Advanced Algorithms for Tomographic Imgaging

Current LIDAR-based ATR systems generally fail to exploit orientation information preserved in multi-view image datastreams. We propose to transition and adapt a novel technique that FTI developed in 1998 for Air Force high-range-resolution (HRR) radar signature classification to the LIDAR ATR problem. The approach is based on comparative analysis of the temporal LIDAR (or HRR) signatures obtained from multiple target views and on exploitation/fusion of target-related signature redundancies. We have demonstrated that the accuracy and fidelity of the associated algorithm approaches 100% Pd with 0% RFA as the number of independent profiles is increased. However, in the HRR application, the number of available source pulses was constrained severely by hardware limitations. Within the currently considered LIDAR applications, we expect to see considerably more independent LIDAR return profiles for each target; and thus we expect to be able to exhibit superior classifier performance. In Phase I, we will design and develop a software solution to optimally exploit multi-view LIDAR signatures; and adapt our algorithms for demonstration using GFI or in-house data. Phase II will develop advanced LIDAR ATR software to run on low-volume, low-power hardware (COTS components) that can support embedded real-time computation.