Hierarchical Target Detection/Recognition for Foveal Vision

Robust and scale/rotation/aspect-invariant target detection/recognition is of great importance for image based sensing platforms to be used for guidance applications. This Small Business Innovation Research Phase I program will investigate the feasibility of a hierarchical target detection/recognition approach for robust, semi-affine-invariant target detection and recognition for hierarchical foveal machine vision (HFMV) based targeting systems. The system is organized into three hierarchies: a hierarchy of camera models for different targets at different circumstances, a hierarchy of multiresolution images (foveal images), and a hierarchy of filters to achieve automatic target detection and recognition. Targets are detected based on their appearances by maximum discrimiant filters across multiple scales. The maximum discriminant filters are derived by an integration of a limited number of target appearance templates combined with appropriate camera models through Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Templates are incorporated into Linear Phase Coefficient Composite (LPCC) filters to achieve target detection. The proposed approach promises robust target detection/recognition on individual images and image sequences as long as the change in appearance of the target can be described by affine transformations. Even though it is is developed primarily for visual and infrared (IR) HFMV imageries, the proposed approach can be used for other types of images such as images from Synthetic Aperture Radar (SAR) and Laser Radar (LADAR) sensing platforms. The proposed algorithm can be mapped into feedforward neural networks.