Identification method for abnormal driving behavior of vehicles based on lightweight graph convolution

Identifying abnormal driving behaviors is crucial for enhancing road safety, preventing accidents, and protecting the well-being of all road users. To address the issues of high misidentification and miss rates, as well as prolonged completion times in traditional methods, this study proposes an identification method for abnormal driving behavior based on lightweight graph convolution. The process begins by capturing vehicle images using a camera. To enhance the image quality, a contrast-limited adaptive histogram equalization method is employed to improve the contrast of the vehicle images. These enhanced images are then combined with a Gaussian mixture model for object detection within the vehicle images. Subsequently, a lightweight graph convolution network is constructed, incorporating spatiotemporal feature extraction modules, Ghost modules, and spatiotemporal attention modules. The object detection results are fed into this network to obtain the final identification outcomes for abnormal driving behaviors. The experimental results show that the average misidentification rate of this method for abnormal driving behavior is reduced to 2.92%, and the average missed detection rate is reduced to 4.19%. In addition, the completion time for identifying these behaviors is only between 0.16 seconds and 0.51 seconds.