提出了一种非刚体图象序列特征点对应的新方法,首先根据图象序列帧间的时间间隔很小这一特点,指出特征点在图象帧间的运动具有平滑性,进而提出了一种特征点运动平滑的代价函数。然后将目标跟踪技术应用于特征点的跟踪,利用两个一维卡尔曼滤波器分别对特征点的两个坐标进行预测跟踪。在图象帧间的特征点对应过程中,该文综合考虑了特征点的预测位置与实际位置之间的距离以及特征点的平滑代价函数等因素,从而使特征点的运动轨迹得到正确的延续。特征点的遮挡问题可以通过特征点跟踪过程中对特征点位置的预测得到解决。实验证明,该文方法

In this paper, a new approach for feature points correspondence in non-rigid image sequences is proposed. First of all, according to the very short time interval between consecutive frames in image sequences, the paper points out that the motion of feature points is smooth, and a cost function that describes the smooth degree of motion feature points is presented. Then the object tracking technique is introduced into the tracking of feature points, using two 1D Kalman filter to track the two coordinates of feature points. In the process of feature points correspondence between the two consecutive frames, the distance between the predicted position of feature points and the position of feature points in the next image frame, and the cost function of feature points are used to correctly judge the extension of every motion trajectory. The occlusion problem can be solved by the position prediction in the process of the tracking of feature points. Results show the approach can establish correct correspondence of feature points and can solve the problem of occlusion.