Detection of Dangerous Driving Behavior Based on CG-YOLOv5

In order to avoid huge traffic accidents caused by dangerous driving behavior on the way, it is very important to detect this behavior in real time. In view of the high loss of the existing methods and the interference of the data set background, an improved YOLOv5 detection method is proposed. Firstly, the convolution block attention module mechanism is integrated in the network to enhance the feature expression ability and suppress the background interference in the process of feature fusion, so as to reduce the loss of the algorithm. Secondly, Ghost convolution is used instead of ordinary convolution operation to realize the lightweight of the detection model. The experimental results show that the recognition accuracy of this method is 99.9%, and the reasoning time under the same test data is as low as 3.895s. At the same time, the model is much smaller than the original network model, and can be effectively applied to the real-time accurate monitoring of dangerous driving behavior of drivers in the real environment.