一种基于运动相似熵的人群异常行为检测-信息通信学术期刊网.PDF

研究与开发 一种基于运动相似熵的人群异常行为检测 李斐，陈恳，李萌，郭春梅 （宁波大学信息科学与工程学院，浙江 宁波 315211 ） 摘 要：实现对人群异常事件的检测是图形处理在智能视频监控领域的重要研究内容。提出了一种基于运动 相似性熵（EMS ）的人群异常行为检测算法。该算法在对视频图像进行光流计算的基础上，以底层光流块为 基本单位获取场景运动信息，根据社会网络模型的概念，提出构建场景的运动网络模型（MNM ），完成对场 景粒子运动相似性的划分，并在时间域上计算 MNM 的粒子分布熵值 EMS ，最后将得到的图像熵与设置合理 的阈值相比，判断异常行为是否发生。实验证明，该算法可有效检测异常行为，与其他经典检测算法相比有 较大优势。 关键词：人群异常检测；运动相似性熵；运动网络模型；图像熵；光流法 中图分类号：TP391 文献标识码：A doi：10.3969/j.issn.1000−0801.2017117 Crowd abnormal behavior detection based on motion similar entropy LI Fei, CHEN Ken, LI Meng, GUO Chunmei Institute of Communication Technology, Ningbo University, Ningbo 315211, China Abstract: It is an important research content of graphic processing in the field of intelligent video surveillance to detect abnormal events. An algorithm based on entropy of motion similarity (EMS) to detect abnormal behavior was proposed. Based on the optical flow algorithm, taking the bottom flow block as the basic unit to get the scene motion information, according to the concept of social network model, the construction scene of the motion network model (MNM) was proposed, the division of the scene particles motion similarity was completed, and the distribution EMS of MNM was calculated in the time domain. Finally, the obtained image entropy was compared with the reasonable threshold, to determine whether abnormal behavior occured. Experimental results indicate that the proposed algo- rithm can detect abnormal behavior effectively and show promising performance while comparing with the state of the art methods. Key words: crowd abnormal detection, entropy of motion similarity, motion network mo