next up previous contents
Next: Extraction of Elevation Contours Up: Computer Vision Previous: Dynamic Vision for Space

Digital Libraries for Virtual Environments

F. Beyrouti, F.P. Ferrie Recent advances in laser rangefinding systems have made it possible to supplement conventional video data with accurate measurements of three-dimensional shape. This has led to a flurry of research aimed at building more complex representations for objects that are able to reflect physical attributes and properties. Applications range from the creation of so-called virtual worlds for the purposes of visualization to the representation of workspaces for autonomous robotics. While much of the research effort has been directed at computing these representations, increasing attention is now focused on the problem of their representation in digital media. The objective of this project is to investigate how this kind of multi-media can be structured in the form of digital libraries. There are two key aspects of this work. The first concerns the indexing problem. For example an object can be indexed by its shape, color, texture, attributes contained in a property list, position and orientation in a workspace, etc. For example, these attributes might correspond to the metadata relating to the visual documents. In our case the metadata are often prescribed as well as their relationships to the visual documents. The second aspect concerns how the information contained in these libraries can be dynamically updated to reflect the current state of the environment. For example a sensor monitoring the position and articulation of a manipulator moving through a workspace might be used to update the corresponding parameters in that object's description in a world model database. This requirement introduces an interesting twist on the design of the library as representations must also be adaptive. The situation gets even more complicated when queries are allowed to take place concurently with updating (synchronization problem). To provide focus, a virtual laboratory demonstration will be implemented using the digital library infrastructure. Sensor-derived object models (NCE IRIS projects MS-2 and MS-3) will be accessible to remote clients (e.g. our colleagues at Université Laval) via ATM.


next up previous contents
Next: Extraction of Elevation Contours Up: Computer Vision Previous: Dynamic Vision for Space

Thierry Baron
Mon Apr 7 12:54:24 EDT 1997