Material Classification for Physics-Based Sensor Simulation Using Stereo-Pair Imagery

Current simulators can display very detailed imagery over very large geographic training areas in both visual and sensor channels. To provide a high-fidelity sensor channel, the simulation must be physics-based, and must rely on knowledge of the material properties of the simulation area. The proposed work includes the development and evaluation of algorithms for an advanced classifier and feature extractor that uses both color/multispectral imagery and digital elevation maps. It is specifically targeted to exploit the stereo imagery from the latest generation of earth imaging satellites. The extractor uses an initial multi-scale object-based process to segment the image, followed by an expert system for object classification and feature extraction. The proposed option tasks include the exploitation of height and intensity texture, and the extraction and representation of spatial variability within a material class.