Embedded by design
The camera and inference run on a Raspberry Pi rather than depending on a cloud service, reducing both network dependence and exposure of raw footage.
Project 07 · Team project
Real-Time Facial Anonymization for Embedded Devices
A standalone Raspberry Pi camera pipeline that detects faces and applies Gaussian blur locally for privacy-sensitive video systems.
- Completed
- 2022
- Hardware
- Raspberry Pi 4B
- Stack
- Python · OpenCV
- Built with
- Mark Slattery
Project poster
Privacy processing happens on the camera system.
The system was designed for doorbells, security cameras, and other real-time video sources where identifiable footage should not leave the device by default.
Processing path
Detect, align, decide, and anonymize.
- Detect and trackHaar and LBP cascades locate face and eye regions in each frame.
- Align the faceEye locations help normalize partially obscured or tilted faces.
- Apply policyReference images can identify faces that should or should not be hidden.
- Blur locallyGaussian filtering removes facial detail before footage is stored or displayed.
Selective privacy
The prototype included zones and reference-image matching so anonymization behavior could change by location or recognized identity.
Project boundary
This page presents the original system and poster. It does not send webcam video to a server or claim modern biometric accuracy for the 2022 cascade-based detector.
Inspect the code ↗