No abstract

on roadways. Screening out nonroad areas during image processing is desirable for two reasons: first, vehicle detection and tracking will only be performed within roadway areas, and, therefore, there will be no false detections of vehicle movements in these nonroad areas; second, the computational effort can be reduced because only roadway areas are processed. The term "road mask" describes a screen applied to an image to eliminate irrelevant areas of the image and to focus the computational effort on the roadway itself.As part of the National Consortium on Remote Sensing in Transportation-Flows, a software tool called "tracking and registration of airborne video image sequences" (TRAVIS) has been developed to extract vehicle positions from airborne imagery to assist in the analysis of microscopic traffic behaviors (1, 2). The input into TRAVIS is a sequence of images from the video. TRAVIS registers the image sequences to an initial common reference frame, detects the vehicles in the images, and tracks the vehicles through the image sequence. The output of TRAVIS is a sequence of vehicle pixel coordinates as they are tracked through the image sequence (3). The research here describes a method to generate a road mask from basic vehicle movements in airborne imagery, with the ultimate goal of improving the detection and tracking of vehicles. This is part of a continuing effort to enhance the software. Literature reviewMany algorithms have been developed to extract roadway boundaries for various purposes. Most of them fall into one of three categories: the Hough transformation-based approaches (4, 5), the deformable template-based approaches (6-9), and the line segment groupingbased approaches (10). For example, Bandera et al. use a mean shift-based clustering in the Hough domain to detect line segments in edge images (4). A line segment random window randomized Hough transformation is used to find favorable line segments. Then, a variable bandwidth mean shift algorithm is applied to cluster items in the Hough parameter space. The performance of the algorithm is good in terms of the computation time and the ability to detect line segments. However, there are a set of parameters to calibrate; for example, the minimal line segment length and the minimal gap between two line segments.In contrast, Kluge and Lakshmanan describe a likelihood of image shape algorithm for lane detection based on a deformable template approach, which uses intensity gradient information (6). The lane detection problem is formulated as finding a set of the most likely lane edge parameters. A shape model defines a set of shape parameters, and a likelihood function measures how well a given Estimation of a Road Mask to Improve Vehicle Detection and Tracking in Airborne Imagery Xueyan Du and Mark HickmanThis research proposes a method to estimate a road mask in airborne imagery to improve vehicle detection and tracking. The road mask can remove false detections and reduce computation time. The road mask was estimated on the basis of smoothed (i...

No abstract