Robust Visual Tracking for the Changed Object Appearance

摘要：

In this paper, we propose a visual tracking method for the changed appearance of the tracked object due to varying illumination, pose variants and scale changes. It is based on the improved particle filter. The observation model employs on-line updating appearance model, affine transformation and M-estimation to handle the changed appearance. To improve the performance of re-sampling process of particle filter, we use Kalman filter to integrate the most recent observation into particle filter, and get sub-optimal Gaussian proposal distribution. The smaller computational cost is required to estimate the posterior probability density properly. Experimental results demonstrate the robust performance of the proposed algorithm by tracking in the recorded video sequences.

关键词： Visual tracking Particle filter Observation model, Kalman filter

作者: Qicong Wang Chenhui Yang Yahui Gao Eryong Wu

作者单位: Department of Computer Science Xiamen University Xiamen, Fujian Province, China College of Computer Hangzhou dianzi University Hangzhou, Zhejiang Province, China

会议类型: 国际会议

会议名称: 2008 Sino-European Workshop on Intelligent Robots and Systems(SEIROS08)(第一届中欧智能系统及机器人国际学术研讨会)

会议地点: 重庆

会议语种:英文

页码: 1-6

在线出版日期: 2008-12-11（万方平台首次上网日期，不代表论文的发表时间）