Real-time tracking of non-rigid objects using mean shift

Details Real-time tracking of non-rigid objects using mean shift Real-time tracking of non-rigid objects using mean shift

2000-02-14

2003-07-08

Chang, Jon (Department: 2623)

Image analysis

Applications

Target tracking or detecting

C382S107000, C382S190000, C348S416100

Reexamination Certificate

active

06590999

ABSTRACT:

FIELD OF THE INVENTION
The present invention relates to a method and apparatus for real-time tracking of non-rigid objects appearing as an image in successive image frames. A degree of similarity is computed between a given target in a first frame and a candidate target in a successive frame, in the preferred embodiment the degree being expressed by a metric derived from the Bhattacharyya coefficient. A gradient vector corresponding to a maximization of the Bhattacharyya coefficient is used to derive the most probable location of the candidate target in the successive frame. The invention provides a practical, fast and efficient solution which achieves real-time performance (30 frames/sec.), and is appropriate for tracking a large variety of targets. The invention provides tracking which is robust to partial occlusion, significant clutter, target scale variations, rotations in depth, and changes in camera position (zoom, tilt and pan).
DESCRIPTION OF THE PRIOR ART
The efficient tracking of visual features in complex environments, such as where the camera itself is moving (and thus both the foreground and background images change with each frame), is a challenging task for the vision community. Real-time applications such as surveillance and monitoring, perceptual user interfaces, smart rooms, and video compression all require the ability to track moving objects. The computational complexity of the tracker is critical for most applications, since only a small percentage of a system's resources are typically allocated for tracking, while the rest of the resources are assigned to preprocessing stages or to higher-level tasks such as recognition, trajectory interpretation, and reasoning.
SUMMARY OF THE INVENTION
The present invention describes a new approach for real-time tracking of a non-rigid target. The tracking is based on visual features, such as color and/or texture, where statistical distributions of those features characterize the target. A degree of similarity is computed between a given target in a first frame and a candidate target in a successive frame, the degree being expressed by a metric derived from the Bhattacharyya coefficient. A gradient vector corresponding to a maximization of the Bhattacharyya coefficient is used to derive the most probable location of the candidate target in the successive frame.
More specifically, the method and apparatus in accordance with the present invention for real-time tracking of a target which appears in a plurality of successive image frames, comprises:
a) Developing a statistical distribution as a characterization of a visual feature of each of a given target in a first frame and a candidate target in a successive frame,
b) Computing a degree of similarity between the given target and the candidate target, the degree being expressed by a metric derived from the Bhattacharyya coefficient, and
c) Applying an iterative comparative procedure to the degrees of similarity computed in step b), the iterations being based on a gradient vector corresponding to a maximization of the Bhattacharyya coefficient in order to shift the location of candidate target in the successive frame, to derive as the location of the candidate target in the successive frame that location which has characteristics most similar to the characteristics of the given target in the first frame.


REFERENCES:
patent: 4868871 (1989-09-01), Watson, III
patent: 5946041 (1999-08-01), Morita
patent: 6169817 (2001-01-01), Parker et al.
“The Divergence and Bhattacharyya Distance Measures in Signal Selection” by Thomas Kailath, IEEE Transactions on Communication Technology, Feb. 1967, pp. 52-60.

Affiliated with

Also associated with

Rating

Real-time tracking of non-rigid objects using mean shift does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Real-time tracking of non-rigid objects using mean shift, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Real-time tracking of non-rigid objects using mean shift will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3083445