Research On The Extraction Of Depth Information Of Binocular Visual And Its Key Algorithms

To retrieve the depth information from the disparity which based on the binocular stereo vision, we use the two cameras in different places shooting the same scene to get the two-dimensional image and then calculate the depth information, at last reconstruct the three-dimensional model or get the distance from the object to the camera. Binocular stereo vision has many advantages such as it has simple and inexpensive equipment, high precision and speed, retrieve the depth information without touching the object. So it takes an important role in the UAV power line inspection and the construction robot. This paper introduces the camera calibration, image pre-processing, stereo rectification, stereo matching, depth extraction and three-dimensional model reconstruction, compared with common algorithm and puts forward some improved methods to meet the practical application. The main researches are as follows:1、This paper describes the imaging model of the camera, the transformation relationship between these coordinates, the intrinsic and extrinsic parameters of camera, focus on the Zhang’s chessboard calibration method. Based on Harris corner detection, this paper proposes a new method to accurately determine the coordinates of the corner. This method eliminates the corner out of the chessboard which we do not care about, and the repeated and overlaying corner or not “ideal” corner. At the same time, calculate the corner coordinates accurately, improve the accuracy of the calibration results.2、In view of the stereo rectification which we can’t get the camera intrinsic and extrinsic parameters, this paper focuses on the Hartley’s algorithm, this method must estimate the fundamental matrix and restore the Epipolar geometry. Then the key of estimating the fundamental matrix accurately lies on the exact matching of the matching point pairs. Therefore, a new Features based on Harris-SIFT is used. The Features can reduce the computation time and error matching rate of the feature matching, it can also directly represent the structure characteristic information of the image. Finally, We compared the results of Harris-SIFT Features extraction and SIFT Features extraction in terms of computation time and error matching rate. The results show that Harris-SIFT Feature matching can improve the matching accuracy, thus improve the accuracy of estimate of the fundamental matrix, and ultimately improve the efficiency of the Hartley’s algorithm.3、This paper presents a method that extracting depth information from multiple angles, and then rotating, shifting and overlapping the different depth maps to increase the depth information,researches on the need of stereo matching such as constraint criterion, matching primitives, matching strategy. Focus on three kind of stereo matching algorithm: GC, SGBM, BM. Compared the three algorithms by matching the standard image pairs, analysis the rate of error matching rate and computation time. This paper introduces the Least Squares method of extracting depth information, at the same time. At last, this paper used the color point cloud to reconstruct the color Three-dimensional model.4、This paper designs and implements a binocular stereo vision system to extract the depth information. This system’s interface is sample, the devices are cheap and it can not only get the depth information from image pairs quickly, but also obtain the color Three-dimensional reconstruction model.