The Improved Pedestrian Target Detection and Re-recognition Algorithm

Image processing technology can be applied to detect and recognize the specific pedestrian target. To well detect moving pedestrian targets in complex occlusion scenes, the improved pedestrian detection algorithm based on YOLOv5 and re-recognition algorithm based on foreground mask estimation are proposed in this paper. In the improved pedestrian detection algorithm, aiming at interclass occlusion, a new two-branch fusion detection head is designed, the loss function and sample matching strategy are modified, meanwhile the non-maximum suppression algorithm is optimized. Aiming at in-class occlusion, SoftNMS algorithm is introduced to improve the detection capability in occlusion scene. In the modified pedestrian re-recognition algorithm, to enhance the pedestrian re-recognition rate under intricate backgrounds, HRNetV2 network replaces ResNet50 as the backbone network and three loss functions are optimized. Experimental results show that modified algorithms improves the performance of pedestrian target detection and re-recognition in complex occlusion scenes.