Adaptive terrain classification in field environment based on self-supervised learning

This paper focuses on terrain classification in field environment and proposes a self-supervised terrain classification method which is based on 3D laser sensor and monocular vision sensor to adapt to changes in terrain environment and external conditions. First of all, extract typical traversable areas and typical obstacle areas by analyzing range data from 3D laser sensor and project these two kinds of areas into image space to label the image data. Then extract visual feature from the corresponding image to train classifier to classify the terrain. The experiment results demonstrate that the proposed method in this paper can obtain high classification accuracy and good real-time performance.