traffic-sign-detection

Traffic Sign Condition Detector

A YOLOv26-based object detection model for classifying traffic signs by physical condition. Fine-tuned on street-level imagery for use in road infrastructure monitoring and mapping pipelines.

Model Details

• Architecture: YOLOv26
• Base model: Ultralytics/YOLO26
• Task: Object detection + condition classification
• License: MIT

Dataset

Trained on the traffic-sign-detection-znanc-9hhnw dataset from Roboflow. The dataset consists primarily of traffic sign images captured from a distance, representative of typical street view or dashcam footage.

Classes

good, bad, moderate

Performance

The model performs well on street view imagery where signs appear at a distance, which matches the training distribution. It is well-suited for automated road surveys, mapping applications, and infrastructure audits.

Intended Use

• Street view imagery analysis
• Road sign condition auditing
• Smart city and infrastructure monitoring pipelines

Limitations

• Optimized for signs viewed from a distance; close-up or heavily cropped sign images may produce lower confidence
• Trained on English-language traffic sign imagery; performance on other regions may vary