Research And Application Of The Key Algorithm For Three-dimensional Reconstruction Based On Stereo Vision

Three-dimensional reconstruction is to obtain useful signal from images and get three-dimensional information of the measured object. With the rapid development of computer technology, three-dimensional reconstruction of image has become a research focus in the field of computer vision and is widely used in aerial mapping, visual navigation, motion analysis, medical diagnosis, industrial inspection and other fields. Therefore, the research has important theoretical significance and practical value.Recently, binocular stereo vision method based on Marr's computer vision theory is an important way of three-dimensional reconstruction. With two cameras simulating human eyes, images of the same scene are shot from different viewpoints, and depth of the object can be got by imaging theory of camera, then three-dimensional scene recovery is achieved according to the geometric principles. This paper is based on the theory of binocular stereo vision, focusing on the key technologies of stereo vision measurement: camera calibration, stereo matching, three-dimensional reconstruction, et al. In the camera calibration part, with the consideration of experimental apparatus and experimental environment, Zhang Zhengyou plane camera calibration method is used to obtain the internal parameters, for it is simple and with high precision, In the stereo matching part, several common similarity measurement methods used in region-based matching and feature-based matching are compared through experiments, and the stereo matching method based on Graph Cuts is focused on, in this method stereo matching is transformed into an energy function minimization problem, a grid map is constructed according to energy function, the function is minimized effectively using Graph Cuts theory, and stereo matching is realized. Upon completion of camera calibration and stereo matching, three-dimensional coordinates of the points on the surface of measured object are calculated using binocular stereo vision measurement principle, and the data is converted to depth image. Because of the noise from mismatching in the measurement process, image smooth algorithm is studied to solve the problem. Finally, three-dimensional reconstruction of objects is carried out and the shape is visualization output by OpenGL.In order to verify the feasibility of the algorithm, the every step of three-dimensional reconstruction has been validated through experiments. The method is used in medical diagnosis to measure the lesions and has achieved good measurement results.