Study Of Dense Depth Acquisition Techniques Based On Ground Control Point And Segmentation

Stereo vision is a hot issue in the field of computer vision, shooting the same scene taken from different viewpoints, by processing two or more captured images to establish correspondence between the observed features in order to establish three-dimensional model. With the development of computer technology, stereo vision application of the theory are also expanding, exhibiting its superiority, particularly in the fields which need long-term monitoring and personnel are unavailable.Being sustained by National Natural Science Foundation of China, this paper studied the robust high-precision stereo matching and stereo camera calibration problem detailedly. In the stereo camera calibration, introducted the high-accuracy calibration method Zhang, overcome the shortcomings like, high requirements for equipment and complicated operation, established stereo camera calibration program, and built the practical stereo vision system besed on that; In the stereo matching algorithm, proposed an algorithm besed on the ground control point and image segmentation. Image subregion as the matching primitive, using the subregional feature and block or smooth relationship between the adjacent region to determine the matching energy function, and introducted the dynamic programming algorithm, optimized the defined energy function globally to get the minimum value, ultimately, using the image segmentation information to fill up the mismatch of the occlusion and tectureless region to get the ideal disparity results. And tested in the common standard of the current test platform. The results show that high quality dense disparity map can be obtained using our approach, and the accuracy is reasonable considering the relatively low hardware requirement.Finally, carried out actual experiment based on the established stereo vision system. There have three main steps:first, capture the image using CCD camera. Second, epipolar rectification to correct the image radial distortion. Third, using the proposed algorithm to obtain the corresponding disparity map. So, in the further way, could obtain the platform needed data by analyzing the camera calibration geometric parameters.