Tire Thread Classification

Manual tire inspection is subjective, time-consuming, and inconsistent across inspectors. For a fleet or automotive service operation handling high volumes, manual inspection creates bottlenecks and introduces human error into safety-critical decisions. There was no standardized, automated way to assess tread depth from a photograph and produce a consistent safety classification.

We trained a YOLOv8 object detection model on a labeled dataset of tire tread images, using Roboflow for dataset management, annotation, and augmentation. The model identifies tread wear indicators and classifies tires into safety categories (safe, marginal, replace immediately). The model is served via a FastAPI endpoint that accepts an image and returns the classification result with confidence scores. The Python inference pipeline is optimized for throughput and can process batches of images efficiently.

The model achieves high accuracy on the test set and is running in production, processing tire images automatically and eliminating the need for manual classification at scale. Inspection throughput increased and classification consistency improved significantly compared to manual assessment.