Shopping cart cameras were programmed to monitor hand movements so that when a customer picked up an item, the system could identify it. This feature allowed for tracking of specific actions like adding or removing a product in the cart, thus functioning as a precise customer activity tracker.

We created an action recognition model based on deep learning to identify particular activities of customers, for instance, handling a product. This model identifies and times the pick and putaway processes, which allows retailers to track shoppers’ behavior in real-time.

It recorded each item detail of interaction in real-time, because of which, inventory details were always up-to-date. This feature minimized the time that in-store activities took before the stock records were updated, increasing stock accuracy as a whole.

Each transaction, including addition or deletion, was time-stamped to give the retailer a record of the customers’ activities. This enabled the tracking of specific activities such as the time between the customer picking up an item and placing it into the cart.

The solution was intended to be easily integrated into the current retail management systems. This made it possible for all data recorded by the cameras and models to be integrated with inventory and security systems for operation streams.