Abstract: It is very important to track multi-object in video processing all along. The calculation of observation likelihood is a difficult problem under the circumstance of occlusion by the targets. Method presented by this article adds a parameter of occlusion state for the targets, and calculate the observation likelihood based on the prediction of the occlusion. Apply this method on the video of soccer match in which occlusion occurs frequently, results show that this method can avoid coalescence and disorder of the target labels when occlusion and crowding are not severe.

Key words: Observation Likelihood, Soccer Video, Occlusion, Multi-Object Tracking

摘要： 多目标追踪一直是视频处理中的重要问题，目标互相遮挡的情况下观测置信度的计算是多目标追踪中的难点。本文提出的方法将被遮挡的情况作为目标的一个状态分量，在对遮挡估计的基础上计算目标的观测置信度。将方法应用于遮挡频繁发生的足球比赛视频，结果表明本文的方法在目标遮挡或聚集并不严重时，可以避免目标标号粘合和错乱的问题。

关键词: 观测置信度, 足球视频, 遮挡, 多目标跟踪