A Multi-Scale Vehicle Detection Method for Highways Based on YOLOv5

Aiming at the problem of low accuracy of vehicle detection model due to complex vehicle driving environment, target vehicle scale changes and occlusion in highway video data, a CCA-YOLOv5 model framework based on YOLOv5 is proposed. The CCA attention module is designed in the framework to enhance the model’s attention to key areas, and it is introduced into the YOLOv5 model to enhance the detection ability to adapt to different scales. It is verified on the public dataset BIT-Vehicle and the self-made highway vehicle dataset. The experimental results show that the detection rate and missed detection rate of the vehicle detection model CCA-YOLOv5 in the self-made highway vehicle dataset are 89% and 18%. Respectively, compared with the YOLOv5 model, the detection the rate increases by 21%, the missed detection the rate reduces by 14%, and the detection time of each picture reaches 3.8 milliseconds. It still maintains strong detection performance under the BIT-Vehicle dataset, so the model has high application value.