F. Zhang, A.J. Wang

Pages: 149-160

Abstract
This paper proposes a driver dangerous driving behavior monitoring method based on deep learning. Firstly, construct a driving behavior dataset by collecting images and preprocess the driving behavior data; Secondly, the improved filtering redundant box module algorithm in deep learning is used to process the original information features of the image, and ResNet-50 with residual structure in deep learning is introduced to extract feature information; Finally, the CRITIC weighting method in the objective weighting method is used to calculate the weights of various dangerous driving behaviors, and machine vision is combined to achieve driver dangerous driving behavior monitoring. The experimental results show that the accuracy rate of dangerous driving behavior monitoring in this method is 99.2%, and the monitoring time for driving behavior is 1.2 seconds.
Keywords: deep learning; dangerous driving behavior; filter redundant box; spectral eigenvalues; critic weighting method