Research Of Eagc Refinement Of LIDAR Point Cloud Based On Image

3D reconstruction technology has been a hot research in computer vision andphotogrammetry and remote sensing and other fields, and is widely used in industry,agriculture, medical, cultural heritage protection and other related fields, especially with therapid development of digital city and intelligent city,3D reconstruction of the objects used toobtain the accurate3D models using images will be the focus of the future research anddevelopment trend. As a new technology, with its rapid, accurate, non-direct contact to obtainscanned3D information,3D laser scanning technology is widely used, the point cloud data byscanning can well complete the3D reconstruction, but as a result of scanning device andscanning conditions, not all ideal reconstruction results can be gotten in any case.Taking the image as the research object, according to the related theory ofphotogrammetry, this paper is mainly to study the reconstruction for the solid edge and assistpoint cloud data to complete3D reconstruction, in order to compensate for the inaccuratereflection in the scanning object edges using the3D laser scanning technology and realize theedge refinement of point cloud. This paper focuses on the research of image edge extraction,image matching, relative orientation, intersection and data registration technology and so on.In the stage of the image edge detection, by comparing the effect of several kinds of edgedetection that are commonly used, this paper selects the Canny operator as the extractionoperation and gets ideal results by specific filtering before the image edge extraction andsetting the appropriate threshold in the image edge extraction.In the stage of the image matching, it mainly includes matching the edge points detectedin the previous step and the dense matching in order to reconstruct the ideal image3D modelto fuse with the point cloud data. Among them, feature points are extracted ty grid accordingHarris operator before dense matching. This paper takes a coarse to fine matching strategy,namely searching the homonymous points of the matching points point by point along theepipolar line with the relative orientation of stereo pairs, then complete fine matching with theLeast Squares Image Matching, which ensure the reliability of the matching results.In the stage of the relative orientation and intersection, based on the relative orientationtheory and method of the continuous stereo pairs, this paper adopts the oriented strategy ofdirect-rigorous solution to get accurate relative orientation elements, and obtains the reliable3D model by using the point projection coefficient method of intersection.In the stage of registration between the3D model generated from images with theLIDAR point cloud data, this paper mainly adopts the manual and automatic two registrationstrategy. One method is to manually pick the homonymous points in both images and pointcloud data, then complete registration by calculating the model transformation parameters.Another method is firstly to match the reflection intensity image generated ty point cloud datawith the general image, secondly to index into the homonymous points in the image3D modeland3D laser scanning point cloud data, lastly to compute the transformation parameters tocomplete the automatic registration. Though the strict control on the accuracy of each stage, this paper successfully completethe edge3D reconstruction based on images, and well fuse with the LIDAR point cloud datathrough registration, and realize the3D laser scanning point cloud edge refinement, which hasvery important significance for the application of fine object3D model reconstruction throughpoint cloud data.