福島第一原発での教訓を踏まえた突入撤退判断システム(第5報)

-画像とレーザスキャナを用いた走行可能領域判別手法の構築-

抄録

We develop Rush or Evacuate Judgment System referring for the lesson from Fukushima Daiichi Nuclear Power Plant. In this study, we developed obstacle detection method by deep learning for autonomous control on uneven ground. In this paper, we describe this proposed method and the result of test learning using laser-scanned point cloud images and camera images. These images divided into two data, no or movable obstacles images and immovable obstacles images. As a result, the percentage of correct answers of our classifier for self-test images was 85.04%, and it was found that learning from point cloud images has a possibility of being used for obstacle detection.