A generalized multi-sensor 3D image registration and data fusion method using a multi-resolution approach

Recently surveillance and Automatic Target Recognition (ATR) applications have been increasing as the cost of computing power, needed to process the massive amount of information, keeps falling. Designing and implementing state-of-the-art electro-optical imaging systems to provide advanced surveillance capabilities involves integration of several technologies (i.e. precise optics, cameras, and image-computer vision algorithms for data fusion) into a programmable system. Multi-sensor fusion and integration refers to the combination of data collected from multiple sensors to provide more reliable and accurate information. Registration is the fundamental and complex process of aligning the collected data before the fusion. Several techniques for image registration have been proposed in the literature, but with limited success. In particular, one of the major limitations of existing methods is their lack of accuracy and efficiency. In addition many of these methods suffer from being applications specific. To the best of our knowledge there is no known accurate method in the literature that (a) can work under any scene circumstances/conditions and that (b) can be generalized and extended from a 2Dimensional to a 3-Dimensional space. In this research an efficient and accurate automated image registration with applications to Multi-sensor 3D LADAR imaging is presented. As we show here, the proposed approach is two-fold. First, comparison and matching of scene image/volume small patches of two overlapping 2-D or 3-D data is performed. We show here how the size of the patches is optimally derived. Second, 2D and 3-D Wavelet transforms are applied to these resulting small similar scene patches to extract a number of matching feature points. We show that the advantages of the proposed technique includes its computational efficiency, in comparison to existing methods, and its accuracy in detecting the necessary matching points, which both constitute the most fundamental/crucial but also challenging components of any data fusion/registration system. Finally, demonstration of the theories, analyses, proof of correctness behind the proposed techniques, implementation, and experimental results are presented to show the power and potential of the proposed generalized method that is extendable from 2D to 3D.