ISACS: In-Store Autonomous Checkout System for Retail
summary
An autonomous checkout system combines a 3D store model, multi-person 3D pose estimation, and live inventory signals to match shoppers to products; deployed for 13 months in an 800 sqft convenience store with 52 ceiling cameras and 58 shelf weight sensors, it tracks interactions, triangulates people in 3D, and resolves ambiguous weights with a product layout model, yielding up to 96.4% daily receipt accuracy over ~140 daily transactions.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:19Autonomous Retail! Why? The waiting line
- 1:12How to automate retail?
- 1:52Wait, isn't that what Amazon Go does?
- 2:10ISACS In-Store Autonomous Checkout System for Retail
- 3:03Modeling the physical elements
- 4:10Modeling Interactions to enable spatial-temporal multiple human-object matching
- 5:08Real-World evaluation at an operating store
- 5:25Operating a store with real customers
related talks
23mArtificial Intelligence for Store Assistance
14mComputer vision on edge devices
07mAI in Retail
48mSmart Retail - Vision, Cognition and Intelligence
42mThe Rise of AI
01h 00mAutomation: Amazon’s Supply Chain Optimization Technologies (SCOT)
07mAdvanced analytics and the mystery of the missing jeans
23mAI-Powered Cognitive Retail Transformation
