Pedestrian cross-camera tracking technology aims to extract the same pedestrian target from multiple disjoint surveil-lance videos in complex cross-camera scenarios,and has important practical application value.In the existing research on pedestrian cross-mirror tracking technology,the local characteristics of pedestrian targets are easily interfered by factors such as external lighting conditions,background environment,person angle,human body posture,and occlusion degree.To address this problem,this paper proposes a pedestrian cross-mirror tracking method based on improved multi-feature fusion.Based on the Yolov8 target detection algo-rithm,the SGHM human body area matting algorithm is used to extract the human body area and eliminate useless background infor-mation;and add The adaptive multi-feature fusion network design enhances the tightness of the fusion of key pedestrian features and improves the accuracy of target matching.Experiments have shown that this method has significantly improved its evaluation indicators mAP and Rank-n on the data set Market1501.
关键词
深度学习/行人重识别/多特征融合/跨镜智能追踪
Key words
deep learning/pedestrian re-recognition/intelligent tracking across mirrors
维普
万方数据