Some sites don't want every guard carrying their own phone for attendance โ a manned gate with a shared tablet is simpler to manage. Raksha Kavach's Face Kiosk mode turns an Android tablet into exactly that: a supervisor logs in once, the screen stays in a full-screen camera view, and a registered guard's face is detected and matched automatically to mark their check-in or check-out.
How a supervisor sets it up
A supervisor pairs a tablet to a site using a one-time pairing code issued from the admin panel, then signs in with their own credentials. That session persists โ the kiosk doesn't need re-login for every shift change โ until the supervisor signs out manually or an admin remotely logs the device out from head office.
Registering a guard's face โ from multiple angles
A guard is enrolled by capturing their face from several angles (front, left, right, up, down) directly on the tablet itself, guided step by step. Matching against multiple angles rather than a single front-on photo is meant to make recognition more reliable across the range of head positions a guard might actually be in when walking up to a kiosk, not just looking dead-center at the camera.
What happens during a scan
Face detection and matching both run on the device itself, not by streaming video to a server โ recognition works even during a brief connectivity drop. The kiosk includes a basic liveness check that looks for a genuine eye-blink transition before accepting a match, intended as a first line of defense against someone holding up a printed photo. We want to be upfront here: this is a basic heuristic, explicitly not a certified anti-spoofing system, and it's described that way deliberately rather than oversold.
Server-side validation still applies โ nothing is trusted blindly
A successful on-device match doesn't get recorded as attendance without further checks. Every accepted scan is independently re-validated on the server: is this guard actually deployed at this site today, does the scan time fall inside their scheduled shift window (with a grace period for early arrival or late departure), and is the device's location inside the site's geofence. A scan that fails any of these is rejected with a specific reason rather than silently accepted.
Offline queueing
If the tablet loses connectivity mid-shift, an accepted scan doesn't just fail โ it's queued locally on the device and automatically replayed to the server once the connection returns, preserving the actual scan time rather than recording it as whenever the sync happened to succeed.
Everything else in this feature โ the camera pipeline, the liveness heuristic, the offline queue, the geofence/shift re-validation โ is real, working code. What genuinely still needs field validation is real-world match accuracy on a physical tablet in actual site lighting conditions, which is honestly outside what a lab or office setting can fully confirm.
Admin visibility and alerts
Head office gets a live dashboard of kiosk activity โ device status, today's scan counts, and a recent-activity feed โ plus automatic notifications for situations that need attention: an unrecognized face, a device that's gone offline, low battery on a tablet, or a burst of repeated failed matches that might indicate a hardware or lighting problem.
Conclusion
The Face Kiosk is a genuinely useful option for sites where a shared device makes more sense than individual guard phones, built with real geofence, shift-timing and offline safeguards around the recognition step. We'd rather tell you plainly what's field-proven and what isn't than oversell an accuracy number we haven't actually measured on real hardware yet.