ARPA Image Understanding Workshop 1996 We report on the two areas of our research that are sponsored by ARPA, namely, the retrieval of images from data bases and vision for mobile robots. In image retrieval, we have developed two demos. The rst is based on texture and color information, while the second explores the use of shape information in the presence of occlusions. We plan to merge these two retrieval demos into one once the technologies involved are well understood in isolation. For our next demo, we have stored twenty-thousand images from a stock-photograph collection onto a laser disc recorder. Thumbnails and retrieval indices are stored on a computer hard disk. For robot vision, we have developed, together with Stanford's robotics group and Nomadic (a local robot manufacturer), a robot observer that uses motion planning and visibility graphs to stalk a moving target in an environment cluttered with obstacles. We have also built a depth-from-focus vision system that allows a Nomadic robot to navigate for hours in a crowded environment. The robot avoids obstacles, both static and moving, and turns away from steps, both up and down, in a very reliable fashion.
[1]
Craig Becker,et al.
An Intelligent Observer
,
1995,
ISER.
[2]
Carlo Tomasi,et al.
Obstacle Avoidance Via Depth From Focus
,
1998
.
[3]
Ramesh C. Jain.
NSF workshop on Visual Information Management Systems
,
1993,
SGMD.
[4]
Leonidas J. Guibas,et al.
Shape-based indexing and re-trieval: some first steps
,
1996
.
[5]
Tom Minka,et al.
Vision texture for annotation
,
1995,
Multimedia Systems.
[6]
Michael T. Goodrich,et al.
Geometric Pattern Matching Under Euclidean Motion
,
1993,
Comput. Geom..
[7]
Ketan Mulmuley,et al.
Computational geometry : an introduction through randomized algorithms
,
1993
.
[8]
Sunil Arya,et al.
Approximate nearest neighbor queries in fixed dimensions
,
1993,
SODA '93.