ShelfSense AI

ShelfSense AI is a smart shelf monitoring prototype.

It uses a camera and a locally trained YOLO model to detect products and estimate visible stock levels. The system runs on a Raspberry Pi 5 and shows the results in a Gradio dashboard and on an LCD screen.

RFID badges are used to link shelf changes to users. If stock decreases without an active RFID session, the system creates an alarm with the buzzer and warning feedback.

Raspberry Pi 5

Official Raspberry Pi 5 USB-C 27W Power Supply

USB webcam

Webcam stand

Verbatim MicroSDHC 32GB

TP-Link SD memory card reader

Freenove Project Kit

16x2 I2C LCD screen

RC522 RFID reader with RFID card/tag

Active buzzer

RGB LED

Breadboard

Jumper wires

Resistors

Vida Designs Oxford Cube Bookcase 2 Tier Black Wood

4 mm MDF enclosure / laser-cut box

Scotch double-sided tape

VELAMP MG203 cable clamps

Test products: bottle, can, chips bag, snack bar, toothpaste box, milk carton box

Software: Python, Gradio, PostgreSQL, Roboflow, YOLO model

The estimated price is around 270 euros.

I designed and assembled the physical box for the prototype. The enclosure keeps the ProjectBoard, Raspberry Pi, LCD, RGB LED, RFID reader, buzzer and wiring in a stable setup for the demo.

I connected the Raspberry Pi, USB camera, LCD screen, RGB LED, RFID reader and active buzzer. After wiring the components, I tested each part separately before using them together.

I collected shelf images from the camera angle used in the final prototype. I annotated the products in Roboflow and trained a YOLO model to detect the product classes on the shelf.

I built a Gradio dashboard to show the camera view, stock status, RFID timer and detection results. I used PostgreSQL to store product references, detections, RFID sessions, stock movements and alerts.

I tested the full flow together: camera detection, stock counting, RFID authorization, database updates, dashboard feedback, LCD messages, RGB LED status and buzzer alarms.