Automatic Registration Of Terristrial Lidar Data Using Planar Targets

The registration of point clouds acquired from different stations is the key step in Terrestrial Laser Scanner (TLS) data processing. The quality of the registration not only determines the workflow of data collection, but also affects the accuracy of data processing and modeling. Most existing registration methods are manual or semi-automatic and therefore, low effcient in proceesing. This paper presents a fully automatic, robust registration scheme using man-made tagets. This two-step scheme consists of the automatic recognization of man-made targets followed by the robust coordinate transformation. Our contributions can be detailed as follows:Frist, laser scanning is introduced. A study on the working principle of laser scaner is presented, including the ranging principle, angle measuring principle, scanning principle and how to get the coordinates of point clouds from observations. The lsaer data proceesing technology is studied and some key steps are given.Second, the existing multi-view registration methods developed for TLS data proceesing is briefly reviewed. These methods can be grouped into three categries according to their principles of the registration. The method using man-made targets is choosen in this study as it has the advatanges over the others.Third, the point cloud of targets is recognized fron laser intensity using a threshold dividing method. A scan line based clustering analysis is put forward to segregate the targets from each other. To elilinate the noisy objects which have the similar reflectivity as the man-made targets, the planar characteristics, shape and size are taken into account. An automatic matching algorithm is ued to find out the correspondence targets form different standpoints. Compared with the coordinates derived from IMS, the accuracy is 1.4mm, better than that acquired by manufacturer's software.Fourth, an automatic and robust registeration model is built up. An automatic matching algtithm was used to fine out correspondence targets from diferent standpoints. There are several ways to represent rotation matrix, and each leads to a correspondence coordinate transformation metho, among which the Uniqution method have advatanges over the others. The RANSAC idea is used to form a robust transformation method. Experiment shows a better accuracy of 1.2mm.Finally, a software is programed to realize automatic registration od multi-view point clouds. The function module and data flow is designed and the key techniques such as data reading, data structure and visualization are detailed. The results from the software are showed at last.