基于特征贡献度的Mean Shift 目标跟踪

为有效提高Mean Shift算法的模板匹配精确度,采用基于特征贡献度的Mean Shift目标跟踪方法,对不同贡献度的特征向量赋予不同的权重,以彰显目标特征、抑制背景因素.分别介绍传统Mean Shift目标跟踪算法和基于特征贡献度的Mean Shift算法,并针对多组视频进行实验验证与分析.结果表明:改进后的Mean Shift算法不仅能提高跟踪精度、提升系统的鲁棒性,而且对640 pixel×480 pixel大小的视频处理平均帧速度为22 frames/s,满足实时跟踪要求.

Mean Shift Based on Feature Contribution

To improve template matching accuracy of the Mean Shift framework, we proposed Mean Shift target tracking based on feature contribution. The feature vectors of different contributions are endowed with different weights to highlight the target feature and the background factor. Mean Shift target tracking algorithm and Mean Shift algorithm based on feature contribution are introduced, and the experimental verification and analysis for multi group video are presented. Result shows that the improved Mean Shift algorithm not only improve tracking accuracy, enhanced system robustness, but also maintained an average processing speed as 22 frames/s for a video sized as 640 pixel×480pixel, meet the requirements of real-time tracking.