The increasing accuracy of computer vision technology that is based on deep learning coupled with the declining cost of in-store technology is improving the cost-effectiveness of intelligent retail loss prevention solutions. High-resolution cameras, computers, data storage, and networking together with new AI-powered software can detect both customer and employee behavior that is associated with inventory loss, while keeping false positive results to a minimum. High-resolution cameras give employees a clearer picture over a wider viewing area. However, this solution is still labor-driven.
The real breakthrough technology for loss prevention comes from intelligent software that can be placed in-store. Large retailers and third-party solution providers are investing in the development of computer vision for loss prevention applications. Advances in deep learning models that are trained by using massive amounts of data and large clusters of computers in remote data centers are being applied to retail loss prevention use cases. The outputs from these “trained models” can be used to generate real-time in-store alerts by using less expensive computers that are often found in retail stores for operations such as inventory and POS data aggregation.
High-resolution cameras and software models have been developed exclusively for use in both staffed and SCO lanes. One solution for detecting mis-scans and ticket switching uses data from both the scan terminal and a video stream of the activity at and around the checkout area. The system software attempts to match the items that are being scanned at the terminal with what the camera detects at the checkout lane. Items that are identified in the video stream but do not have a corresponding UPC barcode scan on the terminal are potential mis-scans or ticket swaps. These solutions can also detect items that are left in or under the shopping cart.
Many other applications are being developed for loss prevention in other areas of the store. For example, a series of video frames that show merchandise being placed in a pocket or concealed beneath a garment easily identifies abnormal shopping behavior. Another possible scenario is alerting security personnel if a person loads a shopping cart with an abnormally high number of expensive items. This behavior might indicate that the customer is preparing to avoid all checkout lanes and proceed directly out of the store in a classic “grab and go” theft.
Solutions that are based on the use of intelligent software models that are developed with deep learning techniques address several key issues for retailers. They reduce the need for additional loss prevention staff by operating largely unattended. These solutions can significantly reduce the false positive rate of traditional weight sensor-based systems. Smart vision systems can notify store personnel while a suspected shoplifter or employee commits theft during the transaction in contrast to systems like RFID that operate only at the door.
