Implementation of Occluded Pedestrian Detection Method Based on Improved YOLOv5 in ROS Platform

In this paper, we propose an improved YOLOv5 pedestrian detection algorithm to solve the problems of target missing and low accuracy in the ROS platform. By adding a small detection layer of 160*160, the method improves the detection performance of the model and effectively reduces the false detection rate of occluded pedestrians, especially heavily occluded targets. In order to further improve the detection accuracy, it fuses the underlying features of the backbone network to achieve a path aggregation network with multi-feature fusion. Furthermore, the Soft-DIoU-NMS algorithm is used for post-detection processing to reduce the miss rate. Experiments on a public dataset show that mAP of the modified model, compared to the original network, is improved by 1.5%, and the false detection rate is reduced. The method is applied to the ROS platform, and the implementation results show that the proposed method is efficient for detecting occluded pedestrians, in addition, it is able to achieve real-time detection with an average detection speed of 13.6ms.