Research On Multi-view 3D Surface Splicing Method Based On Artificial Landmarks

Due to the surface of the target beyond the camera view scope, the partial viewof the revolved body surfaces is blocked by other parts, or the measurement precisionrequirements, which make any angle measurement are unable to realize a completethe three-dimensional measurement work of the whole goal?the measurement processof these surfaces are often separated into several pieces or times. The normalization ofevery separated surface data is necessary in order to realize the object wholetopography measurement. Therefore three-dimensional surface splicing is a keytechnology to realize the object topography measurement, and it's also one of theimportant research fields of computer vision and digital image processing.A structured light3D detecting method which based on symbolic M-arrays wasput forward by our group. And our group realized the detecting and reconstruction ofdynamic3D scene in binary light projection. The3D detecting method is suitable forlarge targets which the curvature changes smaller. Because there have no remarkablecharacteristics on the object surfaces, therefore they can hardly be spliced by freesplicing method. Based on the three-dimension detection method, an alternatedsplicing method which based on three landmarks on the target surface is proposed inthis paper. The experiments were carried out, and the result shows that the proposedsplicing method has both the accuracy of certain splicing and the high robustness.The method is that a group of landmarks are set on the surface of the target.Through extracting the landmarks?which make us get the image coordinate of thelandmarks. And the world coordinate of the landmarks are solved according to theimage coordinate of the landmarks. Then we can solve the different perspectivetransformation matrix between the coordinate systems on the basis of the worldcoordinate of the landmarks, which realize the3D surface splicing. The main contextof this paper includes the following respects:(1) The landmarks recognition and location. Artificial landmarks are set on thesurface of the detected object, using image smoothing and image segmentation topreprocess the image of the landmarks. Recognize the landmarks by using the numberof white connected domain of the landmarks. Then locate the center of the landmarksby Least Squares Fitting Method and calculate the3D world coordinates of the landmarks by analyzing the coordinate of the key points. At the same time,3D surfacewas generated by Matlab in experiments.(2) Study on surface splicing method based on three landmarks. This paperstudy the three-dimensional surface splicing based on three landmarks and principle,derived the solving process of transformation matrix. Then unifying the differentworld coordinate systems to the same coordinate system according to transformationmatrix, Also introduces some commonly used surface splicing methods on base ofthree landmarks. Propose the evaluation standards how to judge the image aftersurfaces splicing and provide method of surface smooth processing.(3) Study on optimized selection of landmarks in splicing system includingmany landmarks. In the surface splicing of the nearby surface public zone includedmany landmarks, this paper analyzed the advantage and deficiency of algorithm basedon characteristics and iterative. A method how to choose the best three of all thelandmarks as the splicing landmarks was proposed. This paper also defined theprinciple of optimizing selection about the landmarks, analyzed the impact on theaccuracy by using different landmarks chosen methods and verified the reliability andfeasibility of the principle by a series of experiments.