Outcome and proof
The system is deployed at a live vehicle-entry point. It runs on an edge device, keeps the decision path local, and connects identity-card reading to the physical barrier action.
- Deployed at a live vehicle-entry point
- Runs on an edge device
- No cloud round trip in the decision path
- Denial and review path for uncertain or unauthorized reads
Context and problem
Vehicle entry previously depended on a person at the barrier checking an identity card against an authorization list. That process was slow, repetitive, and tied access to a manual lookup.
My role and responsibility
I designed and implemented the end-to-end computer-vision and access-control pipeline: detecting the card, reading the identity number, checking authorization, and connecting the decision to the barrier relay.
Solution and architecture
- Detect — a YOLO model locates the CNIC in the camera frame.
- Read — the crop is preprocessed and passed through OCR.
- Verify — the extracted number is checked against the authorization database.
- Act — an authorized record triggers the barrier; other outcomes are denied and available for review.
Key engineering decisions
Isolate the card before OCR
Cards arrive with glare, motion, tilt, and inconsistent distance from the camera. The detector gives the OCR stage a controlled crop instead of asking it to read the entire vehicle-lane frame.
Keep the decision local
The pipeline runs at the barrier so entry is not dependent on a cloud inference round trip. The database check remains part of the operational decision path.
Fail closed
A successful detection is not treated as authorization. The barrier opens only after the extracted number maps to an authorized record; uncertain and invalid outcomes remain closed.
Verified results
The verified public evidence is qualitative because operational volumes and identity-reading performance are private client metrics. The system is deployed, runs on edge hardware, and completes the full path from camera input to physical barrier control.
Visual evidence
The system-flow diagram above documents the deployed decision path without exposing a readable national identity card, a client database, or the installation site.
Technology stack
YOLO, OpenCV, Python, OCR, PostgreSQL, and edge-device integration.