Széchenyi Plusz RRF

Automated detection of road hazards such as speed bumps, has become an important area of research due to its potential to improve road safety in autonomous driving. Various techniques have been introduced to detect these hazards using camera vision and artificial intelligence-based image processing methods. However, estimating their distance is still challenging. To address this problem and to satisfy the requirement for real-time on-board data processing, the proposed system has the following properties: (1) high-accuracy road hazard detection by analyzing mono-images and videos with a re-trained YOLO neural network; (2) precise distance measurement utilizing a LiDAR; and (3) efficient local data processing using ROS, implemented on an NVIDIA Jetson AGX Xavier. An important contribution of this paper is introducing multiple classes of road hazards when training the network, instead of only focusing on speed bumps and potholes. Furthermore we have analyzed different LiDAR technologies (standard rotating and non-repetitive circular scanning) to evaluate and compare their precision and to demonstrate that our method can be successfully applied regardless of the scanning pattern of the LiDAR.