Real-Time Attention for Robotic Vision

✍ Scribed by Gal Sela; Martin D. Levine

• Publisher: Elsevier Science
• Year: 1997
• Tongue: English
• Weight: 596 KB
• Volume: 3
• Category: Article
• ISSN: 1077-2014
• DOI: 10.1006/rtim.1996.0057

✦ Synopsis

Real-Time Attention for Robotic Vision

his paper discusses the development and real-time implementation of an algorithm to determine interest points in a scene. The algorithm models interest points as the points of intersection of Tl ines of symmetry between edges in an image and is based on psychophysical experiments of human gaze fixation. A novel, real-time method of computing these interest points is achieved by adopting a symmetry measure which determines the loci of centers of co-circular edges.

The algorithm has been implemented on a parallel network of Texas Instruments TMS320C40 (C40) processors. In addition, the implementation permits the algorithm to operate on non-uniformly mapped log-polar images. This configuration facilitates near real-time performance on images covering a very large field of view.

Results show that the algorithm has practical uses in a number of areas, including face recognition, object tracking and the positioning of an active vision sensor in real-time.