基于SIFT特征的抗遮挡目标跟踪算法

针对多目标跟踪过程中目标易丢失的问题，提出一种基于尺度不变特征变换(SIFT)特征的多目标跟踪算法。利用SIFT特征集，通过设置目标特征留存优先级，实时更新特征集，保存目标近几帧的稳定特征。对于半遮挡导致的物体丢失现象，提出一种根据匹配特征位置关系进行目标分离的方法，可有效标定遮挡发生时的各个目标。该算法无需目标的先验信息，通过留存优先级即可较稳定地跟踪多个目标。实验结果证明其对目标遮挡、尺度变化及形变具有较好的容错性和跟踪鲁棒性。

Anti-occlusion Object Tracking Algorithm Based on SIFT Feature

In order to solve the problem of object losing in the process of multi-object tracking, a new method based on Scale invariant Feature Transform（SIFT） features is proposed. The object image is transformed to SIFT features. By setting the feature reserving priority of preference, the features are updated in real time to store the stable features of recent frame. For the problems induced by occlusion, the method can successfully separate and mark different objects according to coordinate relation of matching feature. The algorithm can realize the stable tracking of multi-objects by feature reserving priority of preference instead of prior information. Experimental results show that the method has strong robust and error-tolerance to objects deformation, scale change and occlusion.