-
-
Real time line graph for rolling average for last 40 frames and unknown headcount bar graph provides analytics on unusual activity.
-
Home page
-
An analytics portal that provides real-time insights on crowd density, known vs. unknown headcounts, and unusual activity detection.
-
Page that displays all unmatched faces that can be accepted or declined. Accepted faces are added to known faces collection.
-
Page that displays known faces. We can remove known faces if we no longer want to authorize that person.
-
Live CCTV footage
Inspiration
In an age where security is paramount, we were inspired to leverage AI for real-time surveillance enhancement. The idea stemmed from the need to automate monitoring and improve the efficiency of security personnel, ensuring timely alerts and robust safety measures. Our goal was to create a system that not only identifies authorized individuals but also flags potential intruders to prevent unauthorized access.
What it does
The Face Recognition and Alert System scans CCTV footage in real-time to detect faces. It logs authorized visitors for future recognition and sends alerts for unknown individuals. The system also provides analytical insights, such as crowd density and the ratio of known vs. unknown headcounts, helping admins make informed security decisions.
How we built it
We developed the Face Recognition and Alert System using a full-stack approach with Node.js, Express.js, React.js, and MongoDB Atlas. The backend handles real-time data processing and communication using WebSockets, while the frontend, built with React.js, offers an intuitive dashboard for admins to monitor and review alerts.
The system processes frames from CCTV footage using the following technologies:
YOLO (YOLOv3) for detecting people in video frames.
MTCNN from facenet_pytorch for accurate face detection.
FaceNet (InceptionResnetV1) from facenet_pytorch to generate embeddings for face recognition. These face embeddings are stored in MongoDB Atlas and searched using Vector Search for efficient and scalable face matching. Authorized faces are logged for future reference, while unknown faces trigger real-time alerts to enhance security. Additionally, the system offers analytical insights, such as crowd density analysis and known vs. unknown headcount, using dynamic visualizations.
Optimized models and MongoDB's Vector Search capabilities ensure faster execution, allowing for quick detection of unusual activities and real-time alerts.
Challenges we ran into
Real-Time Performance: Ensuring low-latency facial recognition from live CCTV feeds was challenging, requiring optimizations in frame capture and model inference.
Data Storage and Scalability: Managing a large volume of facial embeddings and ensuring quick search queries led us to optimize database queries and indexing strategies
Accomplishments that we're proud of
Successfully implemented a real-time alert system that accurately distinguishes between known and unknown faces.
Integrated analytical dashboards providing insights into crowd density and visitor logs. Built a scalable solution with a seamless WebSocket integration for real-time notifications.
What we learned
Gained a deeper understanding of real-time video processing, WebSocket communication, and facial recognition algorithms.
Enhanced our skills in database management and optimizing search queries for large datasets.
Learned the importance of system scalability and efficient resource management in deploying AI-based applications.
What's next for Face Recognition and Alert System
Enhancing Detection Accuracy: Implementing advanced models like YOLOv8 or Vision Transformers to improve face detection and recognition accuracy.
Unusual Activity Detection: Integrating AI models to detect unusual behaviors, such as loitering or sudden crowd surges, enhancing proactive security measures.
Optimizing Model Execution: Accelerating facial recognition models using ONNX and TensorRT for faster inference, ensuring real-time performance even with high-resolution feeds.
Mobile App Integration: Developing a mobile interface for admins to receive alerts and view real-time footage on the go.
Advanced Analytics: Adding features like heat maps, visitor frequency analysis, and behavioral pattern recognition to further enhance security insights.
Edge Deployment: Exploring edge computing solutions to process video feeds locally, reducing server load and improving response times.
Log in or sign up for Devpost to join the conversation.