AI Attendance
Face-recognition attendance that keeps video on campus
A small edge box on your LAN reads the gate and classroom cameras, recognises enrolled students and staff, and sends only tiny attendance events to the cloud. Teachers confirm a pre-filled register in seconds.
- Runs on CPU - no GPU
- Video stays on campus
- Consent before enrollment
Period 1 · Maths · 08:30
Class 5 - A register
- 1Aanya Gupta0.82Present
- 2Aarav Sharma0.77Present
- 3Arjun Mehta0.71Late
- 4Diya PatelOn Leave
- 5Ishaan Rao0.41Review
- 6Kabir SinghAbsent
1 uncertain match needs your check
Confirm registerWhy edge-first
Process video where it is captured
Cloud video processing is expensive, slow on school broadband and a privacy risk. Edge processing flips the economics.
60 Mbps
Uplink needed to stream 30 classroom cameras at 2 Mbps each - plus a cloud GPU for every few schools, and video of minors leaving the premises.
< 1 KB
Per recognition event. Video stays on campus; the cloud only applies attendance rules and sends notifications.
Private by design
Video never leaves the edge. Raw face images are not kept after enrollment by default.
Works offline for days
Recognition continues without internet; events sync later with no loss.
No GPU required
Runs on an affordable mini-PC on your campus.
Recognition pipeline
Detect, track, check, recognise, vote - then emit one event
Every face goes through several checks before attendance is marked - all on a small device on your campus.
Camera feed
From gate, classroom or bus cameras
Motion check
Only looks closely when someone is moving in view
On the edge box (campus or bus) - video never leaves
No GPU needed- 01
Detect
Find faces
Spots every face in view
- 02
Track
Follow each person
Counts each person once as they walk through
- 03
Quality
Size, blur, pose, light
Skips faces that are too small, blurred or turned away
- 04
Liveness
Anti-spoofing
Rejects photos and phone screens - gate & bus
- 05
Template
Secure face template
Converts the face into an encrypted template
- 06
Match
Against enrolled people
Only confident, unambiguous matches are accepted
- 07
Vote
Several frames agree
Multiple frames must agree before marking
- 08
Event
Attendance event
Who, where and when - stored safely until sent
Sync
Securely sent and retried until delivered - never duplicated
EduPresence cloud
Applies attendance rules, updates the register and notifies parents. Never processes video.
Where cameras go
Each camera has one job
Gate cameras run continuously for daily attendance; classroom cameras take scheduled snapshots; bus cameras wake only when the door opens.
Existing CCTV can be reused only if it captures faces clearly enough - most wide-angle CCTV does not.
Cameras are positioned so students face the lens naturally while walking in.
Lighting matters more than the model: avoid cameras facing the sun or doorway glare; add LED light at the gate for winter mornings.
Hardware tiers
Pick the box that fits your campus
Indicative loads for planning. Before purchase we run the pipeline on the exact box with your real camera streams and measure end-to-end latency per frame.
Starter
Compact mini-PC - no graphics card needed
Load
2-4 gate cameras live + classroom round-robin scanning
Use
Small schools, coaching institutes
Standard
Mid-range mini-PC
Load
6-10 live cameras + 40-60 classrooms scheduled
Use
Medium schools
Advanced
AI-accelerated edge device
Load
10-20+ live streams
Use
Large campus, colleges
Enterprise
On-premise server
Load
30+ live streams
Use
Multi-building universities
Bus device
Rugged tablet or compact in-bus unit
Load
1-2 cameras
Use
Every bus
Enrollment & consent
No consent, no enrollment
Children's face templates are the most sensitive data we hold. Enrollment is blocked for any student without recorded biometric consent from a parent or guardian.
- Verifiable parental consent for students (OTP, DigiLocker or signed form, with notice version and timestamp); adult staff consent for themselves
- Capture 5-10 images per person: frontal, ±20° left / right, slight up / down, with and without spectacles, good light
- Blurred or low-quality captures are rejected automatically
- Optional bulk enrollment from ID-card photos (weaker - schedule a proper capture day)
- Secure face templates are created from the captured images
- Raw images deleted after enrollment unless your policy says otherwise
- Campus devices stay up to date automatically; consent withdrawal deletes templates everywhere
1. Consent recorded
Guardian gives biometric consent in the parent app (OTP) or on a signed form.
2. Capture day
Teacher or office captures 5-10 quality-checked images in the admin app.
3. Templates created
Images are converted into encrypted face templates, kept separately for each school.
4. Gallery sync
Campus devices receive the latest templates automatically.
5. Withdrawal honoured
Withdrawing consent deletes the templates from the cloud and every edge device.
Accuracy controls
Built so a child is never wrongly identified
Target: > 95% of enrolled students recognised at the gate each day with a false-accept rate of ~0 at the chosen threshold.
Open-set matching
Unknown faces are rejected, never matched to the nearest student. A match must be confident and clearly better than the next-closest person.
Roster prior
In a classroom we match against that section's roster (~40 people) first - fewer false matches, faster - then fall back to the whole school.
Multi-frame voting
Best 3-5 frames per track; 2+ agreeing frames required for gate and bus events.
Twins & siblings
Pairs with very similar templates are flagged at enrollment and always need teacher confirmation.
Human in the loop
AI pre-fills; teachers confirm period attendance; uncertain gate matches go to a review queue with thumbnails. Nobody is penalised on AI alone.
Per-camera metrics
False accepts / rejects tracked per camera from teacher corrections; thresholds recalibrated per camera.
Masks, sunglasses, helmets
A 'please look at camera' prompt on the gate display, with RFID / QR as the fallback.
Template aging
Children's faces change fast: annual re-enrollment plus automatic updates from high-confidence gate matches.
Anti-spoofing
Passive liveness at the gate and on the bus
No blinking or head-turning required. Every face is checked for liveness before it is matched.
- Stops printed photos and phone-screen replays in milliseconds
- Combined with track behaviour: a real person walks through the capture zone over several frames
- Limit: high-quality 3D masks remain hard - an optional depth / IR camera can be added at high-security gates
Alternatives considered
Why edge face recognition is the primary method
Every other method stays available as a fallback - so each student always has a way to be marked.
Rolling out? Start with a pilot at a single gate with ~300 consenting students. Measure recognition rate, false accepts, time per person and teacher correction rate for 4 weeks, then scale to classrooms and buses.
FAQ
AI attendance questions
Works better together
Explore related modules
Attendance, leave, transport and alerts share one data model - so an approved leave quietly suppresses the absent and missed-bus alerts.
Pilot AI attendance at one gate
We survey your gate, recommend cameras and an edge box, and run a 4-week pilot with consenting students before you scale.
- Free campus survey
- Live in days without hardware
- Hosted in India