Estimation of road surfaces during wheelchair by using the spectrogram

Abstract

The sidewalks we use in our daily lives are made of various surface materials, such as asphalt pavement and concrete pavement. In terms of barrier-free access, Braille blocks are provided on the sidewalks. Barrier-free maps are also available to assist wheelchair users in their daily lives. The Barrier-Free Map shows the location of wheelchair-accessible restrooms and the status of barrier-free facilities at stations and stores. However, it does not indicate the type of road surface, deterioration conditions and other information necessary for wheelchair users to travel. Wheelchair users may experience problems in riding and driving due to uneven road surfaces, bumps, and joints between blocks. However, it is difficult to select a comfortable travel route in advance because the road surface condition cannot be determined without visiting the road. In addition, wheelchair users have the problem of not knowing whether they can drive on their own or need assistance without directly checking the road surface conditions. If the road conditions can be provided to wheelchair users in advance, they can take the most appropriate route. In this study, we report on a road surface estimation method that uses the vibrations generated during wheelchair driving. The following six types of road surfaces were used in the experiment. These road surfaces were wood flooring, two types of asphalt pavement, two types of interlocking block pavement, and a wood deck floor. The wheelchair used was equipped with four vibration sensors for data collection. The vibration sensors were mounted on a frame with casters and tires on its axis. The vibration data collected in the experiment was the result of a 20 [m] straight run on each of the road surfaces. For road surface estimation, a learning model was built for deep learning using spectrum images for the vibration data. As a result, five types of road surfaces could be discriminated with an accuracy of more than 80%.