Showing papers on "Aerial image published in 1982"

Reasoning about success and failure in aerial image understanding

Abstract: Why is an Image Understanding task, such as locating houses in aerial photographs, difficult to do automatically by computer? Three sources of difficulties are examined in this thesis, and a single approach is applied to each: that of reasoning about success and failure. The first problem is the frame problem: to choose the right domain model for a particular image. The second problem is the segmentation technique problem: to choose the right segmentation technique. The third problem is the parameter problem: to choose the correct parameter value for a parameter like a threshold. This thesis presents a unifying approach to these three problems: that of reasoning about success and failure. Reasoning about success and failure necessates an explicit goal representation, and evaluation technique for measuring performance, and search and reasoning techniques for trying to improve performance. This approach is tested in a three level Image Reasoning Program, which has been tried in the domain aerial photographs of suburban housing developments. An Appearance Model Expert examines an Appearance Model and decides what to look for using domain knowledge and the structures located so far. The Operator Expert responds to a query from the AME for a region with certain properties, and chooses an appropriate segmentation technique. Finally, Adaptive Operators use knowledge to adapt low-level parameters to find regions in the image that match particular high-level features. Each component reasons about success and failure using implicit knowledge and evaluations of performance derived from explicit goals and explicit image data. This kind of reasoning enables the Image Reasoning Program to locate objects it would not have otherwise found, improving performance. A more general conclusion is that adaptive techniques are required if high-level knowledge is to appropriately interact with computations at all levels of an Image Understanding system.

Registration of Digitized Aerial Photography with a Digital Map Data Base.

Abstract: : This report investigates the feasibility of automatic registration of a digital aerial image with an existing map data bank. The geometry of an aerial image can be modelled mathematically by a central perspective. The theoretical part presents the background of this model and discusses various strategies to solve the problem of resection in space. We propose a method to automatically recognize areal features of the map in the image and to extract control points for the reconstruction of the imaging configuration. Recognition is done by three algorithms: by correlation, by thresholding and by adaption of a binary mask. Results give estimates for the distance of projected features from actual objects, dislocations caused by variation of the transformation parameters and include experiments performed with an actual aerial photograph. (Author)