基于压缩感知的监控视频多目标自适应跟踪算法

A MULTI-OBJECTIVE ADAPTIVE TRACKING ALGORITHM FOR VIDEO SURVEILLANCE BASED ON COMPRESSED SENSING

针对压缩感知跟踪算法无法自动识别目标和易积累噪声导致目标丢失的问题,提出一种融合图像预处理的压缩跟踪算法。算法在初始化阶段,采用背景建模和背景差分法实现自动识别目标区域和多目标的判定。随后,使用压缩跟踪算法进行多目标的跟踪。在跟踪过程中,采用更新机制,减少噪声的积累,防止目标丢失。更新背景模型,使算法适应光线变化。算法在监控视频库上进行测试,并与压缩跟踪算法进行了对比,实验结果表明,提出的多目标自适应跟踪算法,在主客观性能方面较同类算法具有显著提高,其跟踪精度和鲁棒性方面得到了改善。

Aiming at the problem that the compressed perceptual tracking algorithm can not automatically identify the target and easy to accumulate the noise,this paper proposes a compression tracking algorithm which combines image preprocessing. In the initialization phase,background model and background difference method are used to realize the automatic recognition of the target area and multi-target decision. Subsequently,the multi-target tracking is performed using the compression tracking algorithm. In the tracking process,the use of updated mechanisms to reduce the accumulation of noise,to prevent the loss of goals. Update the background model so that the algorithm adapts to changes in light. The algorithm is tested on the monitor video library and compared with the compression tracking algorithm.Experimental results show that the proposed multi-objective adaptive tracking algorithm has a significant improvement in the subjective and objective performance compared with the similar algorithms,and its tracking accuracy and robustness are improved.