Target-Terrain Classification and Detection Performance

Abstract

The interactive effects of target and background characteristics were investigated. Five multiple-vehicle target conditions that varied in number and configuration were embedded into oblique aerial photographs of real terrain. Target proximity to roads was manipulated, and terrain complexity was evaluated by a subjective scale of scene heterogeneity. Results indicated that performance improved as the number of vehicles comprising a target increased. An interaction between number of vehicles and proximity to roads suggested that local context is a relatively more important performance predictor when searching for single-vehicle targets than when searching for multiple-vehicle targets. The heterogeneity measure was found to be an inadequate predictor of detection performance, and some alternatives are discussed.