Hole-filling algorithm in multi-view stereo reconstruction

To solve the surface reconstruction problem in weak texture or textureless and specular area in multi-view stereo (MVS), a novel hole-filling method is presented for MVS surface reconstruction based on visual hull and depth map fusion. The visual hull and 3D point cloud are taken as input data for fusion. Firstly we extract the voxels that are satisfied the sparsity constraint of 3D point cloud from visual hull, and then in order to get rid of the outer leaf nodes, the constraint of visual hull surface normal and rays is employed. Finally, mean curvature of the surface model is used to eliminate the influence to concave area. The proposed approach has been implemented and the performance of the approach is demonstrated on several real datasets. The results show that it is an effective way to restore the textureless and occluded surfaces of the object. And our algorithm is simple to implement, parameters tunable and robust for different models.