Design of Forward Collision Warning System
using Relative Acceleration Estimation and
Multi-Object Tracking
Bo-Chiuan ChenAssociate Professor, Department of Vehicle Engineering
Director, Vehicle Technology Research Center
National Taipei University of Technology, Taipei, Taiwan
• Education– Ph.D., Mechanical Engineering, U. of Michigan
– M.S., Electrical Engineering, U. Michigan
– M.S., Mechanical Engineering, U. Michigan
– B.S., Naval Architecture and Ocean Engineering, NTU
• Experience– Assistant Professor, Vehicle Eng. Dept., NTUT
– Strategy and Software Engineer, Powertrain Control Dept., Visteon, USA
– Research Assistant, Automotive Research Center, U. of Michigan
• Awards– Outstanding Young Engineer Award, SAE Taipei Section, 2008.
– Excellent Paper Awards, National Conference on Vehicle Engineering, Taiwan, 2008, 2009, 2010.
– Top 5 in Super Mileage Competition, SAE Taipei Section, 2004-2010.
• Research Areas– Active Safety, Vehicle Dynamics and Control, Hybrid Electric Vehicle,
Engine Control, Optimum Control
Self Introduction
Forward Collision Warning System (FCWS) 2
Research Areas
• Active Safety– Electronic Stability Control
– Rollover Prevention
– Forward Collision Warning
– Lane Departure Warning
– Side Collision Warning
– Auto Parking
• Vehicle Dynamic and Control– Light Weight Electric Vehicle
– Electric Differential
– Electric Power Steering
– ABS/TCS
– Semi-active Suspension
• Hybrid Electric Vehicle– Hybrid Electric Scooter
– Power Management System
• Engine Control– Idle Speed Control
– Engine Management System
4 3 2 1
Forward Collision Warning System (FCWS) 3
Statistics
• In U.S., the percentage of rear end collision in all collisions
was about 31.5% in 2009.
Source: Traffic Safety Facts 2009 Forward Collision Warning System (FCWS) 4
Forward Collision Warning System (FCWS)
Human Error
• 80% of drivers attempted no action in rear end collisions.
Source: SAE Paper 1999-01-0817 Source: AVEC 9437953
5
Existing Technologies
Audi
Ford
Honda
Volvo
BMW
Mercedes-Benz
VW
Toyota
Infinity
Forward Collision Warning System (FCWS) 6
Object Detection Systems
• Infrared Laser
– Transmit energy in the THz range (1 THz=1012 Hz).
– Superior angular resolution.
– Limited performance due to atmospheric effects, such as fog and
rain.
– Does not perform well on wet objects or targets whose surface
roughness is the order of the laser wavelength (10.6 microns).
• Microwave/Millimeter wave radar
– Transmit energy in the tens of GHz range (1 GHz = 109 Hz)
– Better adverse weather penetration than active laser systems.
• Camera
– Usable distance accuracy for short-range detection (less than 55 m)
– Poor accuracy for long-range detection due to the pixel resolution
Forward Collision Warning System (FCWS) 7
Forward Collision Warning System
• Autonomous Solutions:
– Identify valid target and measure range, range rate, and vehicle
speed (10Hz or faster).
– Vehicle path prediction
– Issue warning based on
• Time-to-Collision
• Time-headway (time gap)
• Threshold distance
From vehicle speed, road friction,
and human delays such as “blank
time” and “judgment time,” a “safe
following distance” d can be constructed.
RTTC
R
host
RTHW
V
Source: SAE 98PC-417
Forward Collision Warning System (FCWS) 8
Relative Acceleration Estimation
• Lee and Peng (2005) mentioned that the leading vehicle
acceleration is a critical step for developing practical
collision warning/avoidance systems.
• Good estimation of relative acceleration is the key to reduce
the false alarm of FCWS.
– Dagan et al. (2004) calculated TTC from the momentary TTC
defined by Hayward and its derivative, which is closely related to
relative acceleration and can be computed by the scale change in
the image.
– Araki et al. (1996) applied a 3-state Kalman filter to estimate relative
velocity and acceleration.
2
ˆˆ ˆˆ ˆ ˆ ˆ( 1) ( ) ( ( ) ( )), ( )
1 0.5 0
0 1 , 0 , 1 0 0
0 0 1 1
T
k k y k k k R V A
T T
T
x Ax L Cx x
A G C
Forward Collision Warning System (FCWS) 9
Multi-object Tracking
• Moon et al. (2009) proposed primary target selection
– A yaw-rate based subject vehicle’s lane detection, a motion based
analysis, and an integration process.
– Primary target might be changed quite often during transient yaw
motion.
Forward Collision Warning System (FCWS) 10
Proposed Approach
• Recursive Least Squares
R̂ t
t kT
true TTC
estimated TTC
0
R t
rela
tive
dis
tan
ce (
m)
time (sec)
2 ˆˆ T
R R R RR k a k t b k t c k k k φ θ
2 ˆˆ T
D D D DD k a k t b k t c k k k φ θ
ˆ T
R R R Ra k b k c k θ
ˆ T
D D D Da k b k c k θ
2 1 ,T
k t t φ t kT
1
1 1Tk k k k k k
K P φ I φ P φ
ˆ ˆ ˆ1 1Tk k k y k k k θ θ K φ θ
1Tk k k k P I K φ P
1
0 0
3 3
ˆ 0 T
k k
k k k y k
θ φ φ φ
0
X
Y
(0,0)
ˆ 1R k
ˆ 1D k
ˆpredictR k
ˆpredictD k
( )R k
D k
r
overflow when t is large
reset to zero during safe maneuvers
Forward Collision Warning System (FCWS) 11
Proposed Approach
• Variable forgetting factor
– Large forgetting factor is suitable for small relative acceleration.
However, the estimation performance deteriorates with large relative
accelerations.
– Small forgetting factor is suitable for large relative accelerations.
However, it might produce noisier estimations for small relative
acceleration.
– Adjust the forgetting factor according to the estimated relative
acceleration, i.e. variable forgetting factor, might be a good solution.
ˆR R Rd m A ˆ
D D Dd m
2 ˆˆ ˆˆR R Ra b cR k k t k t k
ˆˆ ˆ2 R RV k a k t b
ˆ ˆ2 RA k a k
2 ˆˆ ˆ ˆ2
ˆ
V k V k A k R kTTC k
A k
Forward Collision Warning System (FCWS) 12
Optimization
• Kalman Smoothing
– the forward filtered data contains undesirable time delays,
– a backward Kalman filter is constructed to cancel its effect
– The averaged data is then used as the ground truth.
– The same smoothing procedure is applied to the relative orientation.
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( )
0 1 0 0 0 0 0
0 0 0 0 0 1 01 0 0 0 0 1 0
, , , 1 0 0 1 0 0 00 0 1 0 0 0 1
0 0 0 0 1 0 0
0 0 0 0 0 0 1
KF f KF
KF
T
KF F F L L
f
t t t
t t t
t V t a t R t V t a t
x A x Gw
y Cx Nv
x
A G C N
ˆ ˆ ˆ( 1) ( ) [ ( ) ( )]KF df KF s KFk k k k x A x L y Cx
ˆ ˆ ˆ( 1) ( ) [ ( ) ( )],db sk k k k z A z L y Cz 1[ ]T T
db df df df
A A A A
Forward Collision Warning System (FCWS) 13
Proposed Approach
• Multi-object Tracking
F
(k)
F
(k-1)
H
min
max
0
H
min
F
(k)
0
F
(k-1)
max
min max
if <
if and
possible collision, warning on
else
safe, warning off
else
safe, warning off
end
th
th
TTC TTC
count count
v
r
4secthTTC 3.5secthTTC
2thcount 3thcount
possible collision safe
1 ˆ ˆ ˆˆtan ,R V D
Forward Collision Warning System (FCWS) 14
Experimental Setup
CAN analyzer
VBOX IMU
ibeo LUX laserscanner
Steering angle sensor
Mototron
Data
acquisition
USB
cameraData logging
Analog
CAN Bus
USB
Forward Collision Warning System (FCWS) 15
ISO 15623
Longitudinal target discrimination ability test
Straight road lateral target discrimination ability test
Curved test track and target
discrimination ability test
Forward Collision Warning System (FCWS) 16
ISO Test -1
Forward Collision Warning System (FCWS) 17
2 4 6 8 10 12 14 16 18 20 220
10
20
30
40
rela
tive
dis
tan
ce
(m
)
2 4 6 8 10 12 14 16 18 20 22
-5
0
5
rela
tive
ve
l. (
m/s
)
2 4 6 8 10 12 14 16 18 20 22-5
0
5
time (s)
rela
tive
acce
l. (
m/s
2)
3 state kalman
RLS (variable )
Kalman smooth
ISO Test -1
2 4 6 8 10 12 14 16 18 20 22250
260
270
280
290
an
gle
(d
eg
)
relative velocity vector
2 4 6 8 10 12 14 16 18 20 220
1
2
3
4
time (s)T
TC
(s)
conventional
3 state kalman
RLS(variable )
Kalman smooth
warning
8.5s 11s 17s 18s 8.5s 11s 17s 18s
Forward Collision Warning System (FCWS) 18
ISO Test - 2
Forward Collision Warning System (FCWS) 19
ISO Test - 2
2 4 6 8 10 12 14 16 18 20 220
10
20
30
40
50
rela
tive
dis
tan
ce
(m
)
2 4 6 8 10 12 14 16 18 20 22
-15
-10
-5
0
rela
tive
ve
l. (
m/s
)
2 4 6 8 10 12 14 16 18 20 22
-5
0
5
10
time (s)
rela
tive
acce
l. (
m/s
2)
3 state kalman
RLS (variable )
Kalman smooth
2 4 6 8 10 12 14 16 18 20 22250
260
270
280
290
relative velocity vector (main lane)
an
gle
(d
eg
)2 4 6 8 10 12 14 16 18 20 22
260
270
280
290
300
310
relative velocity vector (adjacent lane)
an
gle
(d
eg
)
2 4 6 8 10 12 14 16 18 20 220
1
2
3
4
time (s)
TT
C (
s)
conventional
3 state kalman
RLS (variable )
Kalman smooth
9s 12s 15s 16.5s 9s 12s 15s 16.5s
adjacent
adjacent
adjacent adjacent
Forward Collision Warning System (FCWS) 20
ISO Test - 3
Forward Collision Warning System (FCWS) 21
ISO Test - 3
2 4 6 8 10 12 14 16 18 20 22
10
20
30
40
50
rela
tive
dis
tan
ce
(m
)
2 4 6 8 10 12 14 16 18 20 22
-15
-10
-5
0
rela
tive
ve
l. (
m/s
)
2 4 6 8 10 12 14 16 18 20 22
-10
-5
0
5
time (s)
rela
tive
acce
l. (
m/s
2)
3 state kalman
RLS (variable )
Kalman smooth
2 4 6 8 10 12 14 16 18 20 22250
260
270
280
290
relative velocity vector (main lane)
an
gle
(d
eg
)2 4 6 8 10 12 14 16 18 20 22
260
270
280
290
300
310
relative velocity vector (adjacent lane)
an
gle
(d
eg
)
2 4 6 8 10 12 14 16 18 20 220
1
2
3
4
time (s)
TT
C (
s)
conventional
3 state kalman
RLS (variable )
Kalman smooth
6s 9.5s 13s 15.5s 6s 9.5s 13s 15.5s
adjacent
adjacent
adjacentadjacent
Forward Collision Warning System (FCWS) 22
Expressway Tests
Merging TrafficLane Change
Cut-in Aggressive Lane Change
Forward Collision Warning System (FCWS) 23
Conclusions
• Relative acceleration is considered to improve the accuracy
of TTC estimation.
– Recursive least square technique with variable forgetting factor is
used to estimate coefficients of two second order polynomials for the
relative distance and relative orientation, respectively.
• The region of interest for FCWS is extended from the main
lane to adjacent lanes.
– According to the measured relative distance and relative orientation,
a multi-object tracking algorithm is developed in this research.
– When the TTC is below the threshold value, relative velocity vector
is used to determine if there is an impending threat for collision.
• Experimental results show that the proposed algorithm
– can pass all 3 tests of ISO 15623
– and issue valid warnings to the driver without false alarms for the
expressway tests.
Forward Collision Warning System (FCWS) 24
Future Works
• Automatic evasive maneuver for collision avoidance
Source: Continental Emergency Steer Assist 2010Forward Collision Warning System (FCWS) 25
Q & A
Forward Collision Warning System (FCWS) 26