Geo-registration is the technique of mapping the pixel co-ordinates from images to geocoordinates. Generally, this is achieved by adjusting and aligning the input images with a standard reference image. Geo-registration helps aerial systems in target detection, target tracking as well as exploration. For aerial systems, the information from the cameras may be inaccurate as the parameters of registration between them may not be known precisely. So, simultaneously reflning the registration parameters and performing geo-registration is a huge challenge. In this paper, we propose a solution based on image segmentation and image registration in order to automatically perform pixel geo-registration. The solution is generic and can be applied to any form of image-based sensing across a variety of modalities. In our approach, images are flrst segmented using the active contour methodology and geometric partial difierential equations (PDEs) based on curve and surface evolution theory. Region-based active contours are the preferred model for registration applications as they are able to utilize more image data in the simultaneous registration and segmentation process. The features extracted after the segmentation are more robust to scene changes than the traditional pixel-to-pixel image registration techniques. Since segmentation may be aided by the solution of registration and vice-versa, it is natural to couple the problems and solve them jointly. Our focus has been to combine segmentation speciflcally with image registration in a joint, simultaneous framework where both problems are solved together with continuous and constant feedback rather than solving one problem in isolation and then using the results of the flrst solution to solve the second problem. We develop an integrated iterative approach to unify the techniques of image registration and image segmentation to provide a robust solution for pixel geo-registration. The geo-registration algorithm is currently being incorporated into a simulator framework with visualization for depicting the terrain using 3D graphics.

Unmanned Aerial Vehicles (UAVs) that operate in civil airspace and ∞y autonomously must have regard for the safety of other vehicles in space the UAV must be equipped with technology for obstacle detection and avoidance. One of the problems with UAVs is the detection and avoidance of clouds and other visual obstructions. In this paper, we propose a system for automated detection of cloud edges as well as the sky regions to maintain operation in visual meteorological conditions using image processing. The solution is based on partial difierential equations (PDE) for image processing. Anisotropic difiusion is a wellknown PDE-based technique and has been widely used in image processing for denoising and segmentation. Recently, new anisotropic difiusion techniques such as anti-geometric difiusion models have been developed which are used for adaptive thresholding and denoising. Using techniques for region merging along with anti-geometric difiusion, helps to separate the scene in a few number of signiflcant regions, thus providing an integrated system for segmentation. This technique works well when there is contrast between the cloud regions and the other areas. To improve the contrast, we propose a preprocessing step of normalizing the input color image where each pixel is treated as a vector. The normalized vector is then projected on the Blue channel. The results obtained by using the modifled scene are shown in the paper. An e‐cient real-time algorithm is being developed so that the algorithm may be used with video sequences. The incorporation of the technique in the UAV simulator tool that is being developed by the co-authors is currently being pursued. I. Introduction Safety is a major concern for Unmanned Aerial Systems (UAS) that operate in civil airspace. The UAS are expected to ∞y autonomously avoiding obstacles and ∞ying through challenging atmospheric conditions. It is desirable that the UAS be equipped with means to maintain operation in visual meteorological conditions while having regard for the safety of other vehicles. One of the major problems with UAS is the detection and avoidance of clouds and other visual obstructions. In this paper, we propose a system for automated detection of cloud edges using image processing. The solution is based on using geometric partial difierential