Multisensor Tracking and Recognition of Animate and Inanimate Objects
Abstract
We report on final results arising from this project, in which we proposed to establish a new paradigm for multisensor tracking and recognition of animate and inanimate objects, fusing a model-based methodology with a neural network-based methodology in an integrated and synergistic manner. Important results are report for the four major project areas: (1) Hybrid ATR systems, (2) Human Motion; (3) Multiple Feature Representation; and (4) Detection and Tracking of Moving Obstacles in the Path of a Navigating Robot. Major accomplishments include the development of: (1) a hybrid intelligent architecture that exploits the complementary nature of symbolic and connection/neural reasoning methodologies for more effective object recognition; (2) a comprehensive mathematical framework to measure the gain in classification performance when several classifiers are combined in a linear fashion; (3) the use of localized gating networks in the mixture-of-experts framework; (4) a Bayesian segmentation framework for textured visual images; (5) a multiple fixed camera system for automatic tracking of human motion in indoor environments; (6) the use of stereo fish-eye lenses for autonomous mobile robot navigation and environment mapping, and (7) an algorithm for moving obstacle detection from a navigating robot.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 31, 1998
- Accession Number
- ADA344840
Entities
People
- J. K. Aggarwal
- Joydeep Ghosh
Organizations
- University of Texas at Austin