基于特征加权的快速压缩感知跟踪

FAST COMPRESSIVE SENSING TRACKING BASED ON FEATURE WEIGHTING

基于压缩感知的目标跟踪算法具有简单、实时、高效的特点。快速压缩感知目标跟踪算法FCT(Fast Compressive Tracking)生成目标高维特征未考虑不同尺度滤波器生成特征的有效性,目标与候选样本之间的相似性度量仅考虑简单叠加,在目标受到光照、遮挡等外界因素的影响下易使跟踪结果出现偏差。针对这些问题,提出一种基于特征加权的快速压缩感知跟踪算法。该算法根据滤波器尺度,自适应地分配权值,生成目标高维特征。算法将候选样本各维压缩特征分类为目标压缩特征的可能性与贝叶斯分类器输出相乘,作为目标与候选样本之间的相似性度量。实验结果表明,提出的方法在目标受到光照、遮挡等外界因素的影响下具有更强的鲁棒性。

The target tracking algorithm based on compressive sensing is simple, real-time and efficient. Fast compressive tracking algorithm （ FCT） generates high dimensional features of the target without considering the effectiveness of different scale filter generation features, and the similarity measure between target and candidate sample only considers simple superposition. It is easy to make the tracking result deviate under the influence of external factors such as illumination and occlusion. Aiming at these problems, a fast compressive sensing tracking algorithm based on feature weighting is proposed. The algorithm adaptively assigns weights according to the filter scale, and generates high dimensional features of the target. The likelihood of the algorithm to classify each dimension compression feature of the candidate sample as the target compression feature is multiplied by the Bayesian classifier output, which is used as a measure of similarity between the target and the candidate sample. The experimental results show that the proposed method is more robust to the influence of external factors such as illumination and occlusion.