作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2010, Vol. 36 ›› Issue (11): 185-187. doi: 10.3969/j.issn.1000-3428.2010.11.067

• 人工智能及识别技术 • 上一篇    下一篇

基于运动补偿和MCMC算法的视频目标跟踪

刘鹏威,王汇源,乔 伟,吴晓娟   

  1. (山东大学信息科学与工程学院,济南 250100)
  • 出版日期:2010-06-05 发布日期:2010-06-05
  • 作者简介:刘鹏威(1986-),男,硕士研究生,主研方向:模式识别,计算机视觉;王汇源,教授;乔 伟,硕士研究生;吴晓娟,教授、博士生导师
  • 基金资助:
    国家自然科学基金资助项目(60675024)

Video Target Tracking Based on Motion Compensation and MCMC Algorithm

LIU Peng-wei, WANG Hui-yuan, QIAO Wei, WU Xiao-juan   

  1. (School of Information Science and Engineering, Shandong University, Jinan 250100)
  • Online:2010-06-05 Published:2010-06-05

摘要: 针对视频目标跟踪领域摄像头运动等问题,提出一种基于二次观测模型的马尔科夫链蒙特卡洛(MCMC)粒子滤波算法。第1次观测通过计算相邻2帧的光流场对运动模型实时修正使其逼近真实的运动方程,第2次观测MCMC粒子滤波步骤。二次观测模型利用图像中的光流信息进行运动补偿实现跟踪。时变的运动模型可以有效提高MCMC方法的效率,减少无效的粒子点数,使其能更快速地收敛到真实值。实验表明对MCMC进行运动补偿可以有效处理摄像头运动问题。

关键词: 马尔科夫链蒙特卡洛算法, 粒子滤波, 二次观测模型, 光流场, 运动补偿

Abstract: Aiming at the problems caused by the camera shake in video target tracking, this paper proposes a new Markov Chain Monte Carlo (MCMC) particle filtering algorithm based on a twice-observation model. The first observation is to modify the motion model through optical flow calculation between two consecutive frames in the video to approach the true motion equation of the target, while the second observation is MCMC particle filtering procedure. This twice-observation model makes full use of the optical flow information to perform motion compensation. Time variant motion model can increase the efficiency of MCMC algorithm, reduce the number of ineffective particles, and enable it to converge to the real value faster. Experiment shows that combining motion compensation and MCMC particle filtering can handle camera motion problems effectively.

Key words: Markov Chain Monte Carlo(MCMC) algorithm, particle filtering, twice-observation model, optical flow, motion compensation

中图分类号: