电子学报 ›› 2015, Vol. 43 ›› Issue (3): 591-596.DOI: 10.3969/j.issn.0372-2112.2015.03.026

• 科研通信 • 上一篇    下一篇

基于视频多目标跟踪的高度测量算法

姜明新1,2, 王培昌2, 王洪玉1   

  1. 1. 大连理工大学信息与通信工程学院, 辽宁大连 116024;
    2. 大连民族学院信息与通信工程学院, 辽宁大连 116600
  • 收稿日期:2013-07-15 修回日期:2013-10-09 出版日期:2015-03-25
    • 作者简介:
    • 姜明新 女,1979年出生于黑龙江省.博士,大连民族学院信息与通信工程学院讲师.研究方向为计算机视觉、智能视频监控. E-mail:jmx@dlnu.edu.cn;王培昌 男,1955年出生于辽宁省大连市.博士,大连民族学院信息与通信工程学院教授.研究方向为计算机视觉、计算机控制与检测技术. E-mail:wpc@dlnu.edu.cn;王洪玉 男,1968年出生.大连理工大学信息与通信工程学院教授,博士生导师.研究方向无线传感器网络、计算机视觉.
    • 基金资助:
    • 国家自然科学基金项目 (No.61172058,No.61403060); 中央高校自主基金 (No.DC110321); 辽宁省教育厅资助项目 (No.L2012476); 中国博士后科学基金资助项目 (No.2014M551081); 大连民族学院人才引进项目 (No.20136212)

Height Estimation Algorithm Based on Visual Multi-Object Tracking

JIANG Ming-xin1,2, WANG Pei-chang2, WANG Hong-yu1   

  1. 1. School of Information and Communication Engineering, Dalian University of Technology, Dalian, Liaoning 116024, China;
    2. School of Information and Communication Engineering, Dalian Nationalities University, Dalian, Liaoning 116600, China
  • Received:2013-07-15 Revised:2013-10-09 Online:2015-03-25 Published:2015-03-25
    • Supported by:
    • National Natural Science Foundation of China (No.61172058, No.61403060); Initiative Fund for the Central Universities (No.DC110321); Supported by Education Department of Liaoning Province (No.L2012476); Supported by China Postdoctoral Science Foundation (No.2014M551081); Talent Introduction Project of Dalian Minzu University (No.20136212)

摘要:

本文提出了一种基于视频多目标跟踪的高度测量算法.首先,采用码本模型检测前景,利用图割理论实现对多目标的跟踪.然后,提取每一帧中目标的头部特征点和脚部特征点,根据投影几何的约束关系计算每个目标的高度.最后,融合多帧的测量结果进行数据优化.本文提出的算法不需要对相机进行完全标定,只需计算摄像机的灭点和地平面的灭线,降低了计算的复杂度.实验结果表明,本文提出的算法具有较高的测量精度,对遮挡和运动状态变化具有较强的鲁棒性,同时,能够满足实时性要求.

关键词: 高度测量, 视频多目标跟踪, 投影几何约束, 图割

Abstract:

In this paper, we propose a height estimation algorithm based on visual multi-object tracking.Firstly, the foreground is detected by using codebook modeling algorithm, and multi-object tracking is performed by using graph cuts.Then, the head feature points and the feet feature points of objects are extracted in each frame, and the height of every object is computed according to projective geometry constraint.Finally, data optimization is performed by fusing multi-frame measurement results.This method does not require full camera calibration, but require computing the vanishing point of camera and the vanishing line of ground.Using this method, the computational complexity can be reduced.The experimental results demonstrate that our method has higher measurement accuracy, and is robust to occlusion and the changing of motion state.Meanwhile, it can also achieve real-time performance.

Key words: height estimation, visual multi-object tracking, projective geometry constraint, graph cuts

中图分类号: