Road vehicle object detection based on improved YOLOv8

This study is dedicated to road vehicle object detection in intelligent transportation systems. By improving the YOLOv8 algorithm, dynamic snake convolution, EMA attention mechanism, and AlphaGIoU loss function are introduced to address the challenges faced by vehicle detection in complex road environments. In the study, data collected from multiple channels and the COCO public dataset were utilized to cover road target images under different lighting and climate conditions, and a diverse training sample set was constructed. By implementing strict preprocessing measures, the quality of the training data is ensured. The model training and validation adopt a comprehensive loss function to enhance the model's generalization ability and detection accuracy. For the YOLOv8 algorithm, we optimized the network structure and training strategy, particularly by enhancing feature extraction ability through dynamic snake convolution, improving key feature attention through EMA attention mechanism, and optimizing the balance between regression and classification loss through AlphaGIoU loss function. In the experimental verification phase, tests were conducted on image datasets containing multiple targets such as cars, buses, and pedestrians. After model training and testing, the improved YOLOv8 algorithm achieved a mAP value of 0.628 on the test set, verifying the system's vehicle detection ability in complex environments and providing effective technical support for the optimization of intelligent transportation systems.