Research On Measurement Technology Based On Binocular Stereo Vision

Vision measurement technique is an advanced non-contact technique that has the advantages such as high speed, high accuracy and low cost. Therefore, it has got wide development and application in industry, especially in manufacturing industry for testing products. On the other hand, stereo vision algorithms need to be modified or improved in such aspects as real-time characteristic, robustness and higher accuracy. In this paper, we have studied key theories about three-dimensional measurement technique based on stereo vision. The main concerns are as follows:1．After analyzing and investigating three traditional camera calibration methods including direct linear transformation, radial alignment constraint and Zhang's method, a model plane is produced and a stereovision measuring system is built. Using the system as the platform, the corner coordinate was accurately measured with universal tool maker's microscope. Then, we obtained all intrinsic and extrinsic parameters of Mintron's camera (MTV-2346CA) applying Zhang's calibration method. The reprojection error is within 1．2 pixels in x-axis, direction and 0．8 pixels in y-axis direction.2．Harris corner extracting algorithm is improved in this dissertation. Adopting quadratic surface fitting method, we have obtained sub-pixel corner coordinate and analyzed quantitatively and qualitatively in terms of stability, reliability and anti-noise ability. Results showe that modified Harris method has good location capability. We have also studied three geometry feature extraction methods about line, circle and ellipse. Line has been detected using classic Hough transform. We use two-steps method as a substitute for Hough transform method on efficient extracting and detecting circles. Ellipse has also been detected efficiently by adopting least square fitting method.3．The area correlation algorithms including SAD, SSD and NCC functions have been investigated. By contrast, we adopted dynamic programming and graph cuts to improve the matching speed and the accuracy. Two modified matching algorithms have been proposed, fast stereo matching based on dynamic programming and stereo matching algorithm based on graph cuts. The experiments showed that matching result of the latter algorithm was better than the former, and the probability of mismatching was reduced.4．The proposed method has been applied in engineering and multi-dimensional measurement of complex workpieces are implemented with a minor relative error less than 0．9%. Additionally, 3D surface reconstruction has also been investigated and partial cylinder profiles are reconstructed.