A Contrast Metric for 3-D Vehicles in Natural Lighting
Abstract
Ground vehicles in natural lighting tend to have significant and systematic variation in luminance over the presented area. This arises, in large part, from the vehicle surfaces having different orientations and shadowing relative to the source of illumination and the position of the observer. These systematic differences create the appearance of a structured 3-D object. 3-D appearance is an important factor in search, figure-ground segregation and object recognition. This paper presents a contrast metric based on the 3-D structure of the vehicle, and an analysis of search performance for the Search_2 imagery. The analysis employs the traditional P-infinity-times-negative-exponential model of search time distribution. P-infinity and mean search time are modeled as functions of the target signature. The signature metric is one over the product of vehicle size and contrast. The value of the metric is measured by the ability to account for variance in observed search performance. The 3-D structure contrast metric performs better than RSS contrast, and both perform dramatically better than the area- weighted average contrast. Target height performs better than either target area or square root of area. The signature metric accounts for over 80% of the variance in probability of detection and 75% of the variance in search time as measured in the TNO perception tests. When false alarm effects are discounted, the metric accounts for 89% of the variance in probability of detection and 95% of the variance in search time. The predictive power of the signature metric when it is calibrated to half the data and evaluated against the other half, is 90% of the explanatory power.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2000
- Accession Number
- ADP010542
Entities
People
- G. Gerhart
- G. Witus