Visualizing Road Condition Information by Applying the AutoEncoder to Wheelchair Sensing Data for Road Barrier Assessment

Abstract

Providing accessibility information about sidewalks for people with difficulties with moving is an important social issue. Conventional methods of collecting large-scale accessibility information are based on manpower and have the problem that large-scale information collection is difficult because of the time and money costs. We previously proposed and implemented a system for estimating road surface conditions by using machine learning with measured values of an acceleration sensor attached to a wheelchair. In this paper, we examine the appropriateness of reconstruction errors which are calculated by Convolutional Variational AutoEncoder as the degree of road burden suitable for each user. This paper calculated reconstruction errors from the traveling data of 14 wheelchair users and created a map reflecting the information of the errors. As a result of the evaluation, it was suggested that the reconstruction errors can reflect the degree of the burden on each wheelchair user during traveling.