基于关键帧定位和时空图卷积的异常行为识别

Abnormal Behavior Recognition Based on Key Frame Location and Spatial-temporal Graph Convolution

为提高监控视频中行人异常行为检测效率,提出了结合关键帧定位和时空图卷积的异常行为识别方法。该方法在人体骨架关键点检测的基础上,采用关键点运动特性定位视频中行人异常行为关键序列,利用时空图卷积网络可以提取行人时空特征的优点,在关键帧序列上构建人体骨架时空图,同时建立基于瓶颈残差模块的时空图卷积网络行为识别模型,实现对监控视频中行人异常行为的高效识别。采用自建数据集和公开数据集对该方法有效性进行检验,结果表明,该关键帧定位算法可高效实现异常行为定位,结合基于瓶颈残差模块时空图卷积网络,在减少时空图卷积网络计算复杂度的同时提升了网络性能,能够有效判断行人异常行为。

In order to improve the detection efficiency of pedestrian abnormal behavior in surveillance video,an abnormal behavior recognition method combining key frame positioning and spatio-temporal map convolution is proposed.Based on the detection of key points of human skeleton,this method uses key point motion characteristics to locate pedestrian anomalies in video behavior key sequence,using spatio-temporal map convolutional network to extract the advantages of pedestrian spatio-temporal features,construct the human skeleton spatio-temporal map on the key frame sequence,and establish a spatio-temporal map convolutional network behavior recognition model based on the bottleneck residual module to realize monitoring video efficient identification of abnormal behaviors of pedestrians in China.Selfbuilt data sets and public data sets are used to test the effectiveness of the method.The results show that the key frame positioning algorithm in this paper can efficiently locate abnormal behaviors,combined with the spatio-temporal graph convolution network based on the bottleneck residual module,while reducing the computational complexity of the spatio-temporal graph convolutional network,it improves the network performance and can effectively judge the abnormal behavior of pedestrians.