Sensor Fusion and Motion Energy Data Fusion for Target Classification and Discrimination

In Phase II, Computational Sensors Corp. (CSC) proposes to continue developing the Fisher Information Matrix (FIM) [1][2][3] analysis to support the development of a classifier designed to operate within the spatio-temporal (S-T) feature space provided by CSC's analog VLSI motion-energy [4] processing hardware. The FIM analysis serves as an important tool for quantifying target state estimation accuracy resulting from data fusion approaches combining isotropic and anisotropic spatial filtering, spatio-temporal (S-T) motion-energy filtering, and different sensor modalities. The extent and specifics of the data fusion and processing needed to optimize target parameter estimation must be determined from the target estimation accuracy required by the interceptor guidance law. A FIM analysis can be used to identify target parameter estimation improvements by comparing estimation accuracy based on the raw sensor data to estimation improvements resulting from data processing and data/sensor fusion approaches. In this way, the FIM analysis can be used to guide feature selection and identify a feature space which has a high sensitivity to the target parameters of interest. In previous work, CSC developed a real-time video processing system capable of agile, spatial frequency filtering using low power analog silicon retinas in a multi-chip architecture.