Computer Vision & Edge Biometric Verification
High-speed facial landmark detection and anti-spoofing liveness verification for academic institutions.
System Architecture & Overview
An edge-accelerated computer vision attendance system leveraging lightweight deep neural networks with 3D depth-texture anti-spoofing verification, achieving 99.4% accuracy under varying lighting conditions.
The Challenge / Problem
Manual attendance taking in university halls wastes up to 15 minutes per lecture and is vulnerable to proxy signing.
Engineering Solution
Architected a dual-stream facial recognition camera pipeline running locally on edge hardware with encrypted cloud synchronization.
Technical Hardening & Challenges
Edge hardware optimization, inference speed under 40ms, and robust anti-spoofing against photo and video replays.
Commercial & Production Result
99.4% verification accuracy, 35ms face detection speed, zero successful spoof attempts in trials.
Pipeline & Architecture Flow
OpenCV Stream -> Face Landmark Extractor -> Liveness Classifier -> Vector Embedding Comparison -> SQLite/Cloud Sync.
Applied AI & Engineering Capabilities
System Interface & Analytics
