Alignment of 3D laser scanner point clouds using photogrammetric targets

Abstract

3D laser scanning technology has found to be an excellent method for the modeling and measuring 3D objects (rigid and real life). The 3D point clouds of an object can be acquired within less than one second (0.6s for Minolta VI-910 3D laser scanner) and stored digitally for pre-processing task. Complex mapping of 3D object such as human faces required at least two scanning images to cover the complete facial area (from right ear to left ear, and from hair line to bottom part of the chin) with optimum 3D modeling accuracy. For complete 3D model generation, the scanning images is needed to be registered (combined) and merged together. Existing registration method used corresponding features between the two scanned images as registration primitive and finally 3D transformation algorithm was applied to combine the images. This paper described the use of photogrammetric targets, as registration primitive to combine two scanning images of human face. The so called “paper targets� were setup on the special design photogrammetric control frame where the human face was placed at the middle of the frame during scanning process. The photogrammetric control frame was calibrated using close-range convergent photogrammetry with coded targets and high precision scale bars to determine the precise 3D coordinate of such targets. The targets was also included in the scanning images and represented as point clouds datasets. Via laser scanning images, the centroid of the targets was measured and the 3D transformation algorithm was successfully applied to transform the scanning point clouds from laser scanning coordinate system to photogrammetric coordinate system. The output of the registered point clouds was displayed and processed in reverse engineering RapidForm 2004 software. The accuracy of the method was evaluated using shell-shell deviation analysis technique where the average deviation of the two scanning images was calculated. The results show that the accuracy of existing 3D registration method was recorded as 0.3-0.5mm, while the proposed 3D registration using photogrammetric targets was recorded to be 0.129-0.285mm. The reliability of the photogrammetric targets method was also evaluated and the measured accuracy is from 0.216-0.741mm with one standard deviation.