Research And Implementation On Key Technologies Of Vision Measurement System Based On Light-Section Method

Vision measurement is a non-contact measurement technology which is based on computer vision, the objective of the research involves space location, geometry size calculation and measurements of space attitude. Further more, it can be applied to 3D reconstruction and flaw detection.The main principle of vision measurement is that through image processing, the surface parameters of targets are acquired. In this paper, we aim at proposing a moving pile volume measurement solution, and make an intensive study on camera calibration, sub-pixel central point detection for structure light, motion blur and sub-pixel edge detection.In terms of calibration, we made a deep analysis on traditional calibration algorithms in the first place. Then, considering the situation that a large field of view in practical application needs to be calibrated, we used a novel nonlinear image distortion correction based calibration method. Experimental result shows that by using a distortion correction model, transformation between image coordinates and space coordinates can be realized. It also shows that not only the nonlinear distortion can be corrected satisfactorily but also calibrating precision is attained.In the research of sub-pixel edge detection, we firstly compared several commonly used orthogonal moment based sub-pixel edge detection algorithms, and compared the precision of them theoretically. Then, we proposed an advanced orthogonal moment based sub-pixel edge detection algorithm. It firstly uses gray integral to get the edge gradient direction, and we should determine the blur level along the gradient direction. For the ideal edge, moment based algorithm is able to satisfy our need, but for the significantly blurred edge it will cause a notable error. In this case, we introduced an advanced centroid method.In acquisition of structure light centre points, we used a modified gray fitting method for the real time processing. Experimental result shows that though the proposed method causes a little precision loss, it perfectly ensures the real-time property. Motion blur is also taken into consideration, because it directly affects the precision in acquiring sub-pixel light centre. In this aspect, we introduced a simple way to handle the instance of uniform motion in a straight line image.