Non-parametric local transforms for computing visual correspondence
ECCV '94 Proceedings of the third European conference on Computer Vision (Vol. II)
A Taxonomy and Evaluation of Dense Two-Frame Stereo Correspondence Algorithms
International Journal of Computer Vision
Evaluation of Memoryless Simplification
IEEE Transactions on Visualization and Computer Graphics
Multi-camera Scene Reconstruction via Graph Cuts
ECCV '02 Proceedings of the 7th European Conference on Computer Vision-Part III
Visual User Interface for Spatial Databases Combined with Landscape Images and Sensors
DANTE '99 Proceedings of the 1999 International Symposium on Database Applications in Non-Traditional Environments
Computers and Electronics in Agriculture
Human detection for a robot tractor using omni-directional stereo vision
Computers and Electronics in Agriculture
| Hi-index | 0.00 |
This paper presents an integrated virtual reality geographic information system (VRGIS) based on computational stereo. In this system, three-dimensional (3D) and 360^o panoramic scenes are reconstructed from multiple stereo image pairs acquired with a mobile stereo machine vision system. The spatial coordinates of panoramic scenes are registered to electronic maps to obtain a real scale VRGIS that is capable of efficient retrieval of spatial information and allows for navigation in virtual reality. A software program was implemented for image processing, coordinate transformation, 3D surface model building, data fusion, digital mapping, and database management for the VRGIS. In setting up 3D models for the virtual reality environment, we focused on comparisons of pattern matching algorithms for creating disparity images from stereo pairs, and adopted the graph cut method and the non-parametric local transform method in our system. Following the computation of disparity images, they are transformed into cylindrical coordinates and stitched into a 3D and 360^o panoramic scene. In order to allow for virtual reality navigation, the panoramic scene is further cast into 3D triangular surface model for fast texture mapping using memoryless simplification algorithm. The effects of reducing the number of edges in a 3D surface model on computation time and discrepancy from 3D surface model with no edge reduction were also analyzed. Experimental results show that the discrepancy was about 13% when 80% edge reduction was applied to create the approximate 3D surface model. This indicates a significant computational and memory reduction while the computer graphic display of the scene model was still visually satisfactory. Finally, multiple 3D scenes were integrated into a site on the electronic map permitting the user to navigate through location to location in a virtual reality environment. The developed system may serve as a versatile platform in establishing field information for many applications such as precision agriculture, vehicle automation, geographic visualization, archaeology site modelling, etc.