Retail shelf vision on Triton
One Triton pipeline turning video from SmartBadge wearable cameras into product detection, planogram matching and out-of-stock alerts, with uncertainty filtering so store teams trust the alerts.
Problem
Augmodo’s SmartBadge is a wearable camera that store associates wear while they work, so shelf video is cheap to capture. Understanding it is the expensive part. Retailers want to know what is on the shelf, whether it matches the planogram and what has run out, without a flood of false alarms.
What I did
I shipped a Triton inference pipeline that serves all three tasks from one stack: product detection in shelf video, shelf-to-planogram matching and out-of-stock detection. I added uncertainty filtering to product lookup, so low-confidence matches stop masquerading as empty shelves.
Result
Three production vision tasks behind one serving stack, and fewer false out-of-stock alerts for store teams.
system
scroll sideways on small screens