Research On Multi-view Registration Technology For Large Size And Complex Surfaces

Multi-view3D data registration is the key to obtain the complete3Dshape measurement data and it has a very wide application value in the currentengineering. This paper research on multi-view3D data global registrationmethod and its key technology based on control codeless points, and used thestructured light stereo vision scanning system that has established in thelaboratory.Quickly and accurately positioning mark point center in the image is theprerequisite of realizing multi-view global registration. It analyzes the markpoint that commonly used in current engineering in this paper, select theappropriate mark point, filter to the image noise, adaptive threshold imagesegmentation, recognize and extracted the contour edge characteristic of themark points, least squares ellipse fitting and iterative fitting of the contouredge characteristic of the mark points to position the mark points center. Inthe filtering to the image noise link, compare with three kinds of filter whichare frequently used, finally select the template for3×3of Gauss Filter.Within the mark points feature contour edge recognition, the rules of closure,length, shape and gray are proposed and used to eliminate the feature contouredge of not mark points, and validated by an experiment.It proposed line criterion of coarse matching, coding structured light offine matching and supplementary matching for the leaky matching points inthe aspect of control points matching between the left and right image in theglobal registration, then it achieve the control points of the left and rightimage matching one by one. It also proposed the distance criterion, the anglecriterion and the auxiliary leg criterion based on space invariant feature ofcontrol points in matching the same name control points between one viewand the global coordinate system, and it realize the same name control pointsmatching one by one. Finally, through the same name control points and SVD algorithm to calculate the optimal transform matrix from one view to theglobal coordinate system, so it realize the multi-view3D data globalregistration.At last, Use the structure light stereo vision scanning system to scan theshoe last and the anti-collision board of the automobile’s gasoline tank withseveral views and adopt the global registration method in this paper. The shoelast and the anti-collision board of the automobile’s gasoline tank realized themulti-view3D data global registration and take the final registration resultscompared with the results of handheld scanner. The standard deviation is0.167mm and0.266mm and the coincidence degree average error of everyview control points is within0.025mm, so it verified the feasibility andaccuracy of the multi-view3D data global registration method.