Abstract Manager

Maintaining the integrity of traffic signs is critical for ensuring road safety and effective navigation. Traditional manual inspections of traffic signs are resource-intensive, time-consuming, and prone to inconsistencies. This research addresses the need for automated, scalable, and accurate method to detect damage on traffic signs using imagery.

The proposed methodology leverages deep learning techniques to identify surface-level damage, including dirt, scratches, peeling paint, graffiti, and vandalism. Advanced 3D modeling tools were employed to simulate realistic damage patterns, ensuring the model is exposed to a wider range of scenarios during training. A comprehensive dataset of over 20,000 images, encompassing undamaged signs, actual damage scenarios, and simulated surface damage across more than 300 traffic sign classes, was assembled. To ensure the model's generalization capabilities, 65% of the test dataset includes out-of-domain images depicting real-world damage, offering a rigorous evaluation beyond controlled, simulated scenarios. The proposed model enhances state-of-the-art deep learning models, including convolutional neural networks (CNNs) and vision transformers (ViTs) by introducing domain specific filters into the approach. The proposed approach achieved a test accuracy close to 90% on real-world damaged images. Analysis of the confusion matrix highlighted challenges such as false positives caused by low-quality training images and false negatives resulting from complex out-of-domain damage types.

This research automates the traditionally manual process of assessing traffic sign conditions while addressing gaps in existing studies, particularly in integrating simulated and real-world datasets under diverse conditions. Unlike existing literature, the framework provides a custom domain-focused scalable solution for traffic sign damage detection. Future work will explore integrating image-based approaches with LiDAR data to create a comprehensive traffic sign condition assessment system, improving damage localization and enabling proactive maintenance planning to enhance road safety.