Monocular camera-based 3D point cloud reconstruction and traffic sign detection using vision transformers and YOLOv8

This study presents a novel method for extracting point clouds of traffic signs using a single monocular camera sensor. Traditional light detection and ranging (LiDAR) techniques, although highly accurate, are expensive, require integration with cameras for segmentation tasks, and increase overall system complexity. The proposed approach is significant as it enables the generation of accurately segmented point clouds without relying on a LiDAR sensor, which was not available to the research group. The solution is flexible, allowing substitution with equivalent algorithms for monocular depth estimation, image segmentation, camera calibration, and global positioning system (GPS) association. Furthermore, the integration of machine learning techniques is proposed for traffic sign classification.