上海交通大学学报(自然版) ›› 2014, Vol. 48 ›› Issue (07): 1026-1032.

• 自动化技术、计算机技术 • 上一篇    下一篇

基于极线约束SIFT特征和粒子滤波的目标跟踪算法

刘士荣1,2,吴楚1,3,张波涛1,2,章海兵1,2
  

  1. (1.杭州电子科技大学 电气自动化研究所,杭州 310018;  2. 检测仪表与自动化系统集成技术教育部工程研究中心,杭州 310018; 3. 华为技术有限公司杭州研究所,杭州 310007)
     
  • 收稿日期:2013-06-28 出版日期:2014-07-28 发布日期:2014-07-28
  • 基金资助:
    国家自然科学基金资助项目(61175093)
     

A Tracking Algorithm Based on SIFT Feature and Particle Filter with Epipolar Constraint

LIU Shirong1,2,WU Chu1,3,ZHANG Botao1,2,ZHANG Haibing1,2
  

  1. (1. Institute of Electrical Engineering and Automation, Hangzhou Dianzi University, Hangzhou 301118, China; 2. Research Center of Detecting Instruments and Automation Systems Integration Technology ofthe Ministry of Education, Hangzhou 310018, China; 3. Hangzhou Institute, Huawei Technologies Co.
    Ltd., Hangzhou 301118, China)
  • Received:2013-06-28 Online:2014-07-28 Published:2014-07-28

摘要:

针对采用颜色或边缘等特征的目标跟踪算法所存在的跟踪效果不稳定的问题,提出了一种基于极线约束尺度不变特征变换(SIFT)和粒子滤波的目标跟踪方法. 该方法采用SIFT特征向量构建目标模型,引入极线约束改善目标匹配精度,采用粒子滤波算法获得SIFT特征向量的候选目标模型,利用似然函数计算目标模型与候选目标模型间的相似性. 实验结果表明,该方法可解决目标与背景颜色相似时的跟踪失败问题,且对目标外形与位姿发生变化具有较好的适应能力.
 
 

关键词: 目标跟踪, 极线约束, 尺度不变特征变换, 粒子滤波

Abstract:

To deal with the instability in the target tracking method based on color or edge features, a tracking algorithm based on SIFT features
 and particle filter with epipolar constraint was proposed in this paper. In this method, the target template was established using the vectors of the SIFT feature points, and the epiploar constraint was used to improve the target matching accuracy. A particle filter method was employed to establish the candidate template of SIFT feature vectors, and a likelihood function was used to calculate the similarity between the candidate template and the target template. Experimental results show that the proposed algorithm can track the target when the color of target is similar to one of the background, and it still works well when the posture or shape of the target changes.
 

Key words: target tracking, epipolar constraint, scaleinvariant feature transform (SIFT), particle filter

中图分类号: