Shan, J., & Ural, S. (2015). Performance measure that indicates geometry sufficiency of state highways: Volume II—Clear zones and cross-section information extraction (Joint Transportation Research Program Publication No. FHWA/IN/JTRP-2015/07). West Lafayette, IN: Purdue University. http://dx.doi.org/10.5703/1288284315529
Evaluation method employed for the proposed corridor projects by Indiana Department of Transportation (INDOT) consider road geometry improvements by a generalized categorization. A new method which considers the change in geometry improvements requires additional information regarding cross section elements. Part of this information is readily available but some information like the embankment slopes and obstructions near traveled way needs to be acquired. This study investigates available data sources and methods to obtain cross-section and clear zone information in a feasible way for this purpose. We have employed color infrared (CIR) orthophotos, LiDAR point clouds, digital elevation and surface models for the extraction of the paved surface, average grade, embankment slopes, and obstructions near the traveled way like trees and man-made structures. We propose a framework which first performs a support vector machine (SVM) classification of the paved surface, then determines the medial axis and reconstructs the paved surface. Once the paved surface is obtained, the clear zones are defined and the features within the clear zones are extracted by the classification of LiDAR point clouds.
SVM classification of the paved surface from CIR orthophotos in the study area results with a classification accuracy over 90% which suggests the suitability of high resolution CIR images for the classification of paved surface via SVM. A total of 21.3 miles of relevant road network has been extracted. This corresponds to approximately 90% of the actual road network due to missing parts in the paved surface classification results and parts which were removed during cleaning, simplification and generalization process. Branches due to connecting driveways, adjacent parking lots, etc. were also extracted together with the main road alignment as by-product. This information may also be utilized if found necessary with further effort to filter out irrelevant pieces that do not correspond to any actual branches. Based on the extracted centerline and classification results, we have estimated the paved surface as observed on the orthophotos. Based on the estimated paved surface centerline and width, we have generated cross section lines and calculated the side slopes. We have extracted the buildings and trees within the clear-zones that are also defined based on the reconstruction of the paved surface. Among 86 objects detected as buildings, 14% were false positives due to confusion with bridges or trees which present planar structure.
road extraction, LiDAR Point Cloud classification, building extraction
Joint Transportation Research Program
Indiana Department of Transportation
West Lafayette, Indiana
Date of this Version