A Raspberry Pi 5 + Sony IMX500 security system with on-device face recognition, local AI scene descriptions, and full control from Telegram — no cloud AI, no subscription.
The system runs on a Raspberry Pi 5 paired with the Raspberry Pi AI Camera — a Sony IMX500 sensor with its own on-chip neural network accelerator. Person detection and pose estimation happen directly on the sensor; face recognition and scene analysis run on the Pi's CPU, entirely offline.
One camera, two modes, all controlled from a phone:
Guided 3-angle enrollment, then silent logging for known faces and instant alerts for strangers.
Send /scene and a local vision model describes what's happening, using the recognized person's name.
Burst photos to Telegram and an optional GPIO siren when an unrecognized face is confirmed.
Switch to IMX500 HigherHRNet pose estimation — 17-keypoint skeletons and basic action classification.
Password-protected MJPEG stream over ngrok, viewable from any browser, anywhere.
Scheduled or manual quiet hours mute the siren while Telegram alerts keep coming through.
A walkthrough of the system in action — face recognition, alerts, and live scene analysis.