Car Detection AI - YOLO26 Object Detection

🎯 Try It Now

⚠️ Coming Soon: Backend API integration. For now, this is a visual demo of the interface. The actual detection will be live in the next update.

Upload an Image

Drop a JPG image or click to select

Detection Results

📖 About This Project

This is a demonstration of a custom-trained YOLO26 object detection model built to identify cars and number plates in images. The model was trained on 500+ hill climb and street car photographs captured at various UK motorsport events.

Why YOLO26? It's the latest version of the YOLO series, offering:

✅ End-to-end NMS-free inference (faster deployment)
✅ Up to 43% faster CPU inference than v11
✅ Better accuracy on small objects
✅ Edge-optimized for real-world deployment

🏗️ Architecture & Learning

This project demonstrates a complete MLOps pipeline:

1. Data Labelling

500 images labelled in Label Studio with bounding boxes for cars and number plates.

2. Model Training

YOLO26 fine-tuned on custom dataset. Achieved strong accuracy on diverse car photos.

3. API Development

FastAPI backend serving predictions. Scalable and production-ready.

4. Containerization

Docker deployment. Easy to run anywhere: laptop, cloud, edge devices.

5. Cloud Deployment

AWS Lambda + S3 + CloudFront. Serverless, scalable, cost-effective (£2/month).

6. Frontend

Single-page web app. Users upload images, get instant detections with confidence scores.

⚙️ Technology Stack

Model YOLO26 Medium

Backend FastAPI

Deployment AWS Lambda

Hosting S3 + CloudFront

Framework PyTorch

Labelling Label Studio

Training Metrics

Dataset: 500 labelled images (hill climb + street cars)
Classes: Car, Number Plate
Training Time: ~20 minutes (M1 Pro)
Inference Speed: ~50-100ms per image
Model Size: 50MB

🔗 Links & Resources

GitHub: View the source code
Blog: Read the full writeup
Twitter: Follow for updates

Real-time Car Detection