On-Line Boosting Tracking Based on Color Information and Haar-Like Features
Lufeng Yao, Jianzhong Wang, Fan Yang · 2013
In view of the learning and improvement ability of on-line boosting algorithm framework, implemented a target tracking system which mixed color information and Haar-like features together. To solve the shortage of the Haar-like features, first used color histogram equalization to the frame image to enhance the image's color information, then used Bhattacharyya coefficient to calculate the similarity of the candidate targets and the previous target, selected the one which had the maximum similarity as the tracked target and compared the similarity and confidence's variance to determine whether to update the target's color histogram or not. According to experiment, this method makes a good tracking in complex background and situation when the target takes a large rotation, better improves the tracking accuracy, and has certain robustness.