基于压缩感知的多特征实时跟踪

针对基于压缩感知的目标跟踪算法中存在的特征单一,在目标纹理变化、光照变化较大时跟踪不稳定、易丢失目标的问题,提出了多特征联合的实时跟踪算法。该算法以多个矩阵作为压缩感知中的投影矩阵,将压缩后的数据作为特征来提取出跟踪所需的多种特征。在更新过程中,针对不同特征在跟踪过程中的稳定性不同,采取不同速度的更新方法,使得在目标环境变化时跟踪的鲁棒性仍然很高。对不同视频的测试结果表明,提出的方法在目标运动、旋转、纹理变化和光照变化的情况下跟踪准确,在目标大小为70 pixel×100 pixel时平均帧速为23 frame/s,满足实时跟踪的要求。与单一特征的压缩感知算法相比,本算法在目标纹理和光照变化很大的情况下仍能完成稳定的实时跟踪。

Real-time tracking using multiple features based on compressive sensing

As traditional tracking algorithm based on compressive sensing can extrack few features and fails to track targets stably in textures and lightings changed,a real-time tracking algorithm using multi-features based on compressive sensing is proposed.The algorithm uses multiple matrixes as the projection matrix of the compressive sensing, and the compressed data as the multiple features to extract the multiple features needed by track. Because the feature stability is different in tracky processing,different update levels are taken to maintain the tracking robustness in varied target conditions. The proposed algorithm is tested with variant video sequences and the results show that the algorithm achieves stable tracking for the target moved or the light changed,and average computing frame rate is 23 frame/s when the target scale is 70 pixel×100 pixel.Obtained results satisfy the requirements of real-time tracking. As compared with the compressive tracking with single kind of feature, the algorithm can track stably under big changed lightings and target textures.