摘要
针对复杂场景中人体遮挡导致人体异常行为检测效果差、实时性低的问题,提出一种基于生成对抗网络的人体异常行为检测算法。通过生成对抗网络预测视频帧,引入感知网络提取复杂场景视频流中的人体运动特征,并建立异常行为判决函数,实现对异常行为的准确检测。实验结果表明,该算法在复杂场景情况下可准确检测出视频中的异常行为,检测精度可达到96.7%,相比于时空自动编码器异常行为检测算法提升了5.5%;对于视频流的检测速度达到25 FPS,可实现对人体异常行为的实时检测。
In order to deal with the problem of human body occlusion in complex scenes where the detection effect of human body abnormal behaviour is poor,and the real-time performance is low,a human body abnormal behaviour detection algorithm based on generative adversarial network is proposed.In this algorithm,predictive frames are generated by generating adversarial networks,the perceptual network is introduced to construct the motion loss function,to effectively extract the human motion characteristics in the video stream of complex scenes,and to establish the abnormal behaviour judgment function in order to realize the accurate detection of abnormal behaviour.Experimental results show that the proposed method can accurately detect the abnormal behaviour in the video in complex scenes,and the accuracy can reach 96.7%,which is 5.5%higher than the spatiotemporal auto-encoder anomaly behaviour detection algorithm.The speed of video stream detection can reach 25 FPS,and therefore the real-time detection of abnormal behaviour of human body can be realized.
作者
唐浩漾
张小媛
王燕
杨青
TANG Haoyang;ZHANG Xiaoyuan;WANG Yan;YANG Qing(School of Automation,Xi'an University of Posts and Telecommunications,Xi'an 710121,China)
出处
《西安邮电大学学报》
2020年第3期92-97,共6页
Journal of Xi’an University of Posts and Telecommunications
基金
西安市科技计划项目(201805040YD18CG24)
陕西省教育厅专项科学研究计划项目(18JK0702)。
关键词
生成对抗网络
人体异常行为检测
感知网络
generative adversarial nets
human abnormal behaviour detection
perception network
