Research on free-form surface stereo matching method based on improved Census transform

The feature points obtained from free-form surface 3D reconstruction method which based on grid projection all distribute along the projected grid lines only, and there is none feature point on the surface among the projected grid lines, which lead to the low accuracy reconstruction. An improved Census transform stereo matching method based on adaptive window is presented to reconstruct the surface among the projected grid lines. The method incorporates feature-based method and region-based method. The grid lines projected on the experimental subject surface are utilized to guide Census transform, the Hamming distance computed by dynamic disparity is used as the matching cost, and the higher accuracy matching points are obtained after filtering. The proposed method is applied on the binary vision 3D reconstruction to validate the improved reconstruction accuracy of surface among the projected grid lines.