Project Idea: AI-Powered Multi-Camera Face Recognition & Smart Attendance System
I want to build a real-time, edge-AI based face recognition and attendance system using Axelera AI Metis and Voyager Wingman.
The main goal is to create a system that can process live RTSP streams from multiple CCTV/IP cameras, recognize registered people, track them across frames, and automatically maintain attendance records — while performing AI inference on the Axelera Metis accelerator instead of relying heavily on the host CPU.
### What I want to build
The system will have the following pipeline:
RTSP CCTV Camera
↓
Video Processing
↓
Face/Person Detection
↓
Face Recognition
↓
Object/Face Tracking
↓
Identity Matching
↓
Attendance Recording
↓
Web Dashboard
The system will initially support one camera and then be designed to scale to multiple RTSP cameras.
### Key Features
• Real-time face detection and recognition
• Multiple registered people
• Multiple reference images per person
• Face embedding database
• RTSP IP camera support
• Automatic camera reconnection
• Face/person tracking
• Recognition confidence score
• Duplicate attendance prevention
• Date and time based attendance
• Camera-wise attendance
• CSV/database attendance records
• Web dashboard for monitoring
• System and camera error logging
• 24/7 operation capability
• Scalable architecture for multiple cameras
### Why Edge AI?
Instead of sending camera footage to the cloud, I want the AI processing to happen locally using Axelera Metis.
This provides:
• Lower latency
• Better privacy
• Reduced network bandwidth
• Local processing of sensitive camera data
• More predictable real-time performance
### How I will use Voyager Wingman
Voyager Wingman will be an important part of the development process.
I will use Wingman to:
• Explore the Voyager SDK
• Identify suitable Metis-supported models
• Build and configure inference pipelines
• Create and validate pipeline YAML configurations
• Develop the Python application
• Integrate RTSP streams
• Configure AIPU resources
• Debug SDK and runtime errors
• Optimize CPU and Metis usage
• Build the multi-camera architecture
• Test and improve the complete system
Rather than manually building everything from scratch, I want to use Wingman as an AI engineering partner throughout the project and document the prompts, decisions, debugging process, and improvements.
### Hardware
Host:
• Intel Core i5-14400
• Ubuntu Linux
• Axelera AI Metis M.2 accelerator
Input:
• RTSP IP/CCTV cameras
### Development Approach
I plan to develop the project incrementally:
1. Metis hardware and Voyager SDK validation
2. Single-camera inference
3. Face detection
4. Face recognition and embedding database
5. Tracking
6. RTSP integration
7. Attendance system
8. Web dashboard
9. Multi-camera support
10. Performance and reliability optimization
The final goal is a practical edge-AI application that can be deployed in environments such as offices, factories, educational institutions, and other controlled-access areas.
### Why I think this project is useful
Face recognition systems are often built around cloud or CPU/GPU processing.
I want to demonstrate how an affordable edge-AI accelerator such as Axelera Metis can perform the complete vision workload locally while maintaining real-time performance and protecting camera data.
The project will also demonstrate how Voyager Wingman can help a developer go from an idea to a complete deployable edge-AI application.
I am especially interested in exploring how far the system can scale in terms of RTSP camera count while maintaining useful recognition accuracy and real-time performance.
I will document the development process, challenges, Wingman prompts, performance results, and final implementation so that the project can be reproduced by other developers.
