High-accurate Matching Algorithm Research In Binocular Stereo Reconstruction

3D reconstruction is more and more important with the development of 3D print and virtual reality and it’s a challenging mission to acquire the accurate model of an object. Many kinds of methods have been put forward in recent years, but calculating the model by a serial of calibrated images is the most convenient way. In this paper, we present a stereo matching approach under the condition of multi-resolution combined with the multi-constraints to compute the 3D model of an object.In the first step, we calibrate the cameras respectively, and then execute the stereo calibration and rectification procedure. This will be helpful when searching corresponding points. To reduce the computation complexity, we cut the object from the background. Then we build the image pyramid and stereo matching happens from the top level to the lowest level. In each level, we use the NCC method to finish the original matching and then apply the three constraints to reject the mismatch points and mark them in the disparity image. Next we execute the stereo matching procedure for the marked pixels again. As the disparity map still contains some noise, we implement a weighted median filter algorithm to de-noising. At last, we execute the disparity refinement to acquire an accurate and smooth disparity map. During the re-match procedure, we propose to choose the pixel in the eight-neighborhood which has the highest match cost to guide the matching of center pixel. In the disparity refinement, we design an anisotropic smoothing kernel to smooth the depth map and keep the detail at the same time. Combine with the calibration parameters; we can compute the 3D point cloud. In order to get rid of the outliers, we adopt a statistical filter to reject them. Finally, we estimate the normal vector and use the Possion reconstruction method to generate the mesh model.We test six datasets for our algorithm and found that we can generate accurate mesh model for an object. The accuracy and the completeness are all reach the top level in the world.