We value your privacy
We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept", you consent to our use of cookies. Privacy Policy.
Edge-Based Biometric Access Control
Engineered a high-performance, offline-capable face recognition system with liveness detection for edge devices. Integrated and optimized state-of-the-art lightweight models (SCRFD, GhostFaceNet) using ONNX Runtime for CPU inference, achieving millisecond-level precision. Developed a database-agnostic MVP with a FastAPI backend, FAISS vector search, and WebSocket-based real-time feedback, ensuring secure, low-latency authentication without cloud dependency.
Key Metrics
Technologies
The Challenge
ESA Automation needed to implement secure, hands-free authentication for their industrial HMI panels. The solution required high accuracy face recognition that could run entirely on-device (ARM Cortex-A53) without internet connectivity, while maintaining strict latency requirements (<100ms) and preventing spoofing attacks.
Our Solution
We engineered a complete processing pipeline from scratch, integrating existing state-of-the-art (SOTA) lightweight models to achieve commercial-grade performance on constrained edge hardware. The solution evolved from a rigorous Proof of Concept (POC) to a production-grade Minimum Viable Product (MVP) designed for diverse industrial IT infrastructures.
Key Results
Achieved 99.8% Recognition Accuracy on the LFW Benchmark
Reduced Inference Time to 85ms on Target Hardware
Successfully Blocked 100% of 2D Photo Spoofing Attacks
Deployed across 5,000+ Industrial Units Worldwide