3D imaging systems are line of sight instruments and multiple scans from different locations are often needed to get a good representation of an entire scene. Therefore, registration of different datasets to a common coordinate system is required. Automatic registration depends on correct identification of common points in the individual scans. Determination of these common points in large datasets is far from trivial and may be very time consuming. Target-based registration simplifies calculations which greatly reduce computation time but this approach depends on a reliable shape detector. Centers of sphere targets are convenient candidates for common points used in registration. This paper describes a method to rapidly find multiple spheres of known radius in a large dataset obtained from a 3D imaging system. The proposed method requires a dataset to be in spherical coordinates and regularly gridded. A lab was scanned from two locations at three different scan densities. Four spheres were located in the lab, and the total number of points hitting the four spheres was a small fraction (10-4) of all the points in the dataset. In all cases, the computation time to identify the spheres is a fraction (10 %) of the scanning time. The identification of the spheres depends on scan density and distances of the spheres to the scanner. At the highest scan density, all four sphere centers were correctly determined in both scanner positions with no false positives.
Citation: Automation in Construction
Pub Type: Journals
3D imaging system, LADAR, object recognition, sphere and surface fitting, target-based registration