How does the vision system recognize products?

How the vision system works, what can/cannot be stocked, how long does the transaction take

What does the vision system use to detect what was purchased?

Visual sensors use a mix of packaging shapes, colors, sizes and location in the cabinet to accurately detect which item(s) a customer has taken.

Micromart uses multiple inward-facing AI-powered visual sensors to recognize what items are taken out based on these packaging images. These sensors use a mix of packaging shapes, colors, sizes, and cabinet location to accurately detect which item(s) a customer has taken. Each of these factors acts as a signal to our system to accurately detect what was purchased by a customer.

Our system does not use weights, RFID, or barcode scanning. Customers can return products to the cabinet in a different location than where they were originally placed with minimal impact on product detection accuracy.

How long does it take to process a transaction? 

A human reviewer may review transactions to ensure billing and inventory accuracy before finalization. This process may take up to a few hours after the customer has used the system. 

What items can / cannot be stocked?

Small items that can be entirely covered by one hand are not permitted to be stocked. Items must be visually distinct in shape or color.

  • Items must be greater than 4” and cannot be completely fit in the palm of one’s hand.
    • For example, a "5-Hour Energy Shot" or a Box of Tic-Tacs are not compatible with Micromart. Smaller items (like Chapstick or travel size Tylenol may be acceptable if they are in a larger blister pack)
  • Items must be visually distinct in shape or color. If items are similar, place them on opposite sides of the shelf