The detection of vehicles in airborne downward looking infrared linescan imagery

The system the authors propose uses a novel arrangement of filters (which are readily available in hardware) for detecting and segmenting potential vehicles. A set of vehicle-nonvehicle discriminating features are then extracted and used as the basis for classification. Knowledge of the platform's position, the sensor geometry and its noise characteristics are exploited to remove unnecessary dimensionality from the feature space and keep the variability of individual features to a minimum. Classifier fusion and hierarchical classification strategies have been compared with standard approaches. The performance of the system is evaluated and problem areas highlighted.