Abstract: As target tracking algorithm based on compressive sensing can extract few features and fails to track targets stably when textures or lightings change much, a target tracking algorithm based on the complementary feature weighting of compressive sensing is proposed. The algorithm extracts two types of complementary texture features and gray average features using two random measurement matrices, and calculates these two types of features’ weight according to the classification results, using the selected large weights feature to find the target in next frame. Because the feature stability is different in track processing, different update levels are taken. Results of tests on variant video sequences show that the proposed algorithm is capable of accurately capturing the tracking target, and obtained results satisfy the requirements of real-time tracking. Compared with the related algorithms, the proposed algorithm can hold a stronger robustness when target textures or lightings change much.

Key words: compressive sensing, target tracking, complementary feature, multi-feature weighting, large weights feature

摘要： 针对基于压缩感知的目标跟踪算法中存在特征单一，在目标纹理或光照变化较大时跟踪不稳定的问题，提出了基于压缩感知的互补特征加权目标跟踪算法。该算法通过两个随机测量矩阵提取出两类互补的纹理特征和灰度均值特征，计算这两类特征对样本的分类结果并更新特征的权值，使用所选取的大权值特征寻找目标在下一帧的位置。在分类器更新过程中，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。对不同视频的实验结果表明，提出的算法跟踪准确，且满足实时性的要求。与相关算法相比，新算法在目标纹理或光照变化很大的情况下具有更强的鲁棒性。

关键词: 压缩感知, 目标跟踪, 互补特征, 特征加权, 大权值特征