Y.X. Chu, J. Zhang, L. Yang

Pages: 133-144

Abstract
It is of great significance to effectively identify the dangerous driving behaviors of drivers in real time. In order to improve the accuracy of driving behavior classification and the efficiency of risk identification, this paper proposes a risk identification method of road traffic dangerous driving behavior based on sliding window feature fusion. Firstly, the influencing factors of driving danger scenes are obtained, and the driving behavior characteristics are extracted according to the sliding window method; Then, the feature multiplication is used to calculate the time feature, and the spatial feature under each channel is calculated. The feature fusion method is used to achieve the spatio-temporal feature fusion; Finally, the risk identification function is constructed according to the ConvLSTM cascade method, and the risk identification result is obtained. The experimental results show that the classification accuracy of this method can reach 96.93%, the recognition accuracy is 93%, and the time consumption is less than 30s. The method of road traffic dangerous driving behavior risk identification based on sliding window feature fusion has better accuracy of driving behavior classification and better risk identification efficiency.
Keywords: sliding window; feature fusion; dangerous driving behavior; convlstm cascade; characteristic multiplication