Computer Engineering and Applications ›› 2017, Vol. 53 ›› Issue (8): 165-169.DOI: 10.3778/j.issn.1002-8331.1510-0065

Previous Articles     Next Articles

Pedestrian abnormal behavior analysis based on optimized sparse reconstruction algorithm

TANG Chunming, LU Yongwei   

  1. School of Electronics and Information Engineering, Tianjin Polytechnic University, Tianjin 300387, China
  • Online:2017-04-15 Published:2017-04-28



  1. 天津工业大学 电子与信息工程学院,天津 300387

Abstract: In order to identify abnormal behavior in the video surveillance, first of all, it tracks the pedestrian, and then analyzes the trajectory to determine whether there is abnormal behavior. In the pedestrian tracking, the Kalman filter and spatial-temporal context algorithm are combined together, which can effectively avoid the shelter problem in complicated background. In the analysis of abnormal behavior, the trajectory is classified according to the shape to get the normal trajectory scenario set. It analyzes the trajectory by optimized sparse reconstruction algorithm and distinguishes normal or abnormal according to the reconstruction residual. The experimental results show that the proposed method has higher recognition rate compared with the original method.

Key words: video monitoring sequence, target tracking, spatial-temporal context, abnormal analysis, sparse reconstruction algorithm

摘要: 针对视频监控中行人异常行为识别问题,首先对行人进行跟踪,然后对跟踪得到的轨迹进行分析,最后判断行人行为是否存在异常。在行人跟踪方面,在时空上下文跟踪算法的基础上结合卡尔曼滤波器,有效改进了复杂背景中的遮挡问题。在异常分析方面,将跟踪得到的目标轨迹按照轨迹形状进行分类,得到场景中的正常轨迹集;将这些轨迹集作为后续处理的训练样本集,通过改进的稀疏重构算法对轨迹进行分析,利用重构误差来判断异常。五段视频序列的测试结果表明,该方法与改进前的方法相比具有较高的识别率。

关键词: 视频监控序列, 目标跟踪, 时空上下文, 异常分析, 稀疏重构算法