基于局部特征匹配的多目标跟踪算法

Part feature matching based multi-target tracking

为解决多目标跟踪过程中因遮挡导致跟踪失败的问题,提出一种基于局部特征匹配的跟踪算法。在卡尔曼滤波跟踪框架下,根据目标数据关联的结果判断目标的状态并进行针对性处理。当目标处于相互遮挡的状态时,利用目标的局部模板在当前帧进行匹配获取目标候选区域,利用改进的距离加权彩色直方图计算候选区域与局部模板的相似度,结合直方图的相似度和卡尔曼预测确定目标的位置。实验结果表明,在满足实时性的要求下,该方法能够有效地处理目标的遮挡问题。

To solve the occlusion problem in multi-target tracking,apart feature matching based tracking algorithm was proposed.In the Kalman filter framework,the state of a targets was confirmed via the result of feature association.Each state was processed specially.When a target was in the state of mutual occlusion,the likely candidate regions were selected through part template matching.Then the similarity between the candidate region and the part template was calculated using the improved distance weighted color histogram.At last,combining the similarity and the Kalman filter prediction,the target’s location was determined.The experimental results show that the proposed algorithm can handle the occlusion problem effectively and meet the real-time requirement.