← All work
Open source · 2025
Vehicle vision system
Open-source computer vision for vehicles: custom-trained YOLOv8 models behind FastAPI for vehicle detection, licence-plate detection and damage detection — built for use cases like insurance claims.
- Make/model accuracy
- 92%
- Inference
- 40 ms / frame
Role
Author
What it is
A complete computer vision pipeline that detects, identifies and analyses vehicles in real-world images — the kind of system behind automated insurance claims, parking automation and vehicle registration.
How it works
- Three APIs, one service: vehicle detection and classification, licence-plate detection, and external damage detection (bumpers, hoods, headlights).
- Custom-trained YOLOv8 models, with PyTorch inference and OpenCV preprocessing.
- FastAPI backend with base64 image input for easy front-end integration, modular model handling, logging of inference and failures, and test scripts.
- Docker deployment.
Results
- 92% make/model recognition accuracy
- ~40 ms per frame inference
Model weights and datasets aren't included in the repo for IP reasons; the README explains how to train and plug in your own.
Computer visionYOLOv8FastAPIPyTorch
Working on something similar?
I'm open to senior and staff roles and to select advisory engagements.