Vertex Space Analysis for Model-Based Target Recognition.
Abstract
We have developed powerful ATR technology that effectively addresses the critical issues of battlefield operation. Inherent in our ATR approach is a treatment of partially occluded, degraded target signatures, clutter and sensitivity to distinguish similar targets. We achieve this capability through judicious choice of features and feature attributes, local feature extraction techniques, and an efficient search strategy that exploits invariant feature attributes using our Vertex Space representation. We use low-level orientation processing combined with consistency reasoning to extract stable features and properties, used as feature attributes. Robust recognition is accomplished by matching sensor features with like model features through the rich hypothesis prediction and verification paradigm of Model Based ATR. We achieve tolerance to clutter by using an efficient search strategy performed in our unique invariant representation, Vertex Space, that reduces both the dimensionality and size of the required search space. Vertex Space mapping results in a reduced representation that serves as a characteristic target signature which is invariant to four of the six viewing geometry degrees of freedom. By conducting the initial search in Vertex Space, the search process is significantly simplified. We have demonstrated model to image feature matching using real IR imagery and models of actual military targets.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1996
- Accession Number
- ADA311784
Entities
People
- Azriel Rosenfeld
- Gary Whitten
Organizations
- University of Maryland