|
Qi Zhao, Ph.D.
Postdoctoral Fellow, Koch Lab Advisor: Dr. Christof Koch
|
[Home] [Research] [Publications] [Talks] [Teaching] [Misc] [Curriculum Vitae]
A Motion Observable Representation Using Color Correlogram and Its Applications to Visual Tracking
|
Introduction
|
This work presents a special form of color correlogram as representation for object tracking and carries out a motion observability analysis to obtain the optimal correlogram in a kernel based tracking framework. Compared with the color histogram, where the position information of each pixel is ignored, a simplified color correlogram (SCC) representation encodes the spatial information explicitly and enables an estimation algorithm to recover the object orientation. In this paper, based on the SCC representation, the mean shift algorithm is developed in a translation–rotation joint domain to track the positions and orientations of objects. The ability of the SCC in detecting and estimating object motion is analyzed and a principled way to obtain the optimal SCC as object representation is proposed to ensure reliable tracking. |
|
|
|
|
Experiments
(click the images to play videos) |
Fig. 1 Car-Chasing Sequences with object rotations, heavy occlusions, background clutters and scale changes [From left to right: (a) (b) (c).] (a) & (b): Tracking results using the original Mean Shift (MS) tracker - It tends to lose track when the car makes turns (4.(a)), or background scene is cluttered(4.(b)). (c): using the SCC tracker with optimal correlogram - It deals with the challenging issues elegantly.
Fig. 2 Person-Cart Sequence with partial occlusions and camera motion Fig. 3 PETS 2001 Sequence Left: using the MS tracker - The tracker can not keep track of the entire objects due to its incapability of detecting rotational motion. Right: using the SCC tracker - the optimal SCC based tracker maintains a secure focus on the object throughout the sequence. The restrictions brought up by certain object shapes and/or camera viewpoints are successfully removed.
Fig. 4 Multiple People Sequence: the optimal SCC based tracker is robust against interactions between objects with similar color distributions, but different color spacial arrangements.
  Fig. 5 Arm Sequence Left: using the MS tracker - The tracker drifts along the lower-arm, due to color similarity in the region. On the other hand, if an elongated kernel is imposed on the whole lower-arm region, it would easily get lost when the part begins rotating. Right: using the SCC tracker - The problem is solved.
Fig. 6 Handset Sequence (source video from Fan and Wu) with wide rotation (more than four revolutions in 135 frames)
Fig. 7 Multiple Human Parts Tracking : satisfactory results indicate the algorithm's potential in being a useful module in any human tracking or behavior analysis tasks.
Fig. 8 Face Tracking: the SCC based tracker succeeds in tracking a wide range of head tilt. |
|
|
|
|
References |
|
A. Vehicle and Pedestrian Tracking 
