Inspection of Road Surface Damage Using Object Detection System

Abstract

This research presents a road surface condition detection system using Object Detection, with automated responses provided through a LINE chatbot. The proposed system consists of two main components.

The first component focuses on developing a model for image recognition of road surface damage conditions by comparing the performance of models trained with the YOLOv5 and YOLOv8 algorithms. The model is trained using a dataset of 790 road surface images, categorized into three types of damage: patched areas, potholes, and cracks. The best-performing model is selected for system implementation.

During the training process, it was found that YOLOv8 required less training time compared to YOLOv5. In terms of accuracy, the YOLOv8-trained model achieved a mean Average Precision (mAP) of 0.827, outperforming the YOLOv5 model, which obtained a mAP of 0.747.

The second component involves deploying the best-performing model into a LINE chatbot. Users can upload road surface images, and the system will analyze the image, returning an annotated version that highlights the damage type along with a textual response specifying the type of road surface damage. This approach helps reduce the time and personnel required for manual road surface condition inspections, enhancing the efficiency of road maintenance operations.