Research on Crime Behavior Recognition Based on Multi-Angle Videos
Lijian Ji, Qing Ye, Peng Lu, Zhiguo Wang · 2024
Crime behavior recognition is a crucial issue in the field of public safety. Traditional single-view video analysis often faces limitations when dealing with complex criminal behaviors. This paper proposes a crime behavior recognition method based on multi-view video analysis. We employed Convolutional Neural Networks (CNNs) to process video data, and the experimental results demonstrate that 3D CNNs offer more accurate recognition compared to 2D CNNs. In the R2Plus1D model (where the 3D convolutional filters are decomposed into separate spatial and temporal components), the accuracy of crime behavior recognition based on multiple angles reached 88%.