Automated Tyre Inspection Through Integrated Text Recognition and Defect Detection

Tyre inspections continue to play a significant role in both vehicle safety and maintenance, but current processes are predominantly manual, time-consuming, and vulnerable to human bias. Automating this task is difficult for two main reasons: the curved geometry of tyre sidewalls, which can make the text difficult to read, and the subtle and varied nature of tyre surface damage. This dissertation presents a novel, unified, dual-pipeline system for automating tyre identification and surface inspection using both optical character recognition-based identification and deep learning-based defect detection. The first pipeline uses geometry-guided text unwrapping to straighten the sidewall text of the tyre, extracts raw text using a cloud-based recognition engine and parses it into structural fields: manufacturer, model, size and Department of Transportation serial code. The second pipeline assesses the condition of the tyre using an EfficientNet-B0 convolutional neural network to identify tyre surface damage, achieving an accuracy of 98.83%. Both pipelines are integrated into a single architecture capable of producing machine-readable and human-readable outputs. The results demonstrate that the system can automatically capture tyre specifications and detect defects from a single image of a tyre sidewall, providing a scalable and objective alternative to manual inspection. These results also provide insight into how AI-based tyre analysis may improve the reliability, traceability and operational efficiency of tyre-related activities across all types of fleets, workshops and service centres.