on the preconditions, and were also
used from the sound sow dataset.
Four tests were needed to inspect the influences of σd and σr, each test with a constant window
width (ww) of 7. Test one starts with comparing the full range of σd and σr, respectively range
0,1 – 20 and 0,1 – 3. For the second test, the range of σr was narrowed down to 0.5 – 1, but
still using the full range of σd. The third test included a range of σd between 1 and 5, with the
same range for σr as test two. Next, two tests were needed to inspect the influence of the
window width. The first window width test (or fourth overall test) uses a range of σd, σr and ww
respectively between 0,1 – 20 ; 0,1 – 3 and 3 – 15. At last, a fifth and final test was constructed
using the three parameters. The reason for this five-partite test will become clear further. Note
that not all values between the predefined ranges are used to construct the charts. Only a few
values are used based on previously done research and small scale pre-tests; furthermore,
researching and comparing each parameter value is time-consuming and the ratio extra data
to time will not provide extra significant decisive information.
Figure 47: Training path ( = recorded path; : walked direction; = sharp turn; = noise)
The influence of σr in test 1 and test 2 are visually clarified via respectively figure 49 and figure
50, each figure representing the extremes of the researched range per test. At the lowest value
of σd , the de-noised paths have the least amount of differences relative to the recorded path.
Only the data-points which indicates strong noise deviations are partly smoothened.
Furthermore, no difference is visually detected between the σr values. When increasing the
value of σd , differences occur between the de-noised paths. When focused on predetermined
critical points (sharp turn, sever noise deviations), several influences become clear:
(1) the size of the difference between two consecutive de-noised paths increases
significant between a σr range of 0,1 and 1,0. Using a σr value greater than 1,0 results in a
decrease of the size of this difference. However, this difference is only visible at critical points.
Concluded is a significant change of the influence of σr in a range of 0,1 and 1,0 regarding the
differences between de-noised and recorded paths and between themselves. Therefore, a
second test was needed. Test two uses a range of σr of 0,5 and 1,0, and the full range of σd.
σr values lower than 0,5 were not used because of problems with overfitting.
(2) Path preservation; increasing σr results more or less in the preservation of the
original critical points. The influence is better distinguishable at low values. σr affects a small
area around the critical points. Thus the smoothing of the path is only visible at a certain range
before and after these critical points. Outside this range, the de-noised path follows the
recorded path parallel or they have a structure which is similar to each other. Furthermore, the
greater σr, the less the de-noised path shows resemblance with recorded path and the larger
the smoothing area around the critical points.
The influence of σd in test 1and test 2 are visually clarified via respectively figure 51 and figure
52, each figure representing the extremes of the researched range per test. At the lowest value
of σr, the de-noised paths have the least amount of differences relative to the recorded path.
At datapoints which indicates strong noise deviations are smoothened. However, in
comparison with σr, there are more differences visible at the lowest value. At the datapoints
indicating noise, very small variations between the de-noised paths are visible. The deviation
between de-noised and recorded path rises as the value of σd rises. This indicates that at the
lowest value of σr, σd already affects and smoothens the recorded path.
When increasing σr, more significant differences are visible. When focused on predetermined
critical points, several influences can be described:
(1) the size of the difference between two consecutive de-noised paths increases
significant in a range of σd of 0,1 and 2,5. Using σd values greater than 2,5 results in a decrease
of this difference until no significant variation is visually detected. Therefore, a second test was
needed. Test three uses a range of σd between 0,5 and 2,5, and a range of σr between 0,5
and 1,0. σr values lower than 0,5 were not used because of problems with overfitting.
(2) Path preservation; increasing σd results faster in a more smoothened path. The term
faster in used because the higher σd, the more critical points are smoothened in comparison
with increasing σr. σd affects an area around the critical points, which is greater in comparison
with the affected area of σr. The smoothing of the path is thus visible over a larger range. This
47
results in a more smoothened de-noised path, therefore, there are fewer resemblances
between de-noised and recorded path. Outside this range, the de-noised path follows the
recorded path parallel or they have a similar structure.
The influence of the window width was tested in test four. Illustrated in figure 53 (test 4a)
and figure 54 (test 4b), window width values of 3, 9 and 15 were used to compare with σd and
σr. The test was divided in two parts in order to describe the changing window width: 1) Test
4a described the window width using constant σd and a rising σr between two consecutive
charts. This part is mainly used to evaluate the window width in function of σr; 2) Test 4b
describes the window width using a constant σr and a rising σd between two consecutive charts.
This part is used to evaluate the window width in function of σd.
At the lowest value of σd (test 4a), no difference is visually detected with an increasing ww and
this regardless of σr setting. However, at lowest value of σr (test 4b), some small difference
occur when increasing the ww. These differences increase when increasing σd. When
increasing σd or σr in combination with respectively an increasing σr or σd, more visual
differences occur. Overall, the influences of σr and σd as described above are also applicable
for this test, at which also was described that most significant variations between the two
parameters was at critical points (sharp turns and sever noise). However, the influences of σr
and σd dominates respectively in test 4a and test 4b.
1) the size of the difference between two consecutive de-noised paths; The higher each
σ, the more significant differences are visually between two consecutive ww settings. The only
difference between the two subtests regarding the mutual distance is that for test 4a the mutual
difference between two consecutive ww settings rises proportional to an increasing σr; whereas
for test 4b the difference between two consecutive ww settings rises significant between ww=3
and ww=9, but this difference was less great between ww=9 and ww=15.
2) Path preservation; the effect of the two main parameter is analogue for this test
regarding path preservation, but some different effect occur when increase the ww. The
influence of both σ will mainly be reinforced with an increasing ww. However, this reinforcement
is disadvantageous for the path preservation, mostly because the de-noised path is
smoothened over a larger area. For example, at highest ww, sharp turns are still recognizable
in the de-noised paths, but the distance between recorded and de-noised path becomes too
great at the centre of the turn. In other words, at high values of each parameter, the de-noised
path is represented at a short-cut regarding to the recorded path.
The fifth and final test represent the ideal range of each parameter. An ideal setting can be
chosen from this range to use for the lameness testing. The range is based on the findings of
test one to four. Each parameter is a final time evaluated, in order to see small changes
between different settings. For further calculations in this project, following ranges for each
parameter will be used: 1) for parameter σd a range between 0,8 and 2,0 was used; 2) for
parameter σr a range was defined between 0,5 and 0,8 and 3) for the window width only one
value, namely ww = 5, was used, mostly because the values above and below ww=5 showed
significant changes between the de-noised paths.
48
Figure 49: Test 1: influence of σr on the recorded path (Figure 49a (left): ww = 7, σd = 1, σr : 0,1 – 3,0; Figure 49b (right): ww = 7, σd = 20, σr : 0,1 – 3,0)
a) b)
a) b)
Figure 50: Test 2: influence of σr on the recorded path (Figure 50a (left): ww =7, σd=1, σr : 0,5 – 1,0; Figure 50b (right): ww = 7, σd = 20, σr : 0,5 – 1,0)
49
Figure 51: Test 1: influence of σd on the recorded path (Figure 51a (left): ww = 7, σr = 0,1, σd : 0,1 – 20; Figure 51b (right): ww = 7, σr = 3, σd : 0,1 – 20)
Figure 52: Test 2: influence of σd on the recorded path (Figure 52a (left): ww = 7, σr = 0,5, σd : 0,1 – 20; Figure 52b (right): ww = 7, σr = 1,0, σd : 0,1 – 20)
a) b)
a) b)
50
Figure 54: Test 4a: influence of the window width on the recorded path (Figure 54a (left): σd = 20, σr = 0,1, ww: 3 – 15; Figure 54b (right): σd = 20, σr = 3, ww: 3 – 15)
a) b)
a) b)
Figure 53: Test 4b: influence of the window width in relation to the dominant σ on the recorded path (Figure 53a (left): σd = 20, σr = 1, ww: 3 – 15; Figure 53b (right): σd = 2,5, σr = 3, ww: 3 – 15)
51
5.2.2. Lameness comparing
Due to the influence of several parameters on the IPS-data, the comparing of the health
statuses of the sows would more likely be influenced by the choice of a certain parameter
settings. This paragraph will describe in detail the three health statuses using a predefined
parameter setting. Therefore, parameter-related decisions and relations can be eliminated.
Many parameters are already introduced in this paper regarding the path of the sow, namely
window width, σd and σr. Now, the moving average is the fourth parameter that could affect
the processing of the speeddata. Figure 55 shows the raw speeddata, using a MA equal to
one, in other words, no smoothing of the speeddata. Comparing figure 55 with figure 58a,
illustrates the effect of an increasing moving average, resulting in a more smoothed behaviour
of the curve. The abrupt speed changes are reduced to a lower level, making visual analysation
more practical. Note that a greater moving average could result in more over-filtering of the
speed-data, which results in a loss of essential data. Figure 56 compares the recorded
speeddata of the walking phase represented as a MA of one and five. The greater the moving
average the lower the difference between the minima and maxima of the data. The difference
between two consecutive speedvalues showed no significant difference (p = 0,474) between
the two MA. No significant difference of variation was found between the three tests (p = 0,895)
and between the three lameness statuses (p = 0,626). By taking these previous results into
account, a moving average equal to five is used for further processing of the speed data.
Figure 55: Speed (m/s) of each test ( = test 1, = test 2, = test 3, = average), divided in the three phases (dark shade ( ) = walking phase, light shade ( ) = slowing down phase, normal shade ( ) = standing still phase) for moving average = 1 (source: sound sow 5 1,5 0,7)
Figure 56: Boxplot representations of the variation of the recorded speedvalues (while walking) of each health status (setting red (left): 1 5 1.5 0.7; setting blue (right): 5 5 1.5 0.7, respectively MA ww σd σr)
Figure 57: Boxplot representation of the speed of each health status (setting: 1 5 1.5 0.7 respectively MA ww σd σr)
53
Lameness comparing will only be using the IPS-data from each sow based on a predefined
parameter setting. Therefore, the number of different conclusions will be narrowed down to
only a few, which uses the predefined setting. The choice went to the setting with following
parameters: ww = 5, σd = 1,5, σr = 0,7 and MA = 5. This choice is based on visual analysation
using previous predetermined conditions, namely de-noise effect, path preservation and low
values of over-filtering. Note that this setting only will be used for visual representation of the
speed.
An overview of the three health statuses is represented in figure 58, each supported with a
speedchart and a walked path representation. Each of the three tests of the sound and hind
lame sow were valid and could be used for data exploration. Only two out of three tests were
valid of the front lame sow, because in one of the test no continuous movement was registered.
Therefore, this test was not used for data exploration.
The data used for statistical research is illustrated in figure 57. Statistical analysis will use
setting ww = 5, σd = 1,5, σr = 0,7 and MA = 1. The average speed of the front lame, hind lame
and sound sows was respectively 0,862 ± 0,026 m/s (mean ± s.d.), 0,734 ± 0,004 m/s (mean
± s.d.) and 1,205 ± 0,044 m/s (mean ± s.d.). The speed differs significant between the
lameness statuses (p < 0,001). No significant correlation was found between the speed and
the number of the tests for the front lame sow (r = 0,018; p = 0,893), hind lame sow (r = -0,006;
p =0,954) and sound sow (r = 0,082; p = 0,530).
The X-coordinates of the walked path were tested for possible deviations. When comparing
the data between the different lameness statuses (regardless of a distinction of the number of
the test), no significant difference (P = 0,329) was found between the sound sow and hind
lame sow. The difference was significant between the sound and front lame sows (P < 0,001)
and the front lame and hind lame sows (P < 0,001). Analysing the influence of the number of
the test on the X-deviations, gave a significant correlation for the hind lame sow (r = 0,490; p
< 0,001) and sound sow (r = 0,444; p = 0,003), but no correlation was found for the front lame
sow (r = 0,217; p = 0,088). The influence of the speed on the deviation of X (or vice versa) was
tested for each test, giving no significant correlation for test one (r = 0,076; p = 0,572) and test
two (r = -0.093; p = 0,499), but a correlation between S and ADx for test three (r = - 0.440; p =
0,006).
54
Y (m)
X (
m)
Sound sow
Hind lame sow
Front lame sow
a)
b)
a)
b)
a)
b)
Figure 58: Evolution of the speed (m/s) in time (s) (Figure 58a; top part) divided in three phases (dark shade ( ) = walking phase, light shade ( ) = slowing down phase and normal shade ( ) = standing still phase) and the accompanying paths (Figure 58b; bottom part) of each test ( = test 1, = test 2, = test 3 and = average) for each sow (above = sound sow, middle = hind lame sow and bottom = front lame sow)
The present study showed a real-time positioning system for indoor application, using ultra-
wideband signals with the time-of-arrival technique, integrated in a group-housed gestation
sow barn. Due to the integration of this setup for an agricultural application, which makes
comparing the performances of the system in this setup not that simple.
Hardware setup
This study is part of a bigger project at the ILVO. The present study was needed to fine-tune
and evaluate the data processing step and doing a small scale experiment to test capability
for registration of sow movement. The intention of the bigger project is to evaluate lameness
and movement related activities and mainly to detect lameness via these activities. This study
showed some bottlenecks while using the system in practical environment. The real time
positioning system (RTPS) used in the present study is based on time-of-arrival (TOA;
calculates the distance between tag and anchor, graphical represented as circles) position
estimations technique and uses ultra-wideband (UWB) signals to communicate between tag
and anchors. Communication between the anchors was based on a local area network (LAN).
Farid et al. (2013) indicated the variation of the intersecting point of the “constructed” circles.
Therefore, resulting in a lower accuracy. The present study has a similar conclusion,
furthermore, the variation of this intersecting point increases closer to one of the (minimal
three) anchors. UWB was describes by Jayabharathy et al. (2014) and Li et al (2007) as a low
power, better multipath resistant and fast real-time position estimation. All of the advantages
cited by the two authors were also perceived in this study. The low power consumption of the
tag was ideal for testing. The battery had a capacity of 390 mA and could be used for
approximately five hours, which is fine for this small scale experiment. However, for future
work, a new type of battery has to be used if the duration of the test will be for example several
weeks or one cycle at the gestation barn. The collar used for the test also need further
development. Harsh environment in the barn, mutual contact of the sows and the physiology
of the sow itself make it challenging task and affects the speed of degradation or loss of the
collar. Further testing will be needed to make sure the collar will withstand these extreme
circumstances for further usage.
Data processing and parameter fine-tuning
To process the data received from the IPS, the online application was used. This prototype
application was tested and underwent some adjustment towards this study. The application
will be the foundation for a new data processing platform and will be used for further projects
using the IPS. The program was mostly needed to process the raw data, which was needed
to evaluate each smoothing parameter.
56
A first way of searching for an ideal range of parameters was based on the ACF, which was
used to present a range of parameters based on statistical calculations. Searching for an ideal
range of parameter settings was based on the deviation of ACF(1)x and ACF(1)y . The results
showed that this principle does not guarantee one perfect range of settings which can be used
for every dataset, but several settings could be used depending the used dataset or
furthermore the tested lameness status. Due to the variation of settings, choosing a specific
setting will result in over- or underfiltering depending on the used dataset. Therefore, no further
research was done regarding this principle. Further research could be done based on a more
mathematical point of view, which includes deeper analysis of the ACF procedure.
Another way of searching for ideal range of parameters was via visual analysis of the de-noised
paths. The data was processed via an own developed processing platform, which mostly
constructed the de-noised paths. However the platform made the evaluation more efficient, the
visual analysation was time consuming and less statistical substantiated, which means, the
final range of settings is more likely influenced by personal preferences of the researcher, so
made this processing phase in a way more subjective.
The final range of parameter settings was based on the visual analyses of the de-noised paths.
Evaluation of each parameter was necessary to describe the evolution of the parameter. The
used range for further usage included a) a constant window width of five, a constant value was
chosen due to significant difference between surrounding values; b) a range between 0,8 and
2,0 was chosen for σd; and c) a range for σr was used between 0,5 and 0,8. Variation between
the each setting is relative small. So depending on the user’s preference based on the
magnitude of smoothing, a different setting could be chosen.
The speeddata was analysed and processed using a own constructed excel macro. The
program worked perfectly for small scale date processing and simple walked paths. The
introduction of a distinction between walking, slowing down and standing still phases was
needed because only the walked path was relevant for lameness testing, with the assumption
that slowing down and standing still is considered lameness independent. In the present study,
no further research was done regarding difference of slowing down or standing still due to
lameness. The distinction between the walking phase and the other phases is primarily based
on two criteria, namely a) starting from a specific place (plastic middle wall) and b) ending
based on five or more consecutive speeddatapoints. Both criteria can be used due to
respectively a) the place where the tests took place. If the test took place in another part of the
barn, the principle the starting point should be redesigned; and b) the simplicity of the walked
trajectory (“straight” line) and the synchronic walk-stop procedure. Taking previous arguments
into account, the program will have to be redesigned in the future, mostly because larger
sample sizes and abstract trajectories of the sow will cause problems.
57
Lameness
Each sow had to complete and repeat three test in order to be valid for data analysis. No
problems were found while testing for the sound sow and hind lame sow. However, the third
test of the front lame sow was invalid due to lack of continuous movement. While testing the
front lame sow, the researches saw that the animal was more resisting to walk, cooperate
during the tests and had the appearance that the animal was more tired in relation to the sound
sow, therefore, assuming that the sow would have suffered more from lameness in comparison
to the other sows during testing.
The moving average (MA) was integrated for analyses of the speeddata. The integration
should eliminate extreme speed variations. The MA works like a smoothing filter and reduces
these extremes. Data-analyses confirms the definition of Chou (1989), using a MA reduces the
variations of the speed and increases the smoothing. Both conclusions increase with an
increasing MA. The mean of the speed do not differ between the used MAs, and even applies
for both number of test and lameness status. The study concludes that no significant mutual
differences occur regarding the mean when using a moving average. A final choice of a moving
average equals to five was used for visual analysis of the speed.
The average speed of the sound sow is significant higher than both front lame sow and hind
lame sow. Furthermore, the front lame sow has a higher average speed than the hind lame
sow. In the present study, average speed values of respectively 1,205 m/s , 0,862 m/s and
0,734 m/s were found for the sound, front lame and hind lame sows. Similar conclusions were
found in the studies of Stavrakakis et al. (2015) and Grégoire et al. (2013), namely that a higher
speed was found for sound sows in relations to lame sows. Comparing only the lame sows,
similar conclusions were found in the study of Stavrakakis et al. (2015), indicating a higher
walking speed was found for front lame in relation with hind lame pigs. The only difference
between the studies is the magnitude of the difference between front lame and hind lame. The
present study indicates a significant lower walking speed of a hind lame sow. Note, this
conclusion has to be nuanced due to the small sample size. This suggests that for this specific
sow the hind limb lameness will have more influence on the walking speed due more pain or
because hind lameness makes walking for the animal much harder. On the other hand, some
of the speed related interpretations have to be put into perspective. In the present study, for
each lameness status, only one sow was used for testing. Therefore, speed could be more
sow related than lameness related, suggesting that the animal could always be faster/slower
relative to the other animals even if the animal does not suffer from lameness. Another
influence could be related to the setup of the tests. This study used food to lure the animal to
the other side of the barn. The speed of the sows could be related to the hungriness of the
animal, suggesting the animal would walk faster or slower because of respectively the desire
to eat or the animal is saturated. Fast learning of the animal, knowing food is on the other side
of the barn, could also affect the speed of the animal. A last aspect which could influence the
speed is the number of performed tests, suggesting that each repetition of the walking test, the
speed would drop significantly per test. The drop of speed would be due to tiredness because
of walking or due to the rise of pain because of lameness. In order to investigate the influence
58
hungriness, fast learning, tiredness or an increasing pain, the three test were compared
searching for mutual differences. The search did not confirm any of above quoted arguments,
therefore, no significant increase or decrease of the speed was found with an increasing of
test repetitions. However, due to the invalid third test of the front lame sow, the increase of
repetitions in combinations with no speed increase or decrease could suggest that at one point
the pain has rised to a degree which pain makes walking unbearable for the animal. The
assumption above that front lame sows could suffer more from lameness has to be re-
evaluated to a way that the invalid third test is more likely due to stubbornness of the animal
and/or the accumulation of pain which lead to a point of unbearable pain.
The walked path and possible relations with lameness were tested via the absolute value of
the deviation of the X-coordinates (ADx) to the middle of the barn. Knowing that the shortest
way to get from point A to point B is a straight line, could “extreme” deviations be due to
lameness. These deviations would be visually represented as a curved path and which is in
the present study closer to the sides of the corridor. Data analyses of ADx showed no
significant difference between sound and hind lame sows, but significant differences between
both front lame and sound sows and front lame and hind lame sows. Meaning that the walked
trajectory of the front lame sow deviates from the other two lameness statuses, and results in
a more curved path. Several explanations could be the reason for this deviation. First reason
could be the severe pain resulting in an abnormal trajectory. If pain would be the reason, the
consequence would be the search for protection resulting in relief of pain. Therefore, the sow
could walk closer to the walls of the corridor, forming some kind of protection if the animal
should fall due to severe pain. The search of protection should be further investigated in the
further. Another reason for the curved path of the front lame sow could be due to lameness on
the front limbs in specific. Front limb lameness could make walking in a straight line much
harder than lameness on the hind limbs, which had no significant deviations of ADx.
The influence of the number of performed test on the variation of ADx was tested, needed to
investigate the influence of tiredness, fast learning, hungriness or increasing pain. The present
study found no significant correlation between ADx and the number of the tests for the front
lame sow. However, correlation was found for sound and hind lame sows, meaning an
increasing ADx was found with each repetition. This shows the more the sow repeats the test,
the more the sow deviates from a “straight” path. Influence of tiredness and increasing pain
could cause this increasing variation. Using figure 58, the effect of an increasing testnumber
on the trajectory is more visible for the hind lame sow than for the sound sow, suggesting that
the influence of tiredness, increasing pain and/or more difficult walking will affect the hind lame
sow more. Due to an already significant higher ADx, no interaction was found for the front lame
sow. The relation between ADx and speed was tested for each test. No correlation was found
for test one and two, but a negative correlation was found for test three. This indicates an
increase of the repetition shows the speed is inversely proportional to ADx. However, no
correlation was already described above between speed and the number of tests. Thus, the
correlation between the speed and ADx for test three will more likely be influenced by a
significant higher ADx for test three.
59
7. Conclusion
The purpose of present study was to firstly finalise the tuning of an indoor positioning system,
which could register the trajectory (or path) of a sow in a group-housed gestation barn. The
tuning tests showed a decrease of accuracy when the tag was closer to an anchor. Based on
the data while tuning, the ILVO constructed a positioning algorithm which could be used for
further data analysis. The algorithm was integrated in an online data processing application,
needed to evaluate the recorded trajectory and to construct a filtered path. The filtered path is
a better representation of the actual walked trajectory of the animal during testing. The
recorded path differs to the filtered (denoised) path on the subject of a reduction of noise. The
algorithm was used in combination with a bilateral filter. The effect of the bilateral filter on the
data-output depends on three parameters, namely the window width and smoothing
parameters σd (spatial parameter) and σr (range parameter).
In order to choose an ideal setting for the algorithm in the online application, the filtered paths
were visually compared to evaluate each parameter. Two main criteria were used for the
parameter evaluation: path preservation and size of the difference between two consecutive
filtered paths. Each parameter affected the data in a different way. σd showed smoothing over
a larger area, in other words affected more datapoints at the same time, whereas σr affected
the data based on more datapoint specific smoothing. The window width mostly reinforced the
effect of the two other parameters significantly while increasing. Due to the variety of possible
settings and the low value of visual changes at low values, a range of values was chosen for
each parameter to serve as ideal settings.
A final phase of the present study included if the system was capable for practical studies. The
practical test was based on lameness comparing between non-lame and lame, with the
subdivision of hind lame and front lame, sows. The data analysis compared two main factors
for possible difference and abnormalities: a) the speed, which showed significant differences
between lame and non-lame sows; and b) deviation of the X-coordinates, which changed
significant with an increase of repetitions. The deviation was sow depended. Increasing pain,
tiredness, or difficult walking due to lameness could the reasons for mutual differences.
Future studies should focus on a better and less-time consuming parameter fine-tuning
progress. Furthermore, the tuning should be based on a more objective way to search for the
needed smoothing parameters. Next, the usage of the indoor positioning system for lameness
comparing should be used at a much bigger scale. The present study only tested the sow in a
forward movement and within a small time-frame. Future work should include other situations
like turning and sow interaction and this for a much bigger timeframe.
60
References
Abdat M, Wan TC, Supramaniam S (2010) Survey on indoor wireless positioning techniques: Towards adaptive systems. In 2010 International Conference on Distributed Frameworks for Multimedia Applications, pp 1-5.
Aitenbichler E, Muhlhauser M (2003) An IR local positioning system for smart items and devices. In 23rd International Conference on Distributed Computing Systems Workshops, 2003. Proceedings., pp 334-339.
Alarifi A, Al-Salman A, Alsaleh M, Alnafessah A, Al-Hadhrami S, Al-Ammar MA, Al-Khalifa HS (2016) Ultra Wideband Indoor Positioning Technologies: Analysis and Recent Advances. Sensors (Basel, Switzerland) 16: 707
Amundson I, Koutsoukos XD (2009) A Survey on Localization for Mobile Wireless Sensor Networks. In Mobile Entity Localization and Tracking in GPS-less Environnments: Second International Workshop, MELT 2009, Orlando, FL, USA, September 30, 2009. Proceedings, Fuller R, Koutsoukos XD (eds), pp 235-254. Berlin, Heidelberg: Springer Berlin Heidelberg
Anil SS, Anil L, Deen J (2009a) Effect of lameness on sow longevity. Journal of the American Veterinary Medical Association 235: 734-738
Anil SS, Anil L, Deen J, Baidoo SK, Walker RD (2007) factors associated with claw lesions in gestating sows. Journal of swine health and production 15: 78-83
Anil SS, Deen J, Baidoo SK, Wilson ME, Ward TL (2009b) Evaluation of the supplementation of complexed trace minerals on the number of claw lesions in breeding sows. Proceedings of the 12th Biennial Conference of the Australasian Pig ScienceAssociation
Bahl P, Padmanabhan VN (2000) RADAR: an in-building RF-based user location and tracking system. In Proceedings IEEE INFOCOM 2000. Conference on Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064), Vol. 2, pp 775-784 vol.772.
Bahl V, Padmanabhan V, Balachandran A (2000) Enhancement to the RADAR user location and tracking sytem. Microsoft Research
BARA M, McGOWAN M, O'BOYLE D, CAMERON R (1993) A study of the microbial flora of the anterior vagina of normal sows during different stages of the reproductive cycle. Aust Vet J 70: 256-259
Barnett JL, Hemsworth PH, Winfield CG, Fahy VA (1987) The effects of pregnancy and parity number on behavioural and physiological responses related to the welfare status of individual and group-housed pigs. Appl Anim Behav Sci 17: 229-243
Barsocchi P, Chessa S, Micheli A, Gallicchio C (2013) Forecast-Driven Enhancement of Received Signal Strength (RSS)-Based Localization Systems. ISPRS International Journal of Geo-Information 2: 978
61
Barsocchi P, Lenzi S, Chessa S, Furfari F (2012) Automatic virtual calibration of range-based indoor localization systems. Wireless Communications and Mobile Computing 12: 1546-1557
Beauregard SH, H. (2006) Pedestrian dead reckoning: a basis for personal positioning. In Proceedings of the 3rd Workshop on Positioning, Navigation an Communication
Benini L, Farella E, Guiducci C (2006) Wireless sensor networks: Enabling technology for ambient intelligence. Microelectronics Journal 37: 1639-1649
Bonde M, Rousing T, Badsberg JH, Sørensen JT (2004a) Associations between lying-down behaviour problems and body condition, limb disorders and skin lesions of lactating sows housed in farrowing crates in commercial sow herds. Livest Prod Sci 87: 179-187
Bonde M, Rousing TJ, Badsberg H, Sørensen JT (2004b) Associations between lying-down behaviour problems and body condition, limb disorders and skin lesions of lactating sows housed in farrowing crates in commercial sow herds. Livest Prod Sci 87
Bos E-J, van Riet MMJDI, Maes DDI, Millet SDI, Ampe BDI, Janssens GDI, Tuyttens FDIX (2016) Effect of rubber flooring on group-housed sows' gait and claw and skin lesions. (2016) JOURNAL OF ANIMAL SCIENCE
Bos EJ, Nalon E, Maes D, Ampe B, Buijs S, van Riet MMJ, Millet S, Janssens GPJ, Tuyttens FAM (2015) Effect of locomotion score on sows’ performances in a feed reward collection test. Animal 9: 1698-1703
Bouet M, Santos ALd (2008) RFID tags: Positioning principles and localization techniques. In 2008 1st IFIP Wireless Days, pp 1-5.
Cador C, Pol F, Hamoniaux M, Dorenlor V, Eveno E, Guyomarc’h C, Rose N (2014) Risk factors associated with leg disorders of gestating sows in different group-housing systems: A cross-sectional study in 108 farrow-to-finish farms in France. Preventive Veterinary Medicine 116: 102-110
Calderon Diaz J, Boyle L (2014) Effect of housing on rubber slat mats during pregnancy on the behaviour and welfare of sows in farrowing crates, Vol. 53.
Castro P, Chiu P, Kremenek T, Muntz R (2001) A Probabilistic Room Location Service for Wireless Networked Environments. In Ubicomp 2001: Ubiquitous Computing: International Conference Atlanta Georgia, USA, September 30–October 2, 2001 Proceedings, Abowd GD, Brumitt B, Shafer S (eds), pp 18-34. Berlin, Heidelberg: Springer Berlin Heidelberg
Cera KR, Mahan DC (1988) Effect of dietary calcium and phosphorus level sequences on performance, structural soundness and bone characteristics of growing-finishing swine. Journal of Animal Science 66: 1598-1605
Chen Y, Lymberopoulos D, Liu J, Priyantha B (2013) Indoor Localization Using FM Signals. IEEE Transactions on Mobile Computing 12: 1502-1517
Chon HD, Jun S, Jung H, Won An S (2004) Using RFID for Accurate Positioning. Journal of Global Positioning Systems 3: 32-39
Chou Y-L (1989) Statistical analysis for business and economics / Ya-Lun Chou.
62
Chowdhury TJS, Elkin C, Devabhaktuni V, Rawat DB, Oluoch J (2016) Advances on localization techniques for wireless sensor networks: A survey. Computer Networks 110: 284-305
Collin JM, O.; Lachapelle G. (2003) Indoor positioning system using accelerometry and high accuracy heading sensors.
Dandan S, Lin X (2016) A Hybrid Video and RFID Tracking System for Multiple Mice in Lab Environment. In 2016 3rd International Conference on Information Science and Control Engineering (ICISCE), pp 1198-1202.
Dardari D, Closas P, Djurić PM (2015) Indoor Tracking: Theory, Methods, and Technologies. IEEE Transactions on Vehicular Technology 64: 1263-1278
Durand F, Dorsey J (2002). Fast bilateral filtering for the display of high-dynamic-range images. Proceedings of the 29th annual conference on Computer graphics and interactive techniques; San Antonio, Texas. ACM.
Dwiyasa F, Lim MH (2016) A survey of problems and approaches in wireless-based indoor positioning. In 2016 International Conference on Indoor Positioning and Indoor Navigation (IPIN), pp 1-7.
Eckert K. (2005) Overview of Wireless LAN based Indoor Positioning Systems. Department of Computer Science IV, University of Mannheim, Germany.
Elmore MRP, Garner JP, Johnson AK, Richert BT, Pajor EA (2010) A flooring comparison: The impact of rubber mats on the health, behavior, and welfare of group-housed sows at breeding. Appl Anim Behav Sci 123: 7-15
Engblom L, Lundeheim N, Strandberg E, del P. Schneider M, Dalin AM, Andersson K (2008) Factors affecting length of productive life in Swedish commercial sows1. Journal of Animal Science 86: 432-441
Estienne M, Harper A, Knight J (2006) Reproductive traits in gilts housed individually or in groups during the first thirty days of gestation. Swine health prod 14: 241-246
Fang Z, Zhao Z, Cui X, Geng D, Du L, Pang C (2010) Localization in wireless sensor networks with known coordinate database. EURASIP J Wirel Commun Netw 2010: 1-17
Farid Z, Nordin R, Ismail M (2013) Recent Advances in Wireless Indoor Localization Techniques and System. Journal of Computer Networks and Communications 2013: 12
Fernando XN, Krishnan S, Sun H, Kazemi-Moud K (2003) Adaptive denoising at infrared wireless receivers, Vol. 5074, pp 199-207.
Finkenzeller K (2010) RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification, Vol. 3: Wiley Publishing.
63
FODeconomie. (2014) Kerncijfers landbouw.
Galler S, Schroeder J, Rahmatollahi G, Kyamakya K, Jobmann K (2006) Analysis and Practical Comparison of Wireless LAN and Ultra-Wideband Technologies for Advanced Localization. In 2006 IEEE/ION Position, Location, And Navigation Symposium, pp 198-203.
Gezici S, Zhi T, Giannakis GB, Kobayashi H, Molisch AF, Poor HV, Sahinoglu Z (2005) Localization via ultra-wideband radios: a look at positioning aspects for future sensor networks. IEEE Signal Processing Magazine 22: 70-84
Grandjot G (2007) Claw problems cost money. SUS - Schweinezucht und Schweinemast: 28-31
Grégoire J, Bergeron R, D'Allaire S, Meunier-Salaün MC, Devillers N (2013) Assessment of lameness in sows using gait, footprints, postural behaviour and foot lesion analysis. Animal 7: 1163-1173
Grimes JG (2008) GPS Performance Standard. The National Space-Based Positioning, Navigation,
and Timing (PNT) Executive Committee 4: 160
Gu Y, Lo A, Niemegeers I (2009) A survey of indoor positioning systems for wireless personal networks. IEEE Communications Surveys & Tutorials 11: 13-32
Hallberg J, Nilsson M, Synnes K (2003) Positioning with Bluetooth. In 10th International Conference on Telecommunications, 2003. ICT 2003., Vol. 2, pp 954-958 vol.952.
Harle R (2013) A Survey of Indoor Inertial Positioning Systems for Pedestrians. IEEE Communications Surveys & Tutorials 15: 1281-1293
Hatami A, Alavi B, Pahlavan K, Kanaan M (2006) A Comparative Performance Evaluation of Indoor Geolocation Technologies. Interdisciplinary Information Sciences 12: 133-146
Haute TV, Rossey J, Becue P, Poorter ED, Moerman I, Demeester P (2014) A hybrid indoor localization solution using a generic architectural framework for sparse distributed wireless sensor networks. In 2014 Federated Conference on Computer Science and Information Systems, pp 1009-1015.
Heinonen M, Peltoniemi O, Valros A (2013) Impact of lameness and claw lesions in sows on welfare, health and production. Livestock Science 156: 2-9
Hernández N, Ocaña M, Alonso J, Kim E (2017) Continuous Space Estimation: Increasing WiFi-Based Indoor Localization Resolution without Increasing the Site-Survey Effort. Sensors 17: 147
Hightower J, Borriello G (2001) Location Systems for Ubiquitous Computing. Computer 34: 57-66
Hightower J, Want R, Borriello G (2000) SpotON: An indoor 3D location sensing technology based on RF signal strength. UW CSE 00-02-02
Honkavirta V, Perala T, Ali-Loytty S, Piche R (2009) A comparative survey of WLAN location fingerprinting methods. In 2009 6th Workshop on Positioning, Navigation and Communication, pp 243-251.
Jayabharathy R, Varadarajan R, Prithiviraj V (2014) Performance comparison of UWB and Wi-Fi based indoor localization using TDOA. Journal of applied Sciences 14: 1564-1569
Jensen TB, Bonde MK, Kongsted AG, Toft N, Sørensen JT (2010) The interrelationships between clinical signs and their effect on involuntary culling among pregnant sows in group-housing systems. Animal 4: 1922-1928
Jinhong X, Zhi L, Yang Y, Dan L, Xu H (2011) Comparison and analysis of indoor wireless positioning techniques. In 2011 International Conference on Computer Science and Service System (CSSS), pp 293-296.
Jørgensen B (2000) Longevity of Breeding Sows in Relation to Leg Weakness Symptoms at Six Months of Age, Vol. 41.
Jung M, Fischer G, Weigel R, Ussmueller T (2013) Low Power 24 GHz ad hoc Networking System Based on TDOA for Indoor Localization. ISPRS International Journal of Geo-Information 2: 1122
Kaemarungsi K, Krishnamurthy P (2004) Properties of indoor received signal strength for WLAN location fingerprinting. In The First Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services, 2004. MOBIQUITOUS 2004., pp 14-23.
Kaemarungsi K, Krishnamurthy P (2012) Analysis of WLAN’s received signal strength indication for indoor location fingerprinting. Pervasive and Mobile Computing 8: 292-316
Khalajmehrabadi A, Gatsis N, Akopian D (2017) Modern WLAN Fingerprinting Indoor Positioning Methods and Deployment Challenges. IEEE Communications Surveys & Tutorials PP: 1-1
KilBride AL, Gillman CE, Green LE (2008) A cross-sectional study of the prevalence and associated risk factors for capped hock and the associations with bursitis in weaner, grower and finisher pigs from 93 commercial farms in England. Preventive Veterinary Medicine 83
KilBride AL, Gillman CE, Ossent P, Green LE (2009) A cross sectional study of prevalence, risk factors and population attributable fractions for foot and limb lesions in preweaning piglets on commercial farms in England. BMC Vet Res 5
Kim H-S, Seo W, Baek K-R (2017) Indoor Positioning System Using Magnetic Field Map Navigation and an Encoder System. Sensors 17: 651
Kim SH, Kim DH, Park HD (2010) Animal Situation Tracking Service Using RFID, GPS, and Sensors. In 2010 Second International Conference on Computer and Network Technology, pp 153-156.
King T, Lemelson H, Farber A, Effelsberg W (2009) BluePos: Positioning with Bluetooth. In 2009 IEEE International Symposium on Intelligent Signal Processing, pp 55-60.
65
Kleusberg AL, R.B. (1990) The limitations of GPS. GPS World Magazine 1(2): pp. 50-52
Kotanen A, Hannikainen M, Leppakoski H, Hamalainen TD (2003) Experiments on local positioning with Bluetooth. In Proceedings ITCC 2003. International Conference on Information Technology: Coding and Computing, pp 297-303.
Kroneman A, Vellenga L, van der Wilt FJ, Vermeer HM (1993) Review of health problems in group‐housed sows, with special emphasis on lameness. Veterinary Quarterly 15: 26-29
Kuriakose J, Amruth V, Nandhini NS (2014) A survey on localization of Wireless Sensor nodes. In International Conference on Information Communication and Embedded Systems (ICICES2014), pp 1-6.
Kyamakya K, Zapater A, Lue Z (2003) An indoor Bluetooth-based positioning system: concept, implementation and experimental evaluation. International
Ladd AM, Bekris KE, Rudys A, Kavraki LE, Wallach DS (2005) Robotics-based location sensing using wireless Ethernet. Wirel Netw 11: 189-204
LaMarca AdL, E. (2008) Location Systems: An Introduction to the Technology Behind Location Awareness. In Satyanarayanan M (ed.). Morgan & Claypool Publishers, p. 122.
Lanzisera S, Lin DT, Pister KSJ (2006) RF Time of Flight Ranging for Wireless Sensor Network Localization. In 2006 International Workshop on Intelligent Solutions in Embedded Systems, pp 1-12.
Li B, Salter J, Dempster AG, Rizos C indoor positioning techniques based on wireless LAN. LAN, FIRST IEEE INTERNATIONAL CONFERENCE ON WIRELESS BROADBAND AND ULTRA WIDEBAND COMMUNICATIONS: 7
Li Z, Dehaene W, Gielen G (2007) System Design for Ultra-Low-Power UWB-based Indoor Localization. In 2007 IEEE International Conference on Ultra-Wideband, pp 580-585.
Li Z, Liu C, Gao J, Li X (2016) An Improved WiFi/PDR Integrated System Using an Adaptive and Robust Filter for Indoor Localization. ISPRS International Journal of Geo-Information 5: 224
Liu H, Darabi H, Banerjee P, Liu J (2007) Survey of Wireless Indoor Positioning Techniques and Systems. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews) 37: 1067-1080
Liu Y, Yang Z, Wang X, Jian L (2010) Location, Localization, and Localizability. Journal of Computer Science and Technology 25: 274-297
Lu Y, Wei D, Lai Q, Li W, Yuan H (2016) A Context-Recognition-Aided PDR Localization Method Based on the Hidden Markov Model. Sensors 16: 2030
Main DCJ, Clegg J, Spatz A, Green LE (2000) Repeatability of a lameness scoring system for finishing pigs. Vet Rec 147
66
Manapure SS, Darabi H, Patel V, Banerjee P (2004) A comparative study of radio frequency-based indoor location sensing systems. In IEEE International Conference on Networking, Sensing and Control, 2004, Vol. 2, pp 1265-1270 Vol.1262.
Mautz R (2009) Overview of current indoor positioning systems. Geodezija ir Kartografija 35: 18-22
Menzies D, Patison KP, Fox DR, Swain DL (2016) A scoping study to assess the precision of an automated radiolocation animal tracking system. Computers and Electronics in Agriculture 124: 175-183
Millman S (2007) Sickness behaviour and its relevance to animal welfare assessment at the group level, Vol. 16.
Mohapatra D, Suma SB (2005) Survey of location based wireless services. In 2005 IEEE International Conference on Personal Wireless Communications, 2005. ICPWC 2005., pp 358-362.
Nalon E, Conte S, Maes D, Tuyttens FAM, Devillers N (2013) Assessment of lameness and claw lesions in sows. Livestock Science 156: 10-23
Ni LM, Yunhao L, Yiu Cho L, Patil AP (2003) LANDMARC: indoor location sensing using active RFID. In Proceedings of the First IEEE International Conference on Pervasive Computing and Communications, 2003. (PerCom 2003). pp 407-415.
Nuaimi KA, Kamel H (2011) A survey of indoor positioning systems and algorithms. In 2011 International Conference on Innovations in Information Technology, pp 185-190.
Oravainen J, Heinonen M, Seppä-Lassila L, Orro T, Tast A, Virolainen J, Peltoniemi O (2007) Factors Affecting Fertility in Loosely Housed Sows and Gilts: Vulvar Discharge Syndrome, Environment and Acute-phase Proteins, Vol. 41.
Paris S, Kornprobst P, Tumblin J, Fr, Durand d (2007). A gentle introduction to bilateral filtering and its applications. ACM SIGGRAPH 2007 courses; San Diego, California. ACM.
Pittet S, Renaudin V, Merminod B (2008) UWB and MEMS Based Indoor Navigation. Royal Institute of Navigation 61: 369-384
Pluym L, Van Nuffel A, Dewulf J, Cools A, Vangroenweghe F, Van Hoorelbeke S, Maes D (2011) Prevalence and risk factors of claw lesions and lameness in pregnant sows in two types of group housing. VETERINARNI MEDICINA 56: 101-109
Pluym LM, Maes D, Vangeyte J, Mertens K, Baert J, Van Weyenberg S, Millet S, Van Nuffel A (2013) Development of a system for automatic measurements of force and visual stance variables for objective lameness detection in sows: SowSIS. Biosystems Engineering 116: 64-74
Priyantha NB, Chakraborty A, Balakrishnan H (2000). The Cricket location-support system. Proceedings of the 6th annual international conference on Mobile computing and networking; Boston, Massachusetts, USA. ACM.
67
Razavi A, Valkama M, Lohan E (2016) Robust Statistical Approaches for RSS-Based Floor Detection in Indoor Localization. Sensors 16: 793
Richard B (1990) Why is the GPS signal so complex. GPS World Magazine 1(3): pp. 56-59
Sahinoglu Z, Gezici S, Güvenc I (2008) Ultra-wideband positioning sytems. Cambridge University Press
Sahinoglu Z, Gezici S, Gvenc I (2011) Ultra-wideband Positioning Systems: Theoretical Limits, Ranging Algorithms, and Protocols: Cambridge University Press.
Sánchez-Rodríguez D, Hernández-Morera P, Quinteiro J, Alonso-González I (2015) A Low Complexity System Based on Multiple Weighted Decision Trees for Indoor Localization. Sensors 15: 14809
Schenk E, Merchant-Forde J, Lay D. (2010) Sow lameness and longevity. USDA-ARS-MWA Livestock Behavior Research Unit, Purdue University, West Lafayette, IN.
Segers L, Van Bavegem D, De Winne S, Braeken A, Touhafi A, Steenhaut K (2015) An Ultrasonic Multiple-Access Ranging Core Based on Frequency Shift Keying Towards Indoor Localization. Sensors 15: 18641
Shang J, Hu X, Gu F, Wang D, Yu S (2015) Improvement Schemes for Indoor Mobile Location Estimation: A Survey. Mathematical Problems in Engineering 2015: 32
Sharp I, Yu K, Guo YJ (2009) GDOP Analysis for Positioning System Design. IEEE Transactions on Vehicular Technology 58: 3371-3382
Shin JG, Jeon WS, Jeong DG (2016) Anchor-free indoor localization based on floor plan. In 2016 International Conference on Information and Communication Technology Convergence (ICTC), pp 192-197.
Smith B (1988) Lameness in pigs associated with foot and limb disorders. In Practice 10: 113-117
Sonitor. (2015) Sonitor web page.
Stavrakakis S, Guy JH, Syranidis I, Johnson GR, Edwards SA (2015) Pre-clinical and clinical walking kinematics in female breeding pigs with lameness: A nested case-control cohort study. The Veterinary Journal 205: 38-43
Sun G, F. Fitzgerald R, J. Stalder K, A. Karriker L, K. Johnson A, J. Hoff S (2011) Development of an Embedded Microcomputer-based Force Plate System for Measuring Sow Weight Distribution and Detection of Lameness. Applied Engineering in Agriculture 27: 475
Swangmuang N, Krishnamurthy P (2008) An effective location fingerprint model for wireless indoor localization. Pervasive and Mobile Computing 4: 836-850
Thongthammachart S, Olesen H (2003) Bluetooth enables in-door mobile location services. In The 57th IEEE Semiannual Vehicular Technology Conference, 2003. VTC 2003-Spring., Vol. 3, pp 2023-2027 vol.2023.
timedomain. (2012) PulsOn P350 Active RFID tracking system.
Tiranti KI, Morrison RB (2006) Association between limb conformation and retention of sows through the second parity. American Journal of Veterinary Research 67: 505-509
Tomasi C, Manduchi R (1998) Bilateral filtering for gray and color images. In Sixth International Conference on Computer Vision (IEEE Cat. No.98CH36271), pp 839-846.
Tuta J, Juric M (2016) A Self-Adaptive Model-Based Wi-Fi Indoor Localization Method. Sensors 16: 2074
Tuyttens FAM (2005) The importance of straw for pig and cattle welfare: A review. Appl Anim Behav Sci 92: 261-282
Ubisense (2017) Ubisense website.
van der Meulen H, Buré R, R. dK, L. V. (1990) Orienterend onderzoek naar kreupelheid bij zeugen in groepshuisvesting. Instituut voor Mechanisatie, arbeid en gebouwen, Wageningen, p. 18.
Van Steenbergen EJ (1989) Description and evaluation of a linear scoring system for exterior traits in pigs. Livest Prod Sci 23: 163-181
Vorst P, Sommer J, Hoene C, Schneider P, Weiss C, Schairer T, Rosenstiel W, Zell A, Carle G (2008) Indoor Positioning via Three Different RF Technologies. In 4th European Workshop on RFID Systems and Technologies, pp 1-10.
Vossiek M, Wiebking L, Gulden P, Weighardt J, Hoffmann C (2003) Wireless local positioning - concepts, solutions, applications. In Radio and Wireless Conference, 2003. RAWCON '03. Proceedings, pp 219-224.
Want R, Hopper A, Falc V, #227, Gibbons J (1992) The active badge location system. ACM Trans Inf Syst 10: 91-102
Ward A, Steggles P, Curwen R, Webster P, Addlesee M, Newman J, Osborn P, Hodges S. (2002) The Bat Ultrasonic Location System. AT&T Laboratories Cambridge.
Weary DM, Huzzey JM, von Keyserlingk MAG (2009) BOARD-INVITED REVIEW: Using behavior to predict and identify ill health in animals1. Journal of Animal Science 87: 770-777
Werner Ma (2014) Indoor Location-Based Services Prerequisites and Foundations.
Willgert KJE, Brewster V, Wright AJ, Nevel A (2014) Risk factors of lameness in sows in England. Preventive Veterinary Medicine 113: 268-272
69
Xinrong L, Pahlavan K, Latva-aho M, Ylianttila M (2000) Comparison of indoor geolocation methods in DSSS and OFDM wireless LAN systems. In Vehicular Technology Conference Fall 2000. IEEE VTS Fall VTC2000. 52nd Vehicular Technology Conference (Cat. No.00CH37152), Vol. 6, pp 3015-3020 vol.3016.
Xu W (2014) Research and Implementation of Indoor Positioning Based on the Android Mobile Phone. Central China Normal University, Wuhan, China
Yassin A, Nasser Y, Awad M, Al-Dubai A, Liu R, Yuen C, Raulefs R, Aboutanios E (2017) Recent Advances in Indoor Localization: A Survey on Theoretical Approaches and Applications. IEEE Communications Surveys & Tutorials PP: 1-1
Yim J, Jeong S, Gwon K, Joo J (2010) Improvement of Kalman filters for WLAN based indoor tracking. Expert Syst Appl 37: 426-433
Zhang D, Xia F, Yang Z, Yao L, Zhao W (2010) Localization Technologies for Indoor Human Tracking. In 2010 5th International Conference on Future Information Technology, pp 1-6.
Zhang Y, Amin MG, Kaushik S (2007) Localization and Tracking of Passive RFID Tags Based on Direction Estimation. International Journal of Antennas and Propagation 2007: 9
