Overview

In the fall of 2018 I began a new project with Professor Jon Froehlich and the Makeability Lab, using their dataset of crowdsourced labels for sidewalk problems to develop a new system that automatically detects curb ramps, missing curb ramps, surface problems, and obstructions in sidewalks using Google Streetview imagery.

Our system achieved state-of-the-art results, and its performance detecting curb ramps rivals that of human crowd-labelers in our initial findings.

This work was published at ASSETS 2019, where it received the Best Student Paper award, with Esther Jang, Aileen Zeng, Anthony Li, Kurtis Heimerl, and Jon Froehlich.

A copy of the paper is accessible here.