Computer vision based surveillance concept for airport ramp operations

Current research develops a vision-based surveillance system concept suitable for airport ramp area operations. The surveillance approach consists of computer vision algorithms operating on video streams from surveillance cameras for detecting aircraft in images and localizing them. Rough order of magnitude estimates of the number of cameras required to cover the ramp area at a sample airport (Dallas/Fort Worth International Airport) were obtained. Two sets of algorithms with complimentary features were developed to detect an aircraft in a given image. The first set of algorithms was based on background subtraction, a popular computer-vision approach, for change detection in video streams. The second set was a supervised-learning approach based on a model learned from a database of images. The Histogram of Oriented Gradient (HOG) feature was used for classification with Support Vector Machines (SVMs). Then, an algorithm for matching aircraft in two different images was developed based on an approximate aircraft localization algorithm. Finally, stereo-vision algorithms were used for 3D-localization of the aircraft. A 1:400 scale model of a realistic airport consisting of a terminal building, jet bridges, ground marking, aircraft, and ground vehicles was used for testing the various algorithms. Aircraft detection was demonstrated using static and moving aircraft images, single and multiple aircraft images, and occluded aircraft images. Preliminary testing using the in-house setup demonstrated 3D localization accuracy of up to 30 ft.