APNA CAMPUS
Automating classroom attendance with AI-powered face recognition, liveness detection, and real-time CCTV-based student tracking.
- Problem, approach and result explained
- Expert reviewed· 6 reviews
80
Face recognition accuracy
83
Real-time video processing speed
10.6
Average recognition latency
Overview
The problem
In a typical classroom, attendance is often taken manually or through basic face recognition systems. Both approaches have limitations.
Manual attendance consumes valuable teaching time and can lead to human errors. Traditional CCTV-based face recognition also faces challenges because classroom cameras are often mounted at a distance or on the ceiling. Students may turn their faces away, sit in different positions, become partially hidden behind others, or appear too small for accurate face recognition.
As a result, a student may be physically present in the classroom but remain unrecognized by the system.
The bigger problem is that face recognition alone does not understand a person's movement or location inside the classroom. It can identify a visible face, but it cannot independently explain whether someone entered the classroom, moved to a seating area, or remained there.
We identified the need for a smarter solution that combines identity verification with spatial tracking.
Our proposed solution, V-LASER (Virtual Laser), addresses this gap by creating software-based virtual grids, entry and exit lines, and classroom zones over existing CCTV footage. It aims to track people's movement and support the existing face recognition system in associating verified identities with the correct person tracks.
Our objective is to make classroom attendance more intelligent, reliable and transparent while making better use of existing CCTV infrastructure, without requiring physical laser hardware or additional cameras.
The approach
We started by identifying a major challenge in our existing CCTV attendance system: real classroom conditions. When multiple students sit together, their faces may be turned away, partially hidden, or too small for the camera to recognize. This makes it difficult to identify and continuously track each student using face recognition alone. To solve this, we came up with V-LASER (Virtual Laser). It creates software-based virtual grids, tracking lines and classroom zones over existing CCTV footage. Each detected person is assigned a temporary tracking ID, allowing the system to continue tracking their movement even when their face is not visible. Once the student's face becomes clear, our existing face recognition and liveness detection system attempts to verify their identity and associate it with the correct tracking ID. If verification is unsuccessful, the identity remains pending rather than being guessed. In this way, we combine spatial tracking with face recognition to make classroom attendance more reliable, using existing CCTV infrastructure without requiring physical laser hardware or additional cameras.
The result
We developed an AI-powered Smart Attendance Management System that uses live IP CCTV footage to automate student attendance through computer vision and facial recognition.
The system fetches live video streams over the network and processes video frames using an integrated AI pipeline consisting of:
YuNet: Face detection and facial landmark extraction.
EdgeFace: Face recognition using 512-dimensional facial embeddings and cosine similarity.
AuraFAS: Face anti-spoofing and liveness verification.
ByteTrack: Multi-object tracking and temporary identity assignment.
OpenCV: Image preprocessing, low-light enhancement, face alignment and video processing.
The backend is developed using Python, FastAPI and SQLite, while the frontend uses React, TypeScript and Electron. The system supports student identification, attendance verification, faculty-controlled class sessions and database-based attendance management. We also designed an advanced V-LASER concept (Virtual Laser Assisted Spatial Event Recognition) to extend the system with virtual detection zones, entry/exit tracking, spatial event analysis and intelligent alerts without requiring additional physical hardware.
Reflection
During the development of this project, we learned that building a real-time AI attendance system invoariations: Low-light conditions, blur and reduced image quality can affect face detection and recognition.
Different Face Visibility Levels: Since the CCTV camera captures the classroom from a distant ceiling position, some faces are clearly visible, while others appear small or partially obstructed.
Multi-Person Tracking: Maintaining consistent tracking IDs when students overlap, move simultaneously or temporarily disappear from the camera view.
Recognition Reliability: Preventing incorrect identity matches and ensuring that attendance is recorded only after the required verification checks.
Real-Time Performance: Balancing video frame processing, AI inference speed and system responsiveness.lves much more than simply implementing a face recognition model. Integrating live CCTV streams, network communication, computer vision, object tracking, biometric verification and database management presented several technical challenges.
Submitted to Attendance, Mess, Hostel, Repeat: Campus Life, Debugged
- Your idea in one line
Apna Campus is an AI-powered, all-in-one College Management System that unifies student admissions, academics, faculty, attendance, examinations, fees, transport, and campus operations into one intelligent platform for colleges and higher education institutions.
- What will be working by the finale?
1. Role-Based Login & Dashboards: Working authentication with dedicated portals for students, faculty, HODs, administrators, and non-teaching staff. 2. Centralized Student Management: Course-wise admissions, student profiles, enrollment records, and faculty-student mapping. 3. Academic Management: Attendance tracking, timetable management, assignments, internal marks, examination records, and result publishing. 4. Smart Administration: Fee management, document verification, leave requests, approvals, and automated notifications. 5. Campus Operations: Functional modules for library, hostel, transportation, and placement management. 6. Analytics & Reports: Real-time dashboards, attendance summaries, academic performance insights, and administrative reports. 7. Integrated Database & API: A connected React frontend, NestJS backend, and PostgreSQL database with validated APIs and role-based access control. 8. End-to-End Prototype Demo: Demonstration of complete workflows using sample college data, including student registration, faculty actions, approvals, and report generation.
- Tech stack and hardware you plan to use
Frontend: React 18, JavaScript (ES6+), Vite, HTML5, CSS3, Tailwind CSS, and responsive UI components. Backend: NestJS, Node.js, and Express.js for REST APIs and service integration. Database & ORM: PostgreSQL with Prisma ORM. Authentication & Security: JWT-based authentication, role-based access control (RBAC), and secure password hashing. Analytics & Intelligence: Chart.js/Recharts for dashboards, automated reporting, and rule-based insights, with scope for AI-powered analytics. Development & Deployment: Git, GitHub, VS Code, Postman, npm, Docker, and cloud deployment. Integration: REST APIs connecting student, faculty, administration, academic, transport, library, hostel, and placement modules. Hardware Requirements Standard laptops or desktop computers for development, testing, and demonstrations. Internet connectivity for cloud deployment and API integration. No specialized hardware, sensors, or microcontroller boards are required because Apna Campus is a software-based college management platform.
- GitHub repository
- github.com/sahooprasadbishnu1234/CMS.git
- What's working now?
Existing CCTV footage is available for live and recorded video processing. The existing AI architecture includes face detection (YuNet), face recognition (EdgeFace), liveness detection (AuraFAS), and person tracking (ByteTrack). The proposed V-LASER architecture and attendance workflow have been defined.
- GitHub repository
- github.com/Ashish-kumar78/APNA_CAMPUS.git
AI tools used
ChatGPT was used for system architecture planning, developing detailed implementation prompts, understanding AI/ML models, debugging technical issues and improving the integration of the face recognition and attendance pipeline. Claude was used for code analysis, reviewing implementation approaches and improving code structure. Gemini was used for exploring alternative technical solutions, understanding concepts and brainstorming feature improvements. Perplexity was used for technical research, exploring documentation and comparing relevant AI and computer vision approaches. AI tools supported development and problem-solving, while implementation, integration and testing were carried out within the project codebase.