On the Robustness of Efficient Velocity Filter Banks to Registration Error

Detection of pointlike targets in noisy imagery is a difficult task. The velocity filter improves detection performance by integrating information across several frames. Due to absence of target knowledge a bank of Velocity Filters must be used. The required size of the filter bank grows exponentially with the number of frames, prohibiting exhaustive evaluation. This study employs pixel-first and best-first search regimes to reduce the computational demand of the filter bank. Simulations demonstrate that both approaches yield good results and exhibit robustness to registration error. The techniques are successfully applied to real imaging seeker data.