Structure from Motion
Abstract
Analysis of surface boundaries has been extended to situations in which a camera is able to actively track environmental surface points. Two problems were examined - the determination of relative depth at a boundary and the determination of the direction of motion. In both cases, the ability to actively track significantly decreases the complexity of the computations required. An analysis of the computational basis for the visual detection of moving objects has been completed. We have shown that moving object detection can exploit one or more of three general approaches. Each has particular strengths and weakness. Two significant results have been obtained in the area of motion-based segmentation. The first combines motion and contrast information in a boundary detection method that is both more reliable and more accurate than possible using only motion or only contrast. The integration is done in a manner involving little additional computation. Secondly, we have shown how motion information can be used to reduce ambiguity in the recognition of partially occluded objects. Keywords: Robotics; Image understanding; Time-varying image analysis, Visual motion; Optical flow; Segmentation; Computer vision; Artificial intelligence; Space perception.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 17, 1988
- Accession Number
- ADA217577
Entities
People
- William B. Thompson
Organizations
- University of Minnesota