Estimation of Road Snow Accumulation Using Smartphones to Create Snow Cover Maps for Pedestrians

摘要

In this paper, we develop and evaluate a method of estimating the snow accumulation of the road on which the user is walking via machine learning based on the values observed from the acceleration sensor and the gyro sensor mounted in the smartphone. Via method, it is possible to collect sensor values from each pedestrian and estimate the amount of snow on the road surface. By accumulating the estimated data, it is possible to realize a snow cover map and recommend an easy walking route. In the proposed method, we estimated four kinds of roads: ordinary roads, snowy roads, roads stamped to resemble ice burn, and roads on which snow had accumulated considerably. As a result of verification with ten-fold CV using only self-data for each of eight subjects, we achieved an estimation accuracy of about 100% for all subjects. On the other hand, LOSO-CV learning without including self-data showed an accuracy of 55% at best, which was greatly affected by individual differences. According to the results of this verification, the ordinary roads and ice burn are difficult to identify with acceleration sensors and gyro sensors. In addition, in the estimation of three kinds of roads, excluding ice burn, this method realized an accuracy of about 70%.