基于轨迹分析的行人异常行为识别

提出一种基于轨迹分段主题模型的异常行为检测方法。为了解决跟踪偏差引起的轨迹不连续问题,首先使用模糊聚类算法对所有的轨迹进行全局聚类,然后对每一类轨迹采用分段采样的方式对段内轨迹点使用主题模型LDA进行局部聚类;以最大概率的轨迹点作为视觉单词,每类轨迹表示成一系列视觉单词的集合,在此基础上建立局部隐马尔科夫模型HMM;最后通过轨迹匹配的方法进行异常轨迹识别。在CAVIAR数据库上的实验结果表明,该算法能识别多种异常行为,提高了异常行为检测的准确率。

Abnormal pedestrian behavior recognitionbased on trajectory analysis

We present a novel abnormal behavior detection method based on trajectory segment and topic model. In order to solve the problem of trajectory discontinuity caused by track deviation, all trajectories are firstly clustered by the fuzzy clustering algorithm, and then the sampling points of each segment of trajectory class are clustered by the latent Dirichlet allocation (LDA) topic model. The point of maximum probability is used as the visual word, and each trajectory class is represented by a series of visual words. In this sense, the local hidden Markov model (HMM) is established between every two visual words to detect abnormal trajectories through the matching path method. Experimental results show that the proposed method can identify a variety of abnormal behavior, and improve the accuracy of abnormal behavior detection in CAVIAR database.