Page 233

REVISTA IEEE 11

233 José Balsa Barreiro, Dieter Fritsch 3D Modelling of historic urban centres and… Point cloud registration The registration of point clouds is a process whereby the successive point clouds are assembled. In the case of this particular project, there are two successive stages to the assembly process: (a) merging the point clouds obtained by laser scanner to obtain a single point cloud, and (b) merging the latter with the point cloud obtained by aerial photogrammetry techniques. Cyclone software can perform the registration of the point clouds obtained with laser scanner. By applying this procedure to the laser scanner information, a complete point cloud of the entire external block of buildings is obtained, which also includes the surrounding area. However, the information relating to the inner and/or top part of this entire block of buildings could not be obtained due to physical access restrictions. These data gaps are not limited exclusively to these areas, but may also appear in part of the buildings facades due to the presence of trees or other elements that could hinder the laser beam at the time of scanning. For this reason, the previous point cloud must be linked to that obtained by photogrammetry, which contains information on the upper part of the buildings (Figure 5.a). The point cloud generated by aerial photogrammetry is geo-referenced in the Gauss-Krueger coordinate system, which is used as a reference system common to all point clouds by a process called transformation. For that purpose, the different generated clouds must be combined and fitted in position, orientation and scale, by applying a seven-parameter 3D Helmert transformation, comprising three translations (T), three rotations (R) and a scale factor (λ). For the calculation of these parameters, there is a requirement for more than seven equations to be presented, while in practice at least three common points are required between the different point clouds. Generally, the location of these common points in the respective point clouds must be achieved manually, which can lead to less than ideal results. For this reason an alternative is usually chosen involving the application of the Iterative Closest Point (ICP) algorithm to obtain a better result50. It is important to note that this Image 5: (a) point cloud obtained from aerial photogrammetry. (b) Point cloud finally obtained after registration and transformation procedures are carried out 50  BESL, Paul J. and McKAY, Nail D.: «A method for registration of 3-D shapes», IEEE Transactions on Pattern Analysis and Machine Intelligence, n.º 14(2), 1992, pp. 239-256. http://revista.ieee.es


REVISTA IEEE 11
To see the actual publication please follow the link above