Edge Detection for Object Recognition in Aerial Photographs

An important objective in computer vision research is the automatic understanding of aerial photographs of urban and suburban locations. Several systems have been developed to begin to recognize man-made objects in these scenes. A brief review of these systems is presented. This paper introduces the Pennsylvania Landscan recognition system. It is performing recognition of a scale model of the University of Pennsylvania campus. The LandScan recognition system uses features such as shape and height to identify objects such as sidewalks and buildings. Also, this work includes extensive study of edge detection for object recognition Two statistics, edge pixel density and average edge extent, are developed to differentiate between object border edges, texture edges and noise edges. The Quantizer Votes edge detection algorithm is developed to find high intensity, high frequency edges. Future research directions concerning recognition system development, and edge qualities and statistics are motivated by the results of this research.