Study On High-precision Camera Calibration And Robust Stereo Matching

Given two or more images recorded from slightly different perspectives, a shape-from-stereo approach identifies corresponding points in all images that are projections of the same point in a scene. The ultimate goal of stereo vision is to establish the 3-D model of a scene according to the observations of it. How to obtain the depth information of the scene correctly is a crucial step. In a stereo vision system, the depth information is obtained by the matching between different images. After the continued efforts of many scholars, the great progress has been made in stereo vision. A lot of new stereo matching algorithms using the global optimization technologies, such as Graph-cut and Belief-propagation, have been presented in recent years. However, the stereo vision is an ill-posed problem. Considering the influences of textureless regions, occlusions, camera calibration accuracy, repeated scene patterns and other factors, it is still a challenging work to obtain the perfect depth information of a scene.This dissertation studies the problems of both high-precision stereo camera calibration and robust stereo matching in stereo vision from the viewpoint of obtaining the high-precision dense disparity map of a scene.Firstly, we addressed the problem of stereo camera calibration. Our objective is to construct a reliable and fully automatic system to supply a more robust and highly accurate calibration scheme. In our system, a checkerboard pattern is used as the calibration pattern. Firstly, we employ the extended integral image based detector to find the corner points on the corresponding grid patterns from the inputted checkerboard images. Then, an improved Delaunay triangulation based algorithm is used to connect the corner points to form the quadrilaterals that match the squares on the checkerboard pattern in 3-D space. The corner points obtained in such a way are used as the initial estimations of the actual corner points on the checkerboard pattern for calibration. In order to obtain the precise position information of actual corner points, one approach is to find edges of the grid patterns and fit them with lines. In such cases, the final estimation of corner points will be given by the intersections of these lines. However, due to the radial distortions, the edges are actually curved and the lines cannot fit them well. To address the problem, we proposed a global curve fitting method by exploiting the global geometric constraint among corner points. Specifically, according to the camera model adopted, we fit the edge points with a set of cubic polynomial curves rather than lines. The intersections of these curves are then regarded as real corner points in images. Finally, these extracted corner points are used as reference points to perform the computation task of the intrinsic and extrinsic parameters. The experimental results show that the geometrical constraint based method can improve remarkably the performance of the feature detection and camera calibration. At present, the actual 3-D reconstruction accuracy is better than 4‰, and can meet the request of precisely reconstruct a 3-D scene.Secondly, we addressed the problem of stereo matching. We presented a robust stereo matching algorithm based on the inter-regional cooperative optimization. The proposed algorithm uses regions as matching primitives and defines the corresponding region energy functional for matching by utilizing the color Statistics of regions and the constraints on smoothness and occlusion between adjacent regions. In order to obtain a more reasonable disparity map, a cooperative optimization procedure has been employed to minimize the matching costs of all regions by introducing the cooperative and competitive mechanism between regions. Firstly, a color based segmentation method is used to segment the reference image into regions with homogeneous color. Secondly. a local window-based matching method is used to determine the initial disparity estimate of each image pixel. And then, a voting based plane fitting technique is applied to obtain the parameters of disparity plane corresponding to each image region. Finally, the disparity plane parameters of all regions are iteratively optimized by an inter-regional cooperative optimization procedure until a reasonable disparity map is obtained. The experimental Results on Middlebury test set and real stereo images indicate that the performance of our method is competitive with the best-performing stereo matching algorithms and the disparity maps recovered are close to the ground truth data.