An Traffic Flow Detection System Based on Improved YOLOv5 in Complex Weather Conditions

For the traditional way to carry out the traffic flow recognition of the computing platform, it is difficult to achieve the demands of real-time detection in too many models. Therefore, an improved traffic flow recognition method based on YOLOv5 is proposed to address the lightweight model's problem of low detection accuracy. Firstly, we use the dark channel defogging algorithm to preprocess the deep learning samples. Secondly, the Slim-Neck module of GSConv is introduced to replace the Neck part in the YOLOv5 network, which is a lightweight convolution technique with a good balance of accuracy and detection speed. Finally, the CIoU Loss is used as the regression loss function of the target bounding box to speed up the convergence speed and improve the detection effect. The effectiveness of the improved YOLOv5 algorithm has a significant improvement in other detector comparison experiments. The results showed that the improved YOLOv5s are 2.2% more accurate and 14.2% faster than the original YOLOv5 algorithm. The improved yolov5 model can be better applied in practice.