Specific shape building detection from aerial imagery in infrared range

This paper describes an approach to detection of specific shape buildings from the aerial imagery in the infrared range. The proposed algorithm uses contour analysis and is based on a modification of the generalizedHough transform that allows to detect curves defined by a small number of parameters. The main idea is to build atwo-dimensional accumulator array for each possible parameter set of the specified curve, and to combine obtainedarrays into the resultant accumulator array whose local maxima correspond to the positions of the sought objects.The gradient magnitudes of the original image are used to fill the arrays. The closeness of the found contoursto the predefined curve is determined by the morphological analysis of the values calculated by the Canny edgedetector. Filtering detected objects relies on the density of boundaries in their internal area with the ratio of theaverage intensity inside and outside the contour that provides high sensitivity to the specified types of objectsand reduces the number of false alarms of the algorithm. The proposed approach was tested on the problem oflocalization of rectangular buildings and showed the appropriate quality for practical use.