Streamlined Road Sign Maintenance System Using AI and Computer Vision

Abstract

Road signs play a vital role in traffic management and ensuring road safety by providing guidance to the drivers and pedestrians. On the other hand, damaged road signs caused by vandalism, environmental degradation, and other causes result in possible safety hazards, disturbance in traffic order, and also have negative effects on urban aesthetic appearance. In this thesis, an attempt has been made for the design and development of a real-time AI-based system for detecting and classifying street signs into damaged or undamaged categories. Special attention is given to vandalism-related damage, particularly instances involving stickers or graffiti that obscure sign information.

It is based on the YOLOv8 framework, a state-of-the-art object detection algorithm that combines speed and accuracy. It is trained and evaluated based on a custom dataset with over 3800 images, collected from diverse sources and annotated for damage types. The proposed methodology integrates object detection and classification into a streamlined workflow, where the model will first identify street signs and then classify their condition based on visible damage indicators.

Results show the robustness of the system in detecting street signs under various conditions and in distinguishing damaged signs from undamaged ones. The system reaches high precision and recall with significant improvements over traditional computer vision techniques and earlier object detection models. Visual examples and quantitative analyses underpin its performance and practical applicability.

This research will not only contribute to the advancement of object detection and classification technologies but also help to meet an urgent need for automatic road sign maintenance. It offers a scalable and efficient solution and improve overall road safety. The proposed system has the potential to be integrated into smart city infrastructures, complementing existing traffic management and urban maintenance strategies.

Road signs play a vital role in traffic management and ensuring road safety by providing guidance to the drivers and pedestrians. On the other hand, damaged road signs caused by vandalism, environmental degradation, and other causes result in possible safety hazards, disturbance in traffic order, and also have negative effects on urban aesthetic appearance. In this thesis, an attempt has been made for the design and development of a real-time AI-based system for detecting and classifying street signs into damaged or undamaged categories. Special attention is given to vandalism-related damage, particularly instances involving stickers or graffiti that obscure sign information.

It is based on the YOLOv8 framework, a state-of-the-art object detection algorithm that combines speed and accuracy. It is trained and evaluated based on a custom dataset with over 3800 images, collected from diverse sources and annotated for damage types. The proposed methodology integrates object detection and classification into a streamlined workflow, where the model will first identify street signs and then classify their condition based on visible damage indicators.

Results show the robustness of the system in detecting street signs under various conditions and in distinguishing damaged signs from undamaged ones. The system reaches high precision and recall with significant improvements over traditional computer vision techniques and earlier object detection models. Visual examples and quantitative analyses underpin its performance and practical applicability.

This research will not only contribute to the advancement of object detection and classification technologies but also help to meet an urgent need for automatic road sign maintenance. It offers a scalable and efficient solution and improve overall road safety. The proposed system has the potential to be integrated into smart city infrastructures, complementing existing traffic management and urban maintenance strategies.