3D scanning data processing and registration for the construction of urban scenes via image methods

Abstract

ABSTRACTThe difficulty of the 3D registration mainly lies in the huge computation for a precise alignment, especially for large data of urban scenes. In this paper, we proposed an algorithm that converts this 3D problem into 2D case. The main idea is to map the point cloud onto the image. Then, SIFT algorithm is adopted to detect the key points of the images, and key points from two images corresponding to the two point clouds ready to be registered are matched to form several pairs. Next, the key point pairs construct an intrinsic link between the images. From the one-to-one correspondence between pixel and point, this relation can be converted back to 3D space, according to which a transform matrix can be consequently established. The resultant matrix aims at guiding the spatial transform to achieve an ideal 3D registration. Later experiments illustrate that it will obtain a preferable result in much less time.