Self-positioning and feature point detection using sensor fusion in non-GNSS environments

This study presents a YOLOv8m-based object detection model for tunnel inspection, focusing on lighting fixtures and enamel panel joints. A dataset of tunnel images was prepared, annotated, and used to train the model across 150 epochs. Results showed high accuracy and repeatability with standard deviations under 0.1%. These findings indicate the model's potential for improving automated tunnel safety inspections. Future work includes adapting the model to different tunnel structures and integrating Time-of-Flight (ToF) distance imagery for dimensional analysis.