A Neural Network System Based on Lie Group Theory for Multiple Object Recognition and Tracking

Period of Performance: 10/01/1988 - 10/01/1990

$499K

Phase 2 SBIR

Recipient Firm

LNK Corp., Inc.
6811 Kenilworth Avenue, Suite 306
Riverdale, MD 20737
Principal Investigator

Abstract

A NEUTRAL NETWORK SYSTEM IS BEING DESIGNED AND DEVELOPED CAPABLE OF PERFORMING BASIC TIME-VARYING, THREE-DIMENSIONAL (3D) SCENE ANALYSIS AND MULTIPLE OBJECT RECOGNITION/DISCRIMINATION AND TRACKING. RESEARCH IN THE PREVIOUS PHASE HAS DEMONSTRATED THAT THE CAPABILITY FOR ADDRESSING "UNIVERSALS", WHICH ARE INVARIANT UNDER DESIGNATED TRANSFORMATION GROUPS, CAN BE ACHIEVED BY A PREWIRED NEURAL NETWORK DESIGN BASED ON LIFE GROUP THEORY. THE LIE-WIRING NETWORK IS A GENERAL PURPOSE IMAGE TRANSFORMER ACCORDING TO 3D RIGID MOTION. BASED ON THE FINDINGS OF THAT RESEARCH, SIMULATORS OF VARIOUS NEURAL PROCESSES ARE BEING BUILT THAT WOULD BE INVOLVED IN TIME VARYING 3D DYNAMIC SCENE ANALYSIS, SUCH AS MOTION INVARIANT FEATURE DETECTION, SHAPE CONGRUENCE VERIFICATION, SEGMENTATION BASED ON THE COMMON FATE LAW OF GESTALT, MOTION PARAMETER ESTIMATION, OBJECT RECOGNITION/ DISCRIMINATION AND TRACKING. THE TESTS ARE BEING CONDUCTED ON A MASSIVELY PARALLEL COMPUTER AT A REAL-LIFE PROBLEM LEVEL TO DETERMINE THE OPTIMAL PARAMETERS OF THE LIE WIRING, DYNAMIC PATTERN MATCHING, AND OTHER COMPONENTS OF THE SYSTEM. ALONG WITH EXPERIMENTAL TESTS AND THEORETICAL ANALYSES, THE NEURAL NETWORK DESIGN IS BEING REFINED THAT WILL LEND TO A LIE GROUP THEORY BASED ON NEURAL HARDWARE SYSTEM. THE LIE-WIRING NETWORK SYSTEM HAS THE POTENTIAL FOR A ADDITIONAL APPLICATIONS IN GEOMETRIC IMAGE CORRECTION, IMAGE MAPPING, AND REAL-TIME COMPUTER GRAPHICS AND DISPLAY.