Particle Filter-based Multi-part Human Tracking with Failure Adjustment in Video Sequences

Abstract

The study presents a human tracking system in video sequences. To track the target, we. first detect humans in a video according to a Gaussian background model. We then track the humans by using color histogram as the features and using particle filters as the tracking kernel. Since a human is not a rigid object, his appearance may be greatly affected by his motion. In our applications, human bodies imaged are generally large. We decompose each human body into three parts: head, torso, and hip-leg, represent them by three shrunk rectangles, and track them by particle filters. In this way we can reduce possible tracking failures by checking the interrelationship among these three parts. We use support vector machines (SVM) to detect tracking failures and abnormal body parts since the abnormal situations are very diversified and cannot be easily encoded in rules. If a single part is abnormal, its position can be adjusted from the other parts and tracked using the system dynamic model. If two or three parts are abnormal, we re-initialize the tracking process of the three parts around their predicted positions. By testing on 22 video clips from six scenes, the experimental results showed that our three-part tracking system with failure detection and correction can track correctly about 95% persons until the 105th frame. With respect to the body parts, our system has about 95%, 83%, and 91% tracking rates for head, torso, and hip-leg respectively until the 105th frame. The tracking rate of a human increase 20% comparing with that of the whole-body tracker. These rates show the effectiveness of the proposed system.