SmartFace Sentinel

Project Overview

SmartFace Sentinel is a state-of-the-art face recognition system that leverages GPU acceleration, smart caching, and real-time processing capabilities. The system processes real-time video feeds from IP cameras, performs face detection, extracts facial embeddings, and matches them against a database with exceptional efficiency.

Built with a sophisticated Flask-based server architecture, the system handles face recognition, database management, caching, and provides comprehensive API endpoints for seamless integration. The modular design separates client-side face detection from server-side recognition, enabling distributed deployment and easy integration with existing systems.

Key Features & Capabilities

• GPU-accelerated face detection and embedding extraction using CUDA, PyTorch, and FAISS
• SmartPersonCache with LRU eviction policy and similarity-based matching for intelligent caching
• Real-time video processing from multiple IP cameras with frontal face detection filtering
• Dynamic Parquet data management with automatic file monitoring and real-time updates
• Thread pool and queue-based processing for concurrent high-priority recognition tasks
• Comprehensive RESTful API endpoints for recognition, database management, and monitoring
• Multi-embedding storage per person with cosine similarity matching (50% threshold)
• Robust error handling with fallback mechanisms and graceful shutdown capabilities

Technical Implementation

The system integrates InsightFace's buffalo_l model for high-performance face detection and 512-dimensional embedding extraction. Client-side processing uses YOLO for object detection and MediaPipe for facial landmark analysis, ensuring only frontal faces are processed for optimal recognition accuracy.

FAISS (Facebook AI Similarity Search) provides GPU-accelerated similarity searches with GpuIndexFlatIP indexing for rapid nearest-neighbor searches across large datasets. The system implements a configurable GPU memory pool (16GB limit) to prevent memory fragmentation and ensure stable performance under high load conditions.

Results & Performance Metrics

The SmartPersonCache reduces database queries by up to 50% through intelligent caching with a 1-hour duration and 10,000 entry capacity. The system achieves sub-second response times for face recognition tasks, even with large databases, while maintaining thread pool efficiency with 75% CPU core utilization.

Real-time monitoring dashboards provide comprehensive insights into system health, including requests per second, cache hit rates, GPU memory usage, and processing times. The system demonstrates exceptional reliability with comprehensive error handling and automatic fallback to NumPy-based similarity search when needed.