Digital Verified Driver

Edge AI that verifies who is behind the wheel — and watches the road for danger.

Purpose

Identity at the roadside. Safety in the stream.

The camera sees a face and a plate. The edge stack reads plate text locally, extracts a face embedding on the server, and decides: verified match, face-only match, or a clear violation — with images, confidence, and GPS when available. A second Pi path watches for accidents, potholes, and violence.

Version 1 — Auth

Raspberry Pi, IP camera, and LCD for real-time face recognition and license-plate identification against authorized vehicle records.

Version 2 — Cabin

Dual camera modules plus GSM/GPS for in-cabin monitoring, threat detection, and owner alerts with vehicle location.

FastAPI + InsightFace

512-D embeddings, plate compare, multi-driver plates, JWT roles, MongoDB database dvd.

Dashboard

Local 7″ touch UI plus web dashboard — live violations, evidence, trips, map, and Email/Telegram alerts.

Driver verification

Face match against driver_faces. Full match or face_only when plate text is empty and PLATE_STRICT=false.

Plate from the Pi

No server EasyOCR. Plate YOLO → Char YOLO → CNN OCR on device; server cleans and compares.

Violations

wrong_driver, wrong_plate_owner, no_face_detected, no_plate_detected — with decision metadata.

Road incidents

Accident, pothole, violence events with image, confidence, device id, and map coordinates.

Operations

Person registration (photo / JSON / manual embedding), trip log, audit log, access requests.

Notifications

Email and Telegram on serious violations and high-confidence road alerts.

Python FastAPI MongoDB InsightFace YOLOv5 PyTorch OpenCV Raspberry Pi Picamera2 JWT / RBAC Leaflet ReportLab / openpyxl

Collections include users, persons, driver_faces, checks, violations, incidents, devices, vehicles, and audit_logs — served through a JWT-protected dashboard with public Pi endpoints for check-driver, incidents, and heartbeat.

Compute

Raspberry Pi 5

Main processing unit for vision pipelines, peripherals, and software logic. Quad-core Cortex-A76 up to 2.4 GHz, 16 GB LPDDR4X, Wi-Fi/Bluetooth, Gigabit Ethernet, 40-pin GPIO.

IP camera

Uniview IPC2122LB-SF40-A

2MP Mini Fixed Bullet for surroundings and plates. 1080p@30 fps, 4.0 mm lens, Smart IR to 30 m, H.265/H.264, PoE, IP67, RTSP streaming to the Pi.

CSI camera

Raspberry Pi Camera 5MP

OV5647 module on CSI for driver face and plate frames — primary visual input for detection, identification, and verification.

Comms + GPS

SIM808 GSM/GPS

Quad-band GSM/GPRS with integrated GPS over UART. Sends SMS/location alerts on unauthorized use and supports real-time vehicle tracking.

Storage

SanDisk Ultra 256 GB

microSDXC UHS-I (up to 140 MB/s) for OS, models, captured images, and configuration on continuous edge workloads.

Power

Raspberry Pi 5 PSU

USB-C 5 V / 5 A (~27 W) regulated supply for stable capture and inference with cameras and modules attached.

HMI

7″ HDMI touch LCD

1024×600 capacitive display (up to 5-point touch) for local real-time violation output; pairs with the remote web dashboard for logs and evidence.

Emergency

Push button

GPIO momentary switch that pulls GPS location and posts it to the web platform for accidents, medical, or security emergencies.

Auxiliary

Cooling & housing

Cooling fan for sustained Pi load; camera housing for mechanical protection and stable optical alignment.

Evaluated, not adopted

T-A7608SA-H (ESP32 + 4G/GPS) — capable, but costlier than SIM808; ESP32 unused with Pi as host. STEVAL-ASTRA1B — strong AI/IoT candidate, unavailable in the Egyptian market during build.

Edge intelligence