Three-dimensional Reconstruction Of Face Based On Binocular Stereoscopic Vision

Face is the significant information to identify others.Three-dimensional information can be applied in many fields and directions of machine vision.Therefore,it is of great significance to stereoscopicize common face images.Compared with other three-dimensional reconstruction systems,binocular stereo vision system has the advantages of low price and simple operation.This dissertation focuses on the three-dimensional reconstruction method of the planar images collected by the binocular stereo vision system.Based on the depth information that generated by the parallax,the three-dimensional reconstruction is performed by the binocular stereo vision system.First of all,according to the basic principle of binocular stereo vision system and the theory of binocular stereo vision system that parallel configuration,the work of building the binocular stereo vision system was completed.On the basis of Zhang Zhengyou's method,the binocular stereo vision system was calibrated,and the error of the calibration result was analyzed.Second,facial information image pairs from multiple angles were collected by the binocular stereo vision system.This article used image correction method to correct the captured image pairs.According to the basic principle of stereo matching,the advantages and disadvantages of various traditional image matching methods were analyzed.On the basis of completing the image correction,the graph cut matching method was selected as the matching method,and the collected image pairs were matched by this method,then the multi-angle face disparity map was obtained.Finally,disparity maps from different angles were converted into point cloud maps.These point cloud maps were registered and merged to form a dense point cloud map.An algorithm for streamlining the point cloud was proposed.The point cloud was simplified with an improved point cloud simplification algorithm,and a streamlined setting was added to make the simplification effect controllable.The reduced point cloud was meshed with the Delaunay triangulation method to obtain a triangular mesh map of the human face.The texture map was based on the triangular mesh to obtain a realistic 3D face surface.