Abstract: There usually exist large accumulated errors when dynamic deformable three-dimensional（3-D） surfaces are measured from a long calibrated stereo image sequence by global optimization methods,which results in unstable measurement.Thus,this paper presents an effective method based on local image patch expansion technique.Firstly,the proposed method locates the stable matching feature points in stereo and motion images,and sorts the matching regions according to normalized cross-correlation of the matching feature points;Secondly,the method performs expansion from the most reliable region,and fixes the holes produced by self-occlusions and high light;Finally,the method removes the accumulated errors generated during motion and realizes 3-D measuring and tracking accurately.The performance is evaluated on the synthetic rotating textured sphere,and the effectiveness is demonstrated by different real surfaces,which shows that the proposed method can measure complex deformable surfaces and provide useful data for analyzing the motion of dynamic deformable objects.