Targeting Using Multi-Discriminant LADAR Technology (TUMULT)

While automatic target detection and recognition technologies have been developed and proven for a number of sensor modalities (e.g. electro-optical (EO), infrared, and LADAR), none exhibit satisfactory performance when dealing with objects camouflaged by dense forest canopies. The effective detection and recognition of hidden targets requires the application of multiple, co-registered sensor modalities, such as those provided by multi-spectral, multi-hit LADAR sensor systems currently in development. The modalities supported by these emerging sensor technologies provide material-dependent signatures that are well-suited to the detection of concealed targets. However, the high-dimensional feature spaces generated by such sensors increase computational complexity. Worse still, it is likely that the data does not exist in linear subspaces and will therefore defy analysis via traditional linear analysis techniques such as PCA. We propose a system called Targeting Using Multi-Discriminant LADAR Technology (TUMULT). The system derives features from input M-D LADAR data, compresses the feature space using a previously trained nonlinear dimensionality technique and finally classifies image pixels into target or background regions of interest. The capabilities resulting from this effort will significantly improve the Air Force’s ability to detect concealed threats in both future and contemporary warfare scenarios.