Research on Abnormal Driving Behavior Detection Algorithm Based on YOLOv8

To address the issues of low detection accuracy and slow speed in detecting abnormal driving behavior during daily driving, a YOLOv8-ADBD abnormal driving behavior detection algorithm is proposed, which is based on YOLOv8m as the baseline model. By introducing a lightweight visual model based on the ViT architecture, the backbone module of YOLOv8m is reconstructed. To build Slim-Neck, the one-time aggregation method is employed in constructing the cross-level partial network module. This approach minimizes computational complexity and enhances the efficiency of the algorithm. In order to tackle the problem of missed detection of small objects in abnormal driving behavior detection, a new CNN module STD-nsConv is inserted. Additionally, a new P2 layer for small target detection is introduced through upsampling, feature concatenation, etc., to enhance the recognition performance of small objects. The experiments demonstrate that the YOLOv8-ADBD algorithm achieves a 4.5% improvement in mAP50, there has been a reduction of 91.9% in the number of parameters and 85.4% in GFLOPs compared to the original YOLOv8m algorithm. Enhancing both the real-time performance and accuracy of abnormal driving behavior detection is achievable through YOLOv8-ADBD.