MCMC-based multi-object visual tracking was investigated here. To improve the confidence of sampling and perform the iteration effectively,a new approach to multi-object visual tracking was proposed based on reversible jump Markov chain Monte Carlo (RJMCMC) sampling. Given image observation, the tracking problem was formulated as computing the MAP (maximum a posteriori) estimation . The prior proposal distribution of object was developed with the aid of association match matrix, and four types of reversible and jump moves were designed for Markov chains dynamics. The likelihood distribution measure was presented via position-weighted colour hist match between reference objects and candidate objects. The state updating was generated from mean-shift (MS) iteration, rather than from random walk in the MCMC sampling. Experimental results and quantitative evaluation demonstrate that the proposed approach is effective for challenge situations.%研究了基于MCMC的多目标跟踪算法.针对MCMC迭代过程中抽样置信度低以及不能进行有效迭代的问题,提出一种新的基于RJMCMC的视觉多目标跟踪算法.给定观测量,将跟踪问题建模为状态量的最大后验估计(MAP)、关于MAP的先验与似然的估计.借助匹配阵给出了目标先验建议分布,设计了4种马氏链可逆运动方式;似然度量采用随空间加权的颜色直方图匹配衡量.MCMC抽样过程中的状态由MS迭代产生,而不是随机走生成.基于似然度量导出了MS迭代式.实验结果及定量分析评估结果说明了本算法的有效性.
展开▼
