Modular Processing Stages of the Pipe Machine
Abstract
This grant, for $52,000 was applied entirely for the purchase of: 4 Modular Processing Stages for PIPE (frame buffers expanded 4X-deep). The PIPE machine has played a very significant role in our research at the Laboratory for Sensory Robotics. Since November 1987 we have developed real-time PIPE algorithms for the following vision tasks: 1. basic feature extraction such as edges, zero-crossings, gradients, orientations, corners, change detection, log-polar transforms; 2. moving object centroid detection and tracking for a binocular robot eye motion system under neural control (using ADALINES); 3. measurement of visual motion (for moving edges) based on the theory of Convected Activation Profiles by Waxman et al.; image velocity fields are updated at 15 times per second; 4. stereo vision matching based on Prazdny's disparity gradient limit local support algorithm with depth maps generated once per second; 5. preliminary implementation on PIPE of perceptual grouping of features using the Neural Analog Diffusion-Enhancement Layer (NADEL) concept of Waxman & Seibert.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 30, 1988
- Accession Number
- ADA204638
Entities
People
- Allen Waxman
Organizations
- Boston University