An efficient multi-stage hyper-spectral aerial image registration technique in the presence of differential spatial and temporal sensor uncertainty with application to large critical infrastructures & key resources (CIKR) surveillance

Wide area surveillance and reconnaissance in many homeland security applications generally entails an airborne platform providing a desired view of the area of interest. The surveillance needs of large critical infrastructures and key resources that may span for miles, demand a field of view that exceeds that provided by a single airborne sensor. Image registration of aerial imagery can increase the field of view while providing a panoramic view. This increases the field of view without any modification to the optics, focal plane array, and associated hardware. Some of the primary limitations of existing image registration methods are generally (a) image and sensor noise, (b) lack of accuracy in the related correspondence points derivation, and (c) registration algorithms´ computational complexity. In this work, we focus on a multi-stage image registration technique utilizing hybrid similarity measures for partially overlapped aerial imagery, in the presence of sensor uncertainty and noise. We show that the presented algorithm provides a reduction in search space, reducing the computational cost. Additionally, here the system noise is modelled and injected into the images to determine the algorithmic performance as a function of signal to noise ratio (SNR). The presented technique is also evaluated in a scenario where a hyper-spectral or multispectral system may have an image in a noisy part of the spectrum but can be registered by using registration parameters from another part of the spectrum.