主催: 一般社団法人 日本機械学会
会議名: 日本機械学会 北陸信越支部第57期総会・講演会
開催日: 2020/03/08
Secondary accidents due to snow-removal tasks performed by elderly people have become a serious issue in snowy regions. Snow-melting techniques, such as melting machines and road heating, that do not involve the physical removal of snow are becoming increasingly important. Specifically, improving a system that can detect or predict snowfall conditions and melt snow automatically has become necessary. However, conventional automatic snow-melting system has not been efficient owing to the limitation of snowfall-sensor-installation positions. If snow-melting devices can identify appropriate positions and determine the appropriate time to melt the snow like visual judgements made by humans, the efficiency of the snow-melting device can be increased. This research was aimed at detecting snowfall conditions using images from an RGB-D camera and estimating the snow depth using a deep learning technique. We used images and class information for training data to detect objects and snow. The depth data was used to measure the snow depth. We could measure the snow depth with error of less 2cm from 4m away.