Heli-Tele
Goal of this project is the fine-alignment of helicopter-borne aerial video images to a street map. Since only highways and major roads are present in the street map, the alignment process requires the extraction of such highways and major roads from the aerial images. Although GPS information is available with the data, it is often noisy with an offset almost 20 meters error.
Background & Objective: Unsupervised extraction of roads eliminates the need for human operators to perform the time consuming and expensive process of mapping roads from aerial imagery. The road/map alignment has significant applications in transportation. It is used in creating, maintaining, and updating transportation network databases for many different purposes such as infrastructure management, traffic safety analysis, and traveler guidance. Computationally fast algorithms are required to process the increasing volume of aerial imagery. The GPS information is not sufficient enough by itself for an accurate alignment. We integrate the color, texture, edge based road feature detection methods and the available GPS information to compute the alignment parameters.
Technical Discussion: We have developed fully automatic, accurate, and computationally feasible methods to find roads in high resolution aerial video data. Out of several approaches we implemented and tested color and texture-based segmentation, road model enhancement, lane detectors, parameter space transforms for line detection, supervised learning, contour fitting, local orientation histograms, etc for single image road extraction. To take the advantage of the video data, we designed vehicle change detection by video indexing and 3D depth estimation techniques. A fusion step combines road confidences computed by several weak road detectors and non-road region filters.
Publications:
Technology Area: Computer Vision
Modification Date: July 8, 2005