Automated roadway tracking for roadway occlusions

This dissertation presents an automated roadway tracking algorithm that can be used to connect occluded roadways and performs a statistical analysis to evaluate the accuracy of newly extracted roadway centerlines. The characteristics of problems during linear feature extraction are analyzed through comprehensive comparisons of various linear feature extraction techniques and available data sources. A set of operators (i.e., watershed transform and shock graph) is used to detect roadway centerlines from high-resolution panchromatic imagery. A multi-resolution noise removal technique is used to remove non-roadway objects. The change patterns of three different types of roadway surfaces, as criteria, are determined for the automated roadway tracking. The slope variability from secondary searching trajectories is compared with the predetermined roadway surface pattern to determine roadway connections. The curvatures of roadway centerlines adjacent to starting point for the automated roadway tracking are used to connect occluded roadways. The performance of the automated roadway tracking is affected by surface model blunders, misalignment of extracted roadway centerlines, and DEM resolutions. However, the new automated roadway tracking algorithm is unique because it is less sensitive to noise. Finally, the positional accuracy and horizontal alignment of extracted-and-tracked roadway centerlines are evaluated by comparing them with true roadway centerlines. The performance evaluation results show that extracted roadway centerlines become more complete after roadway tracking.