深層学習による点字ブロック検出を用いた自律移動ロボットの走行経路生成

抄録

Our research group is studying the method that use Braille blocks as a graphical map and as landmarks when driving for the purpose of Autonomous running of a robot in an underground passage at Shinjuku Station. In this paper, we present a method for detecting Braille blocks using deep learning, and a method for calculating data about nodes and edges from the detection results. In addition, we evaluated whether the angle of the robot to the guiding blocks could be calculated. The evaluation was conducted by setting up a cart equipped with an RGB camera against the induction block and determining whether the angle between the induction block and the cart could be calculated. As a result, when the induction block and the cart were not parallel, the error in the calculated angle was small. The reason for the large error when the rows of induction blocks and the cart are parallel may be that the method of correcting the coordinates is wrong.