基于时空马尔可夫随机场模型的车辆跟踪算法研究

鲁棒的车辆跟踪是实现交通事件自动检测的重要前提,车辆跟踪中的车辆相互遮挡则是影响车辆跟踪结果的关键因素。针对这一难题,设计自适应的车辆跟踪算法,并依据交通图像序列的时空相关性,根据马尔可夫的基本理论和贝叶斯方法,应用MRF-MAP理论分析框架,并结合了彩色图像序列的纹理信息建立了图像序列的时空马尔可夫随机场模型。采用随机松弛算法中的Metropolis算法来求解时空马尔可夫随机场模型,对车辆跟踪得到的目标标号图进行优化,从而解决车辆跟踪中的遮挡问题。初步实验结果,跟踪不遮挡的车辆时达到的跟踪成功率为95%。遮挡情况时成功率也可达到83%。实验结果表明,该跟踪算法在不遮挡时效果非常理想,在遮挡情况下跟踪鲁棒性也较好。

A Vehicle tracking algorithm based on spatial-temporal Markov random field model

Robust vehicle tracking algorithm is an important precondition for realizing traffic event detection,but occlusion is a key influence factor for vehicle tracking.An adaptive vehicle tracking algorithm is designed to deal with the problem.In addition,according to the spatial and temporal characteristics of traffic image sequences,using basic Markov theory and Bayesian method,and combining texture information of color image sequences,a spatial-temporal Markov random field model of image sequence is built.After object maps are optimized with the Metropolis stochastic algorithm,the occlusion problem in vehicle tracking is solved.Preliminary experimental result,the tracking success ratio when vehicles are not occluded with each other is 95%,and when being occluded the tracking success ratio is 83%.The experimental results indicate that this tracking algorithm is accurate when being not occluded and acceptable when being occluded.