An algorithm for real-time human tracking under dynamic scene

Abstract

To track human target real-time under dynamic scene, an automatic target tracking system based on IP PTZ camera was designed and an algorithm for tracking human target based on the system was proposed. Adaptive three-frame subtraction method was used to detect moving object and Camshift combined with Kalman filter was utilized to track object automatically. As Camshift algorithm suffers from low color saturation and similar color clutter in background, the color histogram of the target was strengthen and color and motion information was integrated to improve the performance. The problem the target was completely occluded had been analyzed and an occlusion handler was given. Experimental results show that the algorithm was robust and reliable in different scenes. © 2010 IEEE.