Research on Vulnerable Road User Detection Algorithm based on Improved Deep Learning

This paper proposes a detection algorithm based on deep learning for Vulnerable Road Users such as pedestrians and cyclists, which is improved on the basis of YOLOv5 network model. (1) Aiming at the problems of low resolution and insufficient information for small targets, a multi-scale feature fusion method is adopted to integrate shallow features with deep features. In this way, the effective information of small target is enriched, and the accuracy of target detection is improved. (2) In view of the interference of image noise, background and other factors, the channel attention method is introduced to strengthen key features and suppress the interference of noise, which can improve the model's attention to small targets and enable the network to better identify blocked targets; (3) Aiming at the problem that the computing speed of the model is difficult to achieve real-time performance, a deep separable convolution optimization method is proposed to reduce the amount of computation, so as to reduce the computing time of the model. The experimental results show that compared with the YOLOv5 model in KITTI data set, the precision of the improved model is increased by 1.26%, the mAP is increased by 0.76 percentage points, and the detection speed is increased by 2f/s.