摘要： 多相机间运动目标的跟踪与识别需要获得尽可能准确的目标区域。针对人群目标的粘连问题，提出一种基于姿态模型的人群目标分割方法。依据人体在运动过程中姿态的变化规律，构造7种出现频率较高的姿态模型。依次对单个目标和联合目标进行模型匹配，获得各个目标的位置、大小以及运动姿态信息。实验结果表明，该方法能有效解决相互遮挡情况下的目标分割问题。

关键词: 姿态模型, 联合相似度, 目标分割, 遮挡目标

Abstract: Object tracking and recognition across multiple cameras requires as far as precise object area. Aiming at the adhesion problem of crowd object, this paper proposes a crowd object segmentation method based on posture model. According to the posture change rule during human movement, it constructs seven kinds of posture model with high probability. Each object position, size and posture message are obtained by model matching for single and combined objects. Experimental results show that this method can solve object segmentation effectively under the situation of occlusion.

Key words: posture model, combined similarity, object segmentation, occlusion object

中图分类号: 

• TP394.1