Research On Three Dimensional Accurate Measurement Method Based On Stereo Vision

As an important branch of computer vision, stereo vision uses one or more cameras from different spatial location to obtain two-dimensional images in the same three-dimensional scene to simulate the human eyes'perception of the external environment. Through the stereo matching of two images to obtain the three-dimensional scene depth and geometry information, stereo vision can reconstruct the three-dimensional shape and position. Because the widely application prospect of stereo vision such as aerospace, industrial production, military application and game industry, stereo vision is still one of the hot topic in the world.Our work focuses on the intrinsic and extrinsic camera parmeters calibration, stereo matching algorithm and error analysis of depth reconstruction algorithm. The main contributions of our work can be summarized as follows:First, the most laboratory camera calibration algorithms require expensive and high precision calibration equipment, so we put forward a kind of calibration model and algorithm without high precision calibration equipment, we present a method to calibrate the intrinsic parameters of camera in laboratory. We take calibration method of ordinary camera and installation error into account in the calibration model. The installation error is modeled by combination of three typical effects:the installation of PTZ initiates status, PTZ and CCD. The proposed procedure consists of a closed-form solution, followed by a nonlinear refining based on maximum likelihood criterion. Both computer simulation and real data have been used to validate our method, and very good results have been obtained. The application in multiple CCD camera image mosaic system is also proved that the algorithm is effective and practical.Second, the wide field monitoring environment may only a few control points can be used for the application of rapid camera attitude calibration, this paper proposes a new real-time method to calibrate two rotation angle of PT camera by using only one control point. This is achieved by assuming that the intrinsic parameters and position of the camera are known in advance. Our method is non-iterative and does not need features matching thus its time efficiency is better. We evaluate our calibration method on various synthetic and real data. The quantitative results demonstrate that the proposed method outperforms other state-of-the-art methods if the intrinsic parameters and position of the camera are known in advance.Third, based on the analysis of primitive matching algorithm and the efficiency of semi-global dynamic programming search algorithm, we propose a semi-global stereo matching algorithm based on block coding. In order to reduce the complexity of the matching algorithm, we divid the image into to several blocks and then block coding separately for each block with gray and gradient direction histogram statistics coding to eight-bit information matching. Matching by semi-global dynamic programming algorithm, from 16 directions to find the optimal matching points, and result a block parallax image. For the boundary effect in parallax image, we use double linear interpolation method to calculate the block within each pixel in parallax and get the final match parallax image. Finally, we apply algorithms to public images database, compared with the state-of-the-art matching algorithm, verified the accuracy and efficiency of our algorithm.Finally, the practical application of the three dimensional measuring error analysis is usually difficult to directly access to the analysis results, we propose a method based on the analytic geometry to analyze all kinds of traditional error sources on the result of 3D measurement. We summed up and simplified all the input errors to five parameters by rotation transformation. Then we use the fast algorithm of midpoint method to deduce the mathematical relationships between target point and the parameters. Thus, the expectations and covariance matrix of 3D point location would be obtained, which can constitute the uncertainty region of point location. Afterwards, we turned back to the error propagation of the primitive input errors in the stereo system and throughout the whole analysis process from primitive input errors to localization error. Finally, extensive experiments are performed to verify the performance of our methods.