吉林大学学报(理学版) ›› 2021, Vol. 59 ›› Issue (3): 609-618.

• • 上一篇    下一篇

基于Mask R-CNN的多目标跟踪算法

张彩丽1, 刘广文1, 詹旭1, 史浩东2, 才华1,3, 李英超2   

  1. 1. 长春理工大学 电子信息工程学院, 长春 130022; 2. 长春理工大学 光电工程学院, 长春 130022; 3. 长春中国光学科学技术馆, 长春 130117
  • 收稿日期:2020-07-07 出版日期:2021-05-26 发布日期:2021-05-23
  • 通讯作者: 刘广文 E-mail:lgwen_2003@126.com

Multiple Object Tracking Algorithm Based on Mask R-CNN

ZHANG Caili1, LIU Guangwen1, ZHAN Xu1, SHI Haodong2, CAI Hua1,3, LI Yingchao2   

  1. 1. School of Electronic Information Engineer, Changchun University of Science and Technology, Changchun 130022, China;
    2. School of Opto-Electronic Engineer, Changchun University of Science and Technology, Changchun 130022, China;
    3. Changchun China Optical Science and Technology Museum, Changchun 130117, China
  • Received:2020-07-07 Online:2021-05-26 Published:2021-05-23

摘要: 针对目前多目标跟踪算法在面对目标频繁遮挡时跟踪效果较差的问题, 提出采用Mask R-CNN作为检测器, 根据检测结果利用Kalman滤波器预测下帧图像中跟踪目标的位置, 用改进匈牙利算法进行数据关联, 并利用轨迹修正方案应对轨迹中断问题. 将该算法在MOT16数据集的各测试集上进行实验, 实验结果表明, 该算法目标跟踪准确率为55.1%, 且针对目标被遮挡问题效果较好.

关键词: 机器视觉, 目标跟踪, 匈牙利算法, 深度学习

Abstract: Aiming at the problem that the current multiple object tracking algorithm had poor tracking effect in the face of frequent target occlusion, using Mask R-CNN as a detector, according to the detection results, the Kalman filter was used to predict the position of the tracking target in the next frame image and the improved Hungarian algorithm was used for data association, and the trajectory correction scheme was used to deal with the problem of trajectory interruption. The algorithm was experimented on each test set of MOT16 dataset, the experimental results show that the tracking accuracy of the algorithm is 55.1%, and the effect of target occlusion is better.

Key words: machine vision, target tracking, Hungarian algorithm, deep learning

中图分类号: 

  • TP391