Research On Technology Of Self-Reference Based Object Digitization

Digitalization is an important research area in the geometric measurement technology. In order to overcome defects of traditional digitalization means, such as lack of flexible, existing scanning dead ends and limitation in object size, this paper does a deep research on self-reference based digitization.An algorithm to extract the center of uncoded circle marks is proposed. Firstly, subpilxel edges are got by combine using of the Snake and the Zernike operator. Then accurate centers are found through robust ellipse fitting method. Exp eriments demonstrate that center detection error is less than0.02pixels.In order to calibrate system parameters accurately and conveniently, a laser plane calibrate method base on stereo vision is developed which utilizes system structure fully. Compared with other popular methods, the main contribution is that the proposed algorithm needs no calibration target. Dense calibration points can be got by projecting structured light to any object, which widen its application scope.Algorithm of self-reference is studied. Firstly, a novel stereo match algorithm using both the eliline constraint and the instrument motion consistency is presented. Then, corresponding marks from two views are got by their topological invariant in different coordinate systems. At last scanning data from two views are registered by found corresponding marks. To improve scanning accuracy, data from all views are refined by global optimization after scanning.Colored point cloud acquisition and its color smoothing are discussed. The color of points reconstructed from current frame is restored by projected to former images in real time. To filter noise color introduced during scanning, an index of color variation is introduced and an offline adaptive smoothing algorithm based on it is presented, which can smooth color while keep feature perfectly. At last, necessary data post processing is studied. Outliers are filtered by combine using of region growing and local outlier factor, and remained clean points are triangulated by a new developed surface reconstruction algorithm. The algorithm presented consists of two steps:an initial triangle mesh is constructed by repeating a simple advancing front rule. Then, initial triangles are subdivided according to surface various and triangle vertexes are refined by means of Multilevel B-spline fitting.