Registration Of Range Images And Planar Regions Extraction From TLS Point Cloud

Terrestrial Laser Scanner (TLS) is a kind of new instrument evolved rapidly in recent years. It captures high density and high definition 3D point clouds from objects efficiently. Because the terrestrial laser-scanning technology is new, there are not much more research works that has been made in this field. This paper focuses on the post-processing algorithm of urban scene range images. Following works have been done in the paper:(1) Multi-view point clouds registration.Normally, several observations will be needed to gain complete information while scanning an object using laser scanner. So, multi-view point clouds registration algorithm is necessary to be studied. At present, a common drawback exists in most registration algorithms: the amounts of calculations for closest points matching are very large. This problem is solved by the method presented in this paper. By using grid interpolated from origin point cloud, efficience of closest points searching is improved. It has been proved by experiments that the method is reliable, easy to be used, and efficient.(2) Segmentation primitive extraction from scan line data.Because of the existing of amount of gross error points in point cloud, many robust algorithms have been adopted to processing range images data. These algorithms often randomly establish large amount of modules firstly, and then identify them individually. This practice is very inefficient. This paper presents a improved RANCAC algorithm which fully utilizes the intrinsic attributions of scan line data. The algorithm is both efficient and roboust. More rational segmentation result can be obtained using the presented algorithm.(3) Fitted plane extraction using segmentation primitives from scan lines.The ultimate target of range image processing is to segment point cloud data and extract geometric features. This paper only focuses on the planar regions extraction. Same with the extraction of segmentation primitives from scan lines, the fitted plane extraction method also fully utilizes the intrinsic properties of range image captured by TLS. The bottleneck for fitting plane using end points of fitted lines segments directly is how to determine the weights of each end point. This paper solves this problem. Furthermore, this paper presents a set of integrated algorithm for invalid fitted planes removing. Experiments proved that perfect segmentation results can be achieved using the presented algorithm.