3D MODELING OF LARGE AND COMPLEX SITE USING MULTI-SENSOR INTEGRATION AND MULTI-RESOLUTION DATA G. Guidi 1, F. Remondino 2, 3, M. Russo 1, F. Menna 4, A. Rizzi 3 1 Dept.INDACO, Politecnico of Milano, Italy 2 Institute of Geodesy and Photogrammetry - ETH Zurich, Switzerland 3 Centre for Scientific and Technological Research - Bruno Kessler Foundation, Trento, Italy 4 Dept. of Applied Sciences, Parthenope University, Naples, Italy
Large and complex site: Pompeii Founded by Samnitic tribes in VII century BC Became roman in the II Century BC The Romans transformed it in a nice vacation place with elegant villas, a Circus, and several amusement places In 62 AD a strong earthquake seriously damaged Pompeii and its monuments In 79 AD the city was suddenly destroyed by a huge eruption of the Vesuvius that covered with meters of ash the whole city For this reason the city is well conserved VII BC II BC 62 AD 79 AD Pompeii history foundation Romans earthquake eruption 2
The modeling project - Reality-based 3D modeling of the Roman Forum in Pompeii with all its scattered archaeological finds and the mural structures facing the area - Identify and solve specific problems in digitizing and modeling large and complex sites with high spatial resolution and accuracy - Intrinsic multi-resolution survey using different 3D techniques (multi-sensor) - Interactive visualization of the final 3D model 3
The area of interest for the modeling project The forum and all the buildings facing it N Large area: 150 m maximum length 80 m maximum width 8 large mural structures Over 300 small finds spread all over the area 150 m 80 m 4
Multi-resolution approach - Level of detail ranging from the geographical scale to the object scale - For each scale the more suitable survey technology was adopted - The resulting geometric resolution varied from 25 cm to 0.5 mm 5
Low-resolution (geographic area framing) Digital surface model (DSM) of about 1 square km around the forum: existing aerial images (1989) image scale: 1:3500 (footprint: 7 cm) ETH multi-photo matcher (SAT-PP) DSM geometric resolution: 25 cm Image registration and georeferencing with GPS campaign Available topographic network Texture mapping of the DSM Pictometry images 15 cm texture resolution 6
Middle-resolution (large or simple structures) 10 D a y s o f a c q u i s i t i o n Leica HDS 3000 laser scanner for long-range acquisitions (3D framing of the forum) Geometric resolution: 5-20 mm Leica HDS 6000 for fast and massive acquisitions REVERSE (3D MODELING acquisition of areas with many occlusions) Geometric resolution: 5-10 mm Close range photogrammetry with digital reflex cameras and interactive modeling Geometric resolution: varying 7
High-resolution (details) Close range photogrammetry with digital reflex cameras and automatic matching (ETH multi-photo matcher CLORAMA) on some detailed areas Geometric resolution: down to 0.5 mm [Remondino et al., 2008 - Development and performance analysis of image matching for detailed surface reconstruction of heritage objects - IEEE Signal Processing Magazine, 25(4)] 8
Cataloguing the ruins Over 300 small finds spread all over the area Each geometrical entity (find) was identified, catalogued, photographed and coded All the modeling work has been referred to such IDs for image storing and models management 9
DATA PROCESSING RESULTS The work Resolution and Details GPS / TOTAL STATION SURVEYING AERIAL IMAGES Calibration and Orientation TOF RANGE SENSORS Calibration and Orientation TERRESTRIAL IMAGES Manual Measurement & Dense Matching Registration, Noise & Overlap Reduction Manual Measurement & Dense Matching Environment mapping Middle-res modeling Details modeling INTEGRATION & ASSEMBLING POMPEII FORUM 3D MODEL ARCHAEOLOGICAL DATABASE 10
Low-resolution: DSM from aerial images 1 km Shaded 0.8 km Color coded Texture mapped 3 existing aerial images used (low-height flight) DSM with 18 Mil points 25 cm spatial resolution 11
Middle-resolution: Laser scanning (TOF) 10 days of scanning in 2 campaigns 66 stations 2 stations georeferenced with GPS 1.2 Bil points acquired Geometric resolution: 5-20 mm Heavy hand cleaning for deleting artifacts (visitors & spurious data) Overlap reduction ICP registrations 100 Mil points used for modeling (1:10 ratio) 12
Middle resolution: Laser scanning (TOF) Subdivision of the point cloud in single sets of data for each archaeological structure (semantic classification) Range-based point cloud used for: + Alignment of the single finds modeled in higher resolution with photogrammetry + Modeling of some structures of particular archaeological interest 13
Middle resolution: Laser scanning (TOF) Range-based models of particular structures of archaeological interest Lari Pubblici Sanctuary Nerone s arch 14
Middle/High resolution: Photogrammetry with interactive modeling Over 3000 images acquired Pre-calibrated cameras Bundle block-adjustment Interactive modeling to define the main geometric entities Geometric resolution (footprint): 0.5-10 mm 3D model Photogrammetric block for a large structure (temple of Jupiter) 15
Middle/High resolution: Photogrammetry with interactive modeling Image-based models of some archaeological finds 16
Middle/High resolution: Photogrammetry with interactive modeling Image-based models of building façades Façade Wireframe Texture model 17
High resolution: Photogrammetry with automated modeling Some areas of interest were modeled with high geometric resolution by means of automated image matching ETH (Chair of Photogrammetry) multi-photo geometrically constrained matcher (CLORAMA): - combination of different image features (grid points, interest points, edges) - matching of features simultaneously in multiple images (>2) - cross-correlation as first approximation, least squares matching for precise measurements - image pyramids and epipolar constraints [Remondino et al., 2008 - Development and performance analysis of image matching for detailed surface reconstruction of heritage objects - IEEE Signal Processing Magazine, 25(4)] 18
High resolution: Photogrammetry with automated modeling Example: Eumachia door bas-relief: 13 m long Ca 100 images 0.2 mm footprint, 0.5 mm surface model 19
High resolution: Photogrammetry with automated modeling Other examples (Zeus face and altar s bas-relief) 20
Geometry vs. texture: models comparisons From laser scanning From interactive photogrammetry 21
Data integration and visualization Laser scanner final point cloud (ca 100 Mil points) georeferenced with GPS Alignment of each photogrammetric model (> 400) into the range-based model (similarity transformation) Rendering and Visualization in Cinema4D and Google Earth 22
Forum s 3D model 23
Forum s 3D model 24
Forum s 3D model 25
Forum s 3D model Over 3000 terrestrial images Team of 8 people 1.2 Bil range points 1 year of data processing 10 days of scanning Geometric resolution from 25 cm to 0.5 mm 26
Forum s 3D model Final model with 610K polygons Offline visualization in Cinema4D (LOD) Interactive visualization on a stereo-display in a virtual theater Linking with Superintendence DB still to be done video 27
Used technologies and multi-resolution data Aerial images Range sensors Terrestrial images Sensors Use Quantity Zeiss RMK A 30/23 DSM of the site at low resolution 3 (scale 1:3500) Geometric resolution Texture resolution 25 cm 5 cm Pictometry Texturing 4-15 cm Leica HDS3000 Leica HDS6000 Canon 10D (24 mm lens, 6 Mpixel) Canon 20D (20 mm lens, 8Mpixel) Kodak DCS Pro (50 mm lens, 12 Mpixel) Nikon D300 (20 mm lens, 12 MPixel) Modeling of entire Forum at middle resolution Modeling of small finds, mural architectural structures, ornaments 21 scans (400 Mil pts) 45 scans (800 Mil pts) 5-20 mm - 5-10 mm - > 3000 0.5-10 mm 0.2-5 mm 28
Conclusions A model of the Pompeii forum (ca 150 x 80 m) has been created from real data (reality-based model, not procedural modeling!) An intensive integration of measurement techniques (photogrammetry and laser scanning) has been adopted, for optimizing the amount of geometrical data Seamless integration of multi-resolution data (geometry and texture) The modeling process allowed to develop each model separately, using the global range-based point-cloud as reference The final model is suitable for real-time rendering in virtual environments 29
Thank you for your attention Fabio Remondino - fabior@ethz.ch 30