講演情報
[Cycle_Path___Side_Walk-08]Research on Hazard Inspection in Pedestrian Spaces Using a Mobile Robot Vehicle
*Yoshiki Kawano1, Ikuo Yamamoto1, Akihiro Morinaga1, Takeki Haraguchi1, Ikkei Nishitake1, Kazunari Hirakawa2, Quoc Thuyet Dang2 (1. Nagasaki University (Japan), 2. Taisei Rotec Corporation (Japan))
キーワード:
Pedestrian-space inspection、Vehicle-type mobile robot、Digital hazard mapping、Autonomous cruising
In Japan today, the aging of social infrastructure is progressing rapidly. Pavement deterioration causes cracks, potholes, and steps, on road, and increases the risk of falls for pedestrians and cyclists. In addition, utility poles and garbage collection stations are installed in narrow roadside strips, creating situations in which pedestrians and personal mobility users are forced to step into the roadway to avoid them, posing a risk of collision accidents. While it is necessary to inspect and improve pedestrian spaces, such as sidewalks and roadside strips, these spaces are widely distributed nationwide. Therefore, manual inspections require substantial time and cost and also face the challenge of inconsistent evaluation. The aim of this study is to digitalize pedestrian-space inspection, and the authors developed an autonomous vehicle-type mobile robot that detects hazardous locations using deep learning and converts the results into a digital map. The vehicle-type robot is autonomously driven to observe pedestrian spaces using various onboard sensors. Using semantic segmentation, the system detects and computes road surface damage, potholes, steps, utility poles, road signs, manholes, garbage collection stations, pedestrian-space width, and road gradient, and integrates these results with GPS-based position information to map them onto a digital map. As a demonstration of experiment, the robot was driven in an actual pedestrian space, and the overall flow from detection of the evaluation items to mapping was verified. As a result, it was confirmed that evaluation items such as utility poles and manholes can be detected and visualized on a digital map by combining them with GPS position information. In addition, it was confirmed that measuring pedestrian-space width using the proposed method, which combines RGB images with 3D LiDAR distance data, is effective. This study presents a framework that integrates deep-learning-based detection results with position information to convert pedestrian-space inspection into a digital map, and provides knowledge that contributes to improving the efficiency and objectivity of inspections.
