摘要
基于TBD(track by detection)框架,使用YOLO网络训练并优化得到了较好的行人检测器,提出一种匹配网络进行多目标的匹配,得到一个准确率较高的行人多目标跟踪框架。为解决行人多目标跟踪中误匹配、目标丢失等问题,提出对于跟踪轨迹模板更新的策略以及对于计算的优化。在MOT数据集上的实验证明,该算法在行人多目标跟踪中取得了较高的准确率,其他多项指标也都达到了较高的水平。
Based on track by detection(TBD)framework,this paper uses YOLO network training and optimization to obtain a better pedestrian detector,and proposes a matching network for multi-target matching,thus obtaining a pedestrian multi-target tracking framework with higher accuracy.In order to solve the problems of mismatch and target loss in pedestrian multi-target tracking,the strategy of updating the tracking trajectory template and the optimization of calculation were proposed.The experiments on MOT data sets show that the proposed algorithm achieves high accuracy in pedestrian multi-target tracking,and other indicators have reached a high level.
作者
樊璐
张轶
Fan Lu;Zhang Yi(School of Computer,Sichuan University,Chengdu 610064,Sichuan,China)
出处
《计算机应用与软件》
北大核心
2021年第4期190-196,214,共8页
Computer Applications and Software