In view of the occlusion between pedestrians in blind field of vision such as traffic intersections,it is of practical significance to detect target pedestrians in complex roads efficiently and accurately.In order to achieve pedestrian detection at nighttime intersection scenes,a pedestrian target detection algorithm based on improved YOLOv5 is proposed,using Non local and PSA modules to improve the Bottleneck CSP of the original YOLOv5 network,which can effectively make up for the inter frame information exchange process of pedestrian features and enhance the feature dependence of long range channels.A deeper 160×160 detection layer and adaptive anthor are designed to improve the boundary regression accuracy of night pedestrian detection.The experimental results show that for the traffic intersection scene under nighttime,the compressed and improved model has high robustness for pedestrian detection,with 14.2% and 12.7% improvement compared to the original algorithm mAP_0.5 and mAP_0.5:0.95 values,respectively,indicating the effectiveness of the proposed algorithm for nighttime pedestrian detection.