基于HOG的实时压缩跟踪研究

Research on real time compression tracking based on HOG

压缩跟踪利用压缩感知理论将Harr类特征投影到低位空间中,实现了实时性跟踪。但该方法提取是关于颜色信息的特征,易受光照影响,在光照剧烈变化的场合会跟踪失败。基于此,提出一种基于梯度方向直方图实时压缩跟踪算法。该算法采用HOG特征取代Harr类特征,增强对光照的不敏感性,提高了跟踪鲁棒性。通过不同视频的测试结果表明,文章提出的方法在光照剧烈变化、形变等情况下能准确地跟踪目标,且平均帧率15 frame/s,基本满足实时性要求。

Compression Tracking projects Harr features onto the lower space to achieve a real-time tracking by using the compressed sensing theory. However, the feature which is extracted by this method is about color character, which is sensitive to light. In the occasion of the light with dramatic changes, it easily leads to the failure tracking. Based on the above discussion, we propose a HOG-based real-time compression tracking. In this paper, we use HOG features to replace Harr features to enhance the insensitivity to light, improve the tracking robustness. Through different video test results ,they show that the proposed method could track the target accurately in the light of dramatic changes and deformation circumstances, and the average frame rate is 15 frame/s, which is basically meet the real-time requirements.