基于光流法优化的人群异常行为检测研究  

Research on crowd abnormal behavior detection based on optical flow optimization

作　　者：陈辛 杨江涛 许新云 CHEN Xin;YANG Jiangtao;XU Xinyun(School of Electronic Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学电子信息工程学院,山西太原030024

出　　处：《现代电子技术》2023年第12期168-174,共7页Modern Electronics Technique

基　　金：国家自然科学基金青年基金项目(61905172);山西省高等学校科技创新项目(2021L295);博士科研启动基金项目(20192015);来晋优秀博士奖励资金项目(20192068);校级研究生教育改革研究课题(JG2022009)。

摘　　要：危害社会公共安全的事件时有发生,但目前使用的监控摄像头存在画面不清晰、需人为查看且耗时长等问题,故研究高效稳定的智能监控系统对维护公共安全和秩序有着重要意义。针对人群中异常行为的检测问题,文中提出一种改进ORB光流算法,根据运动能量的变化对异常行为进行检测和判断。该方法利用SURF中的Hessian矩阵进行特征点提取,使其具有尺度不变性;采用盒式滤波器替换高斯滤波器进行图像降噪,提高计算速度;使用Shi-Tomasi算法对特征点进行筛选,再与金字塔LK光流法进行匹配,提高检测速度和精度;通过计算MRF模型的能量函数来判断人群中异常行为是否发生,并进行异常区域定位。最后,在UCSD数据集中对所提方法进行验证。多种场景下的ROC曲线和AUC值对比结果表明,所提算法具有良好的性能,可以有效地检测出人群中的异常行为。Events that endanger social and public safety occur from time to time,but the currently used monitoring cameras have issues such as unclear images,requiring manual viewing,and long time consumption.Therefore,an efficient and stable intelligent monitoring system is of great significance to maintain public safety and order.In allusion to the problem of abnormal behavior detection in crowd,an improved ORB optical flow algorithm is propsoed to detect and judge abnormal behavior according to the change of motion energy.The Hessian matrix in SURF is used to extract feature points,making it scale invariant.The box filter is used to replace the Gaussian filter for image denoising,improving the calculation speed.The Shi Tomasi algorithm is used to filter feature points,and then match with pyramid LK optical flow method to improve detection speed and accuracy.By calculating the energy function of MRF model,it can judge whether abnormal behavior occurs in the crowd and locate the abnormal area.The proposed method is verified in the UCSD dataset.The comparison results of ROC curves and AUC values in various scenarios indicate that the proposed algorithm has good performance and can effectively detect abnormal behavior in the population.

关 键 词：异常行为 群体检测 LK光流 ORB特征点 MRF模型 智能监控 

分 类 号：TN911.23-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]