Citation

Luo J, Wu G, Wei Z, Boriboonsomsin K, Barth M. Transp. Res. Rec. 2019; 2673(8): 499-507.

Copyright

(Copyright © 2019, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.1177/0361198119842820

PMID

unavailable

Abstract

To support active mobility, extensive work has been focused on planning, maintaining, and enhancing infrastructure, such as sidewalks. A significant amount of these efforts has to go on the setup and maintenance of sidewalk inventory on a certain geographic scale (e.g., citywide, statewide). To address the stated problem, this paper proposes the development of an aerial-image-based approach that can 1) extract the features of sidewalks based on digital vehicle road network; 2) overlay the initial sidewalk features with aerial imagery and extract aerial images around the sidewalk area; 3) apply a machine learning algorithm to classify sidewalk images into two major categories, that is, concrete surface present or sidewalks missing; and 4) construct a connected sidewalk network in a time-efficient and cost-effective manner. A deep convolutional neural network is applied to classify the extracted sidewalk images. The learning algorithm gives 97.22% total predication rate for the test set and 92.6% total predication rate in the blind test. The proposed method takes full advantage of available data sources and builds on top of the existing roadway network to digitize sidewalks.

Language: en