Modeling And Visualization Of Aboriginal Rock Art in The Baiame Cave Sabry El-Hakim, Michel Picard Visual Information Technology, National Research Council Canada John Fryer School of Engineering, University of Newcastle, Australia
Where is Newcastle?
Outline Motivation and requirements. Proposed approach. Movie Some Details Results of Modeling the Baiame cave. Conclusions.
Why 3D Modeling / Digital Archiving for Cultural Heritage ? Documentation in case of loss or damage Educational resource Visualization from new viewpoints Interaction without risk of damage Virtual tourism and virtual museums
Australia’s Aboriginal Rock Art World’s oldest art Unprotected - Fast deteriorating Inaccessible – Not enjoyed by most Why Baiame?
The Baiame Cave Location
Requirements High geometric accuracy Complete detail Photo-realism Low cost Portability
Automated Image-Based Current Techniques Direct Tracing Photography Field Notes Traditional Rock Art Surveying Photogrammetry Standard Mapping Laser Scanning Automated Image-Based Emerging 2-D Only 3-D on selected points only Potential full 3-D with photo-realism
Comments and Objective The efficient integration of image-based modeling and laser scanning, each where best suited, for recording Aboriginal rock art is the objective of this research.
Technology Selection Image-Based Methods Low cost sensor Require features ** Difficult on fine details Object size independent Manageable data size Portable Laser Scanning High cost sensor ** No features required Give fine details Object size a factor Huge data size Bulky ** The problems for this application
Error Sources : Image-based, Laser-scan and Geometric Configurations (see example) Mainly B/D (base/depth) ratio B is the component parallel to the object. B/D Ratio Images 1-3 Good Images 3-2 Poor
- Compression & editing The Proposed Solution Laser Scanning 3D Geometry Capture Digital Imaging Texture Capture Surveying Reference Frame Scans Registration Noise filtering Bundle Adjustment Texture Registration With 3D Geometry Post Processing - Compression & editing - Texture matching Visualization
Rock Art in Baiame Cave - The Movie
Problems With Rock Art Modeling (Image-Based Techniques) Difficult to find ideal locations to take the images (rocks, trees, water,..etc) Insufficient features in multiple images No geometric constraints to increase automation
Problems With Art Rock (Laser Scanning) Texture / color acquisition: most scanners do not provide good solution Registration of scans and scans with texture: require features Noise level, particularly for lower cost scanners: Noise level can be the same size as many geometric details (the rocks)
Best Solution Geometry from laser scanner to get all geometric details Texture from a separate high resolution digital camera
But we must have accurate registration, even without common features or targets, otherwise: Features spanning two images may not align Geometry and color do not match
Texture Mapping Correct registration but texture color / lighting must be corrected Uncontrolled illumination
The Baiame Cave , Data Collection Riegl® Z210i Leica® Total Station + Nikon® Cool Pix digital camera
Surveying Useful Points Not useful: no detectable geometry in the 3D model
Image Locations Top view Available image locations Poor geometric configurations
Commercial Software Tools PolyWorks® 8 3D Modeling ShapeCapture® 5 Bundle Adjustment 3ds max® 5 Rendering Adobe Premier® 6 Movie Editing
Accuracy Numbers Laser scanning: +/- 30 mm Surveying: +/- 2-3 mm Bundle Adjustment: X: +/- 13 mm, Y: +/- 9 mm, Z: +/- 11 mm
Image-Based Model
Scanner Model
Comments on the Results The results of the laser scanner were noisy and did not capture all fine geometric details. This caused problems with registration, particularly since the data had to be smoothed to reduce the noise. The image-based model was not detailed enough and resulted in unrealistic 3D experience.
Conclusions Our approach uses some surveying, laser scanning, digital images with a bundle adjustment to register the texture with geometry without targeted points. The resulting model is realistic and was created in a short time at low cost. It is still not clear what accuracy is really needed. Although rocks do not need to be modeled with high accuracy, poor accuracy will result in visible registration errors.