紀要論文

IGVC Auto-Nav Challengeのための全方位画像を用いた白線認識およびNavigation用ROS componentsの開発

Development of ROS navigation components using omni-directional images-based lane recognition for IGVC Auto-Nav challenge environment

jpn

21879923

https://doi.org/10.15002/00022067

法政大学大学院理工学研究科

法政大学大学院紀要. 理工学・工学研究科編

60

1

6

2019-03-31

Version of Record

This paper describes the development of Robot Operating System (ROS) component for lane recognition and navigation of IGVC Auto - Nav Challenge. To achieve a robust and stable navigation, we propose new lane and obstacle recognition algorithm based on omnidirectional camera. Employing fast NCC based piecewise template matching technique enables robust lane detection regardless of surrounding brightness changes and various lane shape including sharp curves. To recognize surrounding obstacles only vision sensor without LiDAR, we employ YOLOv3-tiny based deep learning algorithm which can avoid obstacle collision. Since ROS has capability of environmental simulation by using Gazebo, rapid prototyping is achieved by applying both simulations and actual experiments which can significantly reduce development period. The effectiveness of the newly proposed algorithm was confirmed by both offline simulation and actual outdoor experiment.

Article

AA12677220