Research On Active Vision3D Reconstruction Based On Code Structured Light And Laser Scanning

To reconstruct3D topography of objects or scene automatically and accurately is the mostimportant issue in the area of machine vision. Some technologies involved have extensiveapplications in the field of advanced manufacturing, reverse engineering, robotics. Activevision technology which combined optical theory and image processing technology gainedcrucial research, for its remarkable advantages, especially to solve the matching problem intraditional stereo vision. However, there are still many unconfirmed theories and algorithms.It depends on the different needs and application backgrounds of different targets to treattargeted. For above mentioned reasons, this thesis mainly focuses on how to recover the3Dinformation from the surface of the object by active vision3D reconstruction methodology,for individual objects and outdoor scenes, which contains the3D reconstruction based onthe pseudo-random code structured light projection and the fusion method of3D laserscanning and camera SFM for3D reconstruction. The details include projector calibration,code feature points detection, image feature matching and point clouds fusion and other keytechnologies.Firstly, in order to handle the calibration of projector efficiently, including the projectorcan not be directly imaging and the parameters can not be easily obtained directly, a simpledirect projection ray calibration methodology is proposed based on the reverse imagingprinciple, so as to control the diffusion of the calibration errors in the system from thecamera to the projector efficiently. A method to solve the spatial coordinates by skew linesmodel is presented based on the thought of ray intersecting location, while the projectionray and imaging rays are not intersected at a point in practice. And it proves that the error issmall by using the midpoints of the common perpendicular of skew lines to approximate thespace points. A new method for feature points matching based on the window uniqueproperty of pseudo-random code image is proposed for3D reconstruction based on thepseudo-random coded structured light projection. In this technique, corner detectionalgorithm and image feature points matching are used. Thus, the matching problem can besolved by finding the correspondence between2D coordinates of feature points in the pixelimage and those in the code of the projected template. Experimental results demonstrate theeffectiveness of these methods.Secondly, an integrated framework of3D reconstruction fusion method based on3D laserscanner and camera-based SFM. for accurate6D SLAM is proposed, in the background ofthe robot6D SLAM,in order to processing the outdoor scene measurement datas that arelarge, easily overlap and have complex surface information. Through the analysis for theglobal consistent registration algorithm of3D laser scanning point clouds and theoptimization algorithms for SFM method to generate dense3D point clouds, with thestrengths and weaknesses of3D laser scanner and camera, a non-synchronous point clouddata fusion algorithms is given. Register by manual selection control points and utilize the result as the initial registration results. Then, the high precise registration3D point cloudsby ICP algorithm. At last, the point clouds in the laser scanning are matched colorinformation which comes from SFM dataset by PMVS method optimization in thedistribution of the color using the nearest point to achieve a3D reconstruction of largeoutdoor scenes. The3D reconstruction experiments of the buildings and outdoor scenesverifies the effectiveness of fusion method.