自律移動ロボットによる道なり走行のための畳み込みニューラルネットワークによる消失点検出

抄録

This paper proposes road following method by vanishing point detection using convolutional neural network (CNN). Some previous road following methods require teaching road features. On the other hand, CNN acquires them automatically and can detect objects in high accuracy and robustness. Network training data is collected in a road, and detection and road following experiment is held in another road. In vanishing point detection, road images are inspected by sliding rectangular regions (Sliding Window Detection). When vanishing point lies on centerline of rectangle region, that region has the highest vanishing point probability. Vanishing point can be detected regardless of weather. Finally, the effectiveness of the proposed method was confirmed by experiment.