An algorithm for pose estimation of vehicles in synthetic aperture radar imagery is presented. The algorithm utilizes robust features and a structured decision process to achieve high precision for pose angle estimation. It is shown that the pose angle could theoretically be recovered from two simple features: the fill ratio and aspect ratio of the segmented target; and this analysis leads to useful nonlinear feature transformations. Four neural networks are used to make estimates conditional on angular region, and then a general neural network makes the best estimate with the feature set including the regional estimates. The algorithm is demonstrated for MSTAR (Moving and Stationary Target Acquisition and Recognition) data. It is relatively robust to scale variations, and significantly more precise than previously published results.
A new algorithm for pose estimation of vehicles in SAR imagery is presented. Using robust features and a
structured decision process, the algorithm achieves high precision. Four neural networks are used to make
estimates conditional on angular regions, and another neural network is used to fuse these estimates. For
the MSTAR Test Sample, the absolute error has a mean of 2 degrees with a standard deviation of 2.1,
which is significantly more precise than previously reported results.
Large cortical models based on differential equations may require significant computations to converge, in addition to
the computations required to simulate learning. Fortunately, sensitivity analysis for such models can be done using the
implicit function theorem (IFT), as shown by McFadden in 1993 for a model with "virtual lateral inhibition" (VLI) in
which inhibition is based on competition for activation, rather than on direct reduction of activation levels. The current
work reviews recent neurobiological work on the nature of inhibition, and also reports new results on numerical issues
that arise in the analysis of VLI models of cortical networks. The IFT technique is at least an order of magnitude faster
than numerical ODE solvers. A new explanation for inhibition based on energy resource sharing is proposed.
Inverse Synthetic Aperture Radar (ISAR) imagery provides an opportunity for 3D reconstruction, because it relies on target motion to provide cross-range resolution and is derived as a temporal sequence. As it moves, the target presents different aspects, which can be integrated to derive the third dimension. Tomasi and Kanade introduced a robust technique for recovering object shape and motion, based on factorization of a matrix that represents the 2D projection equations for a set of points on the target object, as observed in an image sequence. The technique has been applied to orthographic projection Tomasi and Kanade, paraperspective projection Poelman and Kanade, and perspective projection Han and Kanade, but encounters nonlinearities when applied to point perspective projection, which require iterative solution. ISAR projection is naturally well suited for application of the factorization technique because the projection equations are linear. 3D reconstruction may lead to improved performance for automatic target recognition (ATR) procedures and may also be used to enhance human visualization of iamged targets.
Separable filters, because they are specified separately in each dimension, require less memory space and present opportunities for faster computation. Mahalanobis and Kumar1 presented a method for deriving separable correlation filters, but the filters were required to satisfy a restrictive assumption, and were thus not fully optimized. In this work, we present a general procedure for deriving separable versions of any correlation filter, using singular value decomposition (SVD), and prove that this is optimal for separable filters based on the Maximum Average Correlation Height (MACH) criterion. Further, we show that additional separable components may be used to improve the performance of the filter, with only a linear increase in computational and memory space requirements. MSTAR data is used to demonstrate the effects on sharpness of correlation peaks and locational precision, as the number of separable components is varied.