Post on 11-Apr-2017
transcript
VISUAL OBJECT TRACKING
USING PARTICLE CLUSTERING
Harindra W Pradhana
wisnu@intiteknologi.co.id
Content Summary
Key Purpose
Image Processing
Water level model
Color feature
New approach
Object Tracking
Pixel matching
Pixel clustering
Localization & tracking
Performance Analysis
Speed
Accuracy
Conclusions
A little BIT
Key Purpose
Object Tracking
Location relatively from the observer
Low-cost vision sensor
Low resolution
Limited frame rate
Image noise
Effective result
Image Processing
Image Processing – Water level model
Use brightness level
Detects hill and valley
(Harris, 1988)
Image Processing – color features
(Bretzner, 2002)
Use skin color as reference
Extract and calculate color features from RGB
components
Find similar color
Image Processing – New approach
Use both physical space & color space
Both space Euclidean distance measured for clustering
Introduce new color features
Eliminate brightness level better with webcam
Add 255 constants avoid similar distance value on different color
GR+=f 2551
BR+=f 2552
BG+=f 2553
Object Tracking
Object Tracking – Pixels matching
23
2
2
2
1 df+df+df=qp,d
Given particular
threshold
Calculate color
distance
Preview match pixels
Object Tracking – CF comparison
Colour
Green New CF
Orange both
Red Bretzner’s CF
Better detection
Adaptable in poor
lightning
Capable detecting
shiny surface
Object Tracking – Pixels clustering
Given particular
threshold
Count matching pixel
Decide good cluster
Preview the cluster
Object Tracking – Localization &
Tracking
Point out good pixels
in good clusters
Calculate center
weight
Keep last position
Calculate movement
Predict future position
Performance Analysis
Performance Analysis – Speed
Best result :
Resolution =
640x480
Frame rate =
30fps
Block size > 10p
Processing time =
33ms/frame
Performance Analysis – Effectiveness
Higher good
to bad pixel
detection ratio
compared to
Bretzner’s CF
Conclussion
Successfully locate object and track their movement
Clustering produce faster frame processing by
lowering computation size
New color feature effectively increase particle
detection
Capable to adapt poor lighting area by eliminating
brightness component
A Little BIT
BIT – IntiTeknologi
info@intiteknologi.co.id
http://intiteknologi.co.id