中文摘要：

车辆跟踪中普遍存在的车辆遮挡问题直接影响着跟踪的精度，是车辆跟踪研究中的关键问题。在介绍车辆跟踪算法基本原理的基础上，提出了一种基于反向ST—MRF模型的车辆遮挡分割算法。该算法通过反向沿时间轴运用ST．MRF累积图像，优化运动矢量和融合不完整的分割部分，对车辆遮挡进行了比较完美的分割。最后通对比原始ST．MRF算法和反向ST-MRF算法，2者得到目标跟踪的成功率分别为87％和98％。基于反向ST—MRF模型的车辆遮挡分割算法能在交通量比较大，且车辆出现相互遮挡的情况下较准确地获得车辆跟踪数据，为以后的交通事件检测提供重要的数据基础。

英文摘要：

As the common problem in vehicle tracking, vehicle occlusion directly impacts on the tracking precision, which is the crux of vehicle tracking study. Based on introducing the basic principle of vehicle tracking algorithm, a segmentation algorithm of vehicle occlusion based on reversed spatio-temporal Markov model is proposed. Through applying ST-MRF backward along temporal axis, by optimizing motion vectors and merging fragmental segments, the perfect segmentation to vehicle occlusion is obtained. At last, compared the primary ST-MRF algorithm with the revered ST-MRF algorithm, the success rates of target tracking are 87% and 98% respectively. The presented algorithm can obtain accurate tracking data under the condition of large traffic volume and mutual vehicle occlusion, and provide important data basis for traffic incident detection in the future.