A violence detection method based on deep and shallow feature fusion  

作　　者：Lin'en Liu Xuguang Zhang 

机构地区：[1]HangZhou Dianzi University,The Communication Engineering Department,Hangzhou 310018,China

出　　处：《Instrumentation》2024年第4期64-75,共12页仪器仪表学报(英文版)

摘　　要：In the research of video-based violent behavior detection,the motion information in the video is vital for violence detection.How to highlight motion information in videos and integrate spatiotemporal information is an urgent problem that needs to be solved in violence detection.In this paper,we propose a deep learning architecture that integrates shallow features into deep features to strengthen the network's ability to express motion information at a deep level.To enhance the weight of motion information in the network,we design a downsampling module to extract shallow features,fused with the deep features extracted by MobileNet's Blocks.Furthermore we constructed a channel attention module and introduced a Convolutional Long Short-Term Memory(ConvLSTM)module.These two modules aim to redistribute network attention:the channel attention module focuses on channel-level information and the ConvLSTM module emphasizes temporal aspects.Finally,we employ 3D convolution and global pooling to compress the feature sizes,fed into fully connected layers to perform violence detection.Experiments are conducted on three publicly available standard datasets,achieving an accuracy rate of 91%on the surveillance video dataset RWF2000,97.5%on the Hockey fight dataset,and 100%on the movies dataset.Overall,the proposed model demonstrates satisfactory performance in violence detection.

关 键 词：SURVEILLANCE optical flow violence detection deep learning 

分 类 号：G63[文化科学—教育学]