I designed and led the development of an automated face-blurring pipeline for Boston Dynamics, designed to anonymize facial data inadvertently captured by the Spot robot's camera. The goal was to ensure GDPR compliance and protect privacy while processing data efficiently.

Technology Stack: Google Cloud Platform (GCP), Vertex AI, YOLOv5, Kubernetes, Cloud Functions, Terraform.

Functionality: The system involved a mesh of Google Cloud services to detect and blur faces in images captured by the Spot robot. Key steps included:

Data Ingestion: Images and associated data were ingested and stored in Google Cloud Storage.

Face Detection and Blurring: Using a custom YOLOv5 model deployed on Vertex AI, faces were detected and coordinates were sent to a cloud function that applied blurring.

Repacking and Export: The processed images were then repacked and stored back in the cloud, ready for further use or compliance verification.

Performance: The pipeline was capable of handling up to 2 million images per hour, ensuring high scalability and efficiency.

Benefits: This automated pipeline ensured that any facial data captured by the Spot robot was anonymized before being transmitted or used, complying with GDPR requirements and protecting individual privacy.