Automatic Detection of Cracks and Damage in Track Infrastructure Using YOLO and 3D Image Sensor

Rapidly increasing maintenance work due to the aging of railway infrastructure requires maintenance technologies that incorporate advanced techniques for systematic upkeep and preventive reinforcement. This study addresses the limitations of traditional in-person inspection methods for track facilities performance evaluation that consume time and manpower by developing a trolley-type laser imaging inspection device for capturing images of track infrastructure, and an AI-based analysis module for automatically detecting defects and damage in the collected images. The trolley-type device is equipped with a 3D laser image sensor, while the AI analysis module was developed using image labeling and learning through YOLO v7. The developed AI module achieved an accuracy of 91.6%, a recall of 90%, and an F1-score of 90.8%, confirming its capability to objectively and quantitatively detect major defects such as cracks and damage in concrete ties and concrete ballast.