«LASER SCANNING ISSUES FOR THE GEOMETRICAL RECORDING OF A COMPLEX STATUE Maria TSAKIRI1, Charalambos IOANNIDIS1, Alistair CARTY2 1 School of Rural and ...»
In this project, the simplest and most accessible for the Hermes raw dataset technique was implemented; a simple conversion of the proprietary Minolta format data to ASCII PLY files which were then compressed with gzip. Each scan comprising the scan metadata, such as laser power used, focal length and so on, the raw 3D coordinates of the range grid and per-vertex colour information.
This technique, although not the most efficient in terms of storage space, does virtually guarantee the readability and completeness of the archived raw datasets in the future.
In scanning the Hermes statue, 649 scans were acquired resulting in 143,652,299 range samples and 269,178,117 triangles. In total, the raw data requires around 10Gb of space uncompressed, 4Gb compressed. Therefore, it was felt unnecessary to use a more involved data management system.
Cultural Heritage 6
3 PROCESSING ASPECTSThe post-processing of the acquired laser data comprises mainly the tasks of aligning and merging the scans to produce the 3D solid model of the structure.
3.1 Alignment Alignment of scans is performed in order to bring the hundreds of scans of a statue, acquired from different locations, to a common coordinate system. Unlike photogrammetric methods, which make use of reflective targets or pre-signed points as control points, registration of scanned images requires a different approach. There are a variety of techniques used, with the most common being the iterated-closest-points (ICP) algorithm (eg Besl and McKay 1992) which finds matching points on two meshes, computes the rigid transformation that minimises the squared distances of these point pairs and iterates until some convergence criterion is met. Modifications to the original ICP algorithm have been made to improve the rate of convergence and register partially overlapping sets of points (eg Chen and Medioni 1992; Zhang 1994). For complex objects such as statues, an extended ICP algorithm to minimise the sum of squared distances for all views simultaneously using multiple range images is preferred to ensure an even distribution of registration errors between overlapping views.
To produce the final Hermes model, sophisticated in-house software by Archaeoptics Ltd., was used to perform the alignment of all the scans, including a global alignment phase, as well as the integration of each scan into the final model and subsequent hole-filling and mesh repair. The full global alignment of all 649 scans took about 100 hours to perform using a fairly standard PC (2GHz Pentium IV, 1Gb RAM).
The goal of merging is to integrate registered sets of surface measurements into a single 3D surface model. The generic problem of surface reconstruction is to estimate a manifold surface that approximates the unknown object surface from a set of sampled 3D points, without making any assumptions about the surface shape. The two approaches reported most frequently for fusion of multiple overlapping surface measurements into a single surface model are mesh integration (Turk and Levoy 1994) and volumetric fusion (Curless and Levoy 1996).
In the case of Hermes, merging of scans was undertaken by in-house software by Archaeoptics Ltd.
using a hybrid approach of mesh integration with volumetric hole-filling. It was decided that the resampling undertaken during volumetric fusion would get away from “what the scanner saw” and, unless resampled at a computationally crippling level, would lose much of the surface detail we wished to retain. Therefore, we used an in-house mesh integration technique which accurately fuses overlapping scans together.
However, the final stage of processing, hole-filling and mesh repair, is difficult when merely relying on the mesh data itself. This is typically due to holes being non-simple and most often nonplanar either due to poor triangulation choices as input to the fusion algorithms, or poor quality output from the fusion process. To efficiently handle holes with extremely ill-defined boundaries and holes that might span areas of high curvature, we use a variant of the volumetric diffusion method (Davis 2002). This technique ensures efficient and effective automatic hole-filling at both a geometric level and at an aesthetic level. A merged model of the Hermes statue using the above approach is shown in Fig. 4.
Tsakiri, Ioannidis, Carty 7
Fig. 4. A merged model of Hermes statue
4 CONCLUDING REMARKSDocumentation of cultural heritage artefacts using laser scanning is one of the most active areas for the use of this technology, with the construction of 3D digital models as one of its principal objectives. With a significant proportion of museum artefacts made of marble and having complex surfaces, it is important to deal with specific aspects of acquiring, aligning, merging and viewing scanned data.
The Hermes project was performed using an “off-the-shelf”, high resolution, laser triangulation scanner which proved more than sufficient in capturing details smaller than one millimetre. Data capture comprised of 649 scans because there was an overlap averaging around 30% in each scan to avoid holes and missing data due to occlusions. Handling the massive amounts of data in a timeefficient manner is an extremely important problem that is faced only with this type of scanned objects.
Although laser scanning is a preferred method to record data economically and efficiently in order to create 3D models of complex statues, it is important to emphasise the need of using alternatives methods such as photogrammetry to capture data, which can provide control and ensure that the models are geometrically correct. During the Hermes project photogrammetric recording was also performed and a comparison of the models between the two techniques will be made.
REFERENCES Besl, P. and N. McKay (1992). A method for registration of 3D shapes. IEEE Trans. PAMI, 14(2):239-256.
Bitelli G., A. Capra, A. Zanutta (2002). Digital photogrammetry and laser scanning in surveying the Numphea in Pompei. Proc. CIPA WG6 Int. Workshop, Sept. 1-2, Corfu, Greece (CD-Rom).
Chen and Medioni (1992). Object modelling by registration of multiple range images. Image and Vision Computing, 10(3):145-155.
Curless B. and M. Levoy (1996). A volumetric method for building complex models from range images. Computer Graphics Proceedings SIGGRAPH96, ACM, pp303-312.
Davis, J., S. Marschner, M. Garr, M. Levoy (2002). Filling holes in complex surfaces using volumetric diffusion. Proc. 1st Int. Symp. on 3D Data Processing, Visualization, Transmission.
Godin, G., M. Rioux, J-A. Beraldin, M. Levoy, L. Cournoyer and F. Blais (2001). An assessment of laser range measurement on marble surfaces. Proc. 5th Conf. 3DOptical, Oct. 1-4, Vienna, Austria.
Henz G. (2002). PharaohPepi I: documentation of the oldest known life-size metal sculpture using laser scanning in surveying and photogrammetry. In: Proc. CIPA WG6 Int. Workshop, Sept. 1-2, Corfu.
Levoy M., K. Pulli, B. Curless, S. Rusinkiewicz, D. Koller, L. Pereira, M. Ginzton, S. Anderson, J.
Davis, J. Ginsberg, J. Shade, D. Fulke (2000). The digital Michelangelo project: 3D scanning of large statues. In: Proc. SIGGRAPH, 131:144.
Pulli K. (1999). Surface reconstruction and display from range and color data. PhD Thesis, University of Washington, USA.
Turk J. and M. Levoy (1994). Zippered polygon meshes from range images. Proc. Computer Graphics Proceedings SIGGRAPH94, ACM.
Zhang J. (1994)..Iterative point matching for registration of free-form curves and surfaces.
International Journal of Computer Vision, 13(2):111-152.