EAS Gate Detector
What This Is
After looking at thousands of real EAS installations, I want to turn that accumulated visual experience into a normal tool. The first version is simple: people send photos of gates, entrances, exits, and checkout areas; the dataset grows; the model learns to recognize EAS systems from ordinary field photos.
On my current dataset of roughly 1000 images, AM and RF systems are already separated fairly well when the photo is decent. To make the service useful in the real world, the model needs more reality: good shots, bad shots, reflections, glass, advertising panels, partial views, and all the weird angles people actually send.
Pretty photos are useful, but ugly photos are useful too. A good recognition model must survive normal human photography, not a sterile lab.
What To Photograph
Gates
AM, RF, RFID, hybrid, old, new, branded, unbranded, complete or partly visible.
Entrances
Street view, inside view, glass doors, mall corridors, and store-front context.
Checkout Zones
Cashier exits and post-checkout paths where gates stand or can be clearly seen.
RFID And Rare Systems
Ceiling RFID, side modules, built-in blocks, extra metal-detector frames near gates.
No-Gate Stores
Entrances where gates are absent. Negative examples help the model stop hallucinating gates.
Different Quality
Sharp, blurry, close, far, side angle, through glass, bad lighting, ads in the way.
How To Shoot
Take several angles of the same entrance: outside, inside, side view, wide view, and closer details.
Shoot as if you were asking an experienced person to identify the gate from the photo.
Use the normal phone camera if possible, then send the image to the messenger. Telegram's built-in camera can distort the frame.
Avoid faces, private documents, cashier screens, and unnecessary personal information when possible.
If taking a photo is forbidden, unsafe, or likely to create conflict, skip the place. Safety beats one more image.
Sample Frames
Where This Goes
The target is to distinguish AM, RF, RFID, manufacturers, some gate models, and rare installations. Later the dataset can also include labels, tags, alarms, hard tags, RFID blocks, and the rest of the hardware people usually ask about in the same breath.
The bot already tracks accepted photos, duplicates, sync status, and future dataset contribution states. That matters because useful contributors should be visible, not lost in a pile of files.
For Coders And Data Miners
If you know how to collect useful public imagery from maps, panoramas, store pages, reviews, open datasets, or similar sources, I am interested in practical help. The model can already be used as a filter that keeps photos where gates are likely present. What is missing is a clean, repeatable method for finding entrance and checkout-zone images from allowed sources and feeding them into the training loop.
Rewards And Respect
The best helpers will be credited as dataset contributors. When possible, I will also send useful equipment from the shop as gifts. If you have more time than money and you like the topic, this is a real way to help the project move.