Abstract:
A novel method for abnormal event detection is proposed based on multi-scale feature prediction. Firstly, dilated convolution network is used to extract the features of different size receptive fields and fuse them so that address the objects of different scale in video frame. Secondly, a lightweight channel-wise attention module is applied to reduce the impact of background information. Finally, in order to make full use of the context information between video frames, a deep feature prediction module is applied to predict the features of the current moment based on the features of the historical moment, and the prediction error is used for abnormality judgment. Experiments were performed on the two benchmark data sets of USCD Ped2 and UMN to test and evaluate the proposed method. The experiments results show that the proposed method is more effective than other state-of-the-art methods.