A video abnormal behavior detection improvement method was proposed to address issues such as significant variations in target object spatial dimensions and excessive generalization capability for abnormal behavior prediction in traditional video abnormal behavior detection tasks.Different scales of feature information were extracted on the high-level feature map by a multi-scale feature module composed of multiple branches with dilated convolutions,and a cascaded memory enhancement module was used to store normal behavior features to weaken the generalization capability.The coordinated operation of the multi-scale feature module and the memory enhancement module was used to effectively collect and memorize the multi-scale feature informa-tion in normal behavior scenes.The effectiveness of this method was verified through experimental analysis.
关键词
异常行为检测/多尺度特征/多分枝结构/空洞卷积/泛化能力/记忆增强/协同工作
Key words
video abnormal behavior detection/multi-scale features/multi-branch structure/dilated convolution/generalization ability/memory enhancement/collaborative work
维普
万方数据