Traffic Sign Detection & Recognition — YOLOv8

A YOLOv8 object detection model fine-tuned via transfer learning for real-time traffic sign detection and classification. Trained as part of a BS Software Engineering thesis project and subsequently deployed as a public web application.

Live Demo

🚦 Try it on Hugging Face Spaces — supports image upload, video processing, and live webcam capture.

Model Details

Property	Value
Base model	YOLOv8 (Ultralytics)
Training method	Transfer learning
Task	Object detection
Input	RGB image or video frame
Output	Bounding boxes + class labels + confidence scores

Classes (21 traffic sign categories)

ID	Label	ID	Label
0	bus_stop	11	ped_crossing
1	do_not_enter	12	ped_zebra_cross
2	do_not_stop	13	railway_crossing
3	do_not_turn_l	14	red_light
4	do_not_turn_r	15	stop
5	do_not_u_turn	16	t_intersection_l
6	enter_left_lane	17	traffic_light
7	green_light	18	u_turn
8	left_right_lane	19	warning
9	no_parking	20	yellow_light
10	parking

Usage

from ultralytics import YOLO
import torch

# Allow loading of custom YOLOv8 checkpoint
_orig = torch.load
def _safe_load(*args, **kwargs):
    kwargs["weights_only"] = False
    return _orig(*args, **kwargs)
torch.load = _safe_load

# Load model
model = YOLO("best_roboflow.pt")

# Run inference
results = model.predict("your_image.jpg", conf=0.4)
results[0].show()

Use Cases

Driver-assistance systems (ADAS): real-time alerting for speed limits, stop signs, and warning signs
Dashcam footage auditing: automatically flagging missed or obscured signage for road-safety compliance reviews
Autonomous vehicle perception: as a building-block detection module feeding into higher-level driving decision systems

Training Details

Dataset sourced and annotated via Roboflow
Fine-tuned from YOLOv8 pretrained weights via transfer learning
21 traffic sign classes covering common urban road signage

Deployment

This model is deployed as a Streamlit web application on Hugging Face Spaces. See the traffic-sign-detector Space for the full application code and deployment details.

Author

Mian Yahya Gul LinkedIn · GitHub

Downloads last month: 52