Retail shrinkage and industry losses are increasing due to rising inflation. To allow developers to build and release applications to protect against the theft of merchandise, NVIDIA announced three Retail AI Workflows built upon its Metropolis microservices. These AI workflows offer pre-trained AI models and applications required to accelerate development. Moreover, they quickly search thousands of store items for barcodes, cross-camera identification, and identify ticket switching and theft.
Retail store analysis AI workflow allows developers to develop end-to-end retail AI applications to track store performance using customized dashboards. Developers can use and modify the workflow to suit a range of data analytics, including queue analytics such as shopper occupancy, shopper count, the duration of a stay and its direction, heat mapping of the customer’s journey, line crossings, proximity and the regions of significance.
These attributes are easily modified to contain details about specific usage scenarios specific to the individual stores. Stores can use this information to improve staffing, store layout and merchandising and enhance customer experiences to increase sales.
Principal Benefits of the AI Workflow for Retail Store Analytics
Pretrained Models: The AI models used in this workflow are trained to recognize and design feature embeddings for objects. Moreover, they provide behavior analytics and learn as well as others.
Custom Dashboards: Create an end-to-end vision for retail AI applications to analyze store performance by creating customized dashboards.
Flexible Reference Architecture: Users use cloud native microservices with completely customized AI workflow to accelerate the development and implementation of solutions for retail store analytics.
The NVIDIA Retail AI Workflows are available via NVIDIA AI Enterprise, the suite of software comprises:
- Retail Loss Prevention AI Workflow:
The AI models that are part of this process are trained to recognize the hundreds of products that are most often stolen (including alcohol, meat and detergent. It also teaches them to recognize them in the different sizes and shapes offered. With the synthetic data generated using the NVIDIA Omniverse, retailers and independent software providers can tailor and further develop the models for hundreds of thousands of store items. The process is based on the latest techniques for learning using a few shots created in NVIDIA Research which, combined with active learning, recognizes and tracks any new products that shoppers and sales representatives scan during checkout to increase the accuracy of models.
- Multi-Camera Tracking AI Workflow:
Provides multi-target, multiple-camera (MTMC) abilities that enable developers to develop systems that can monitor objects from multiple cameras in the store. This workflow monitors objects and associates of stores across cameras and creates a unique ID for every object. The objects are tracked using visual embeddings or the appearance of objects instead of biometric information to ensure complete privacy for shoppers.
- Retail Workflow of Store Analytics
Computer vision provides insights into store analytics, including traffic patterns in the store, numbers of shoppers with shopping basket aisle occupancy, and much more through custom dashboards.
The workflows are based on NVIDIA Metropolis microservices, a low- or no-code method of creating AI applications. The microservices serve as the building blocks to build sophisticated AI workflows that allow them to scale into production-ready AI applications quickly.
Developers can easily modify and extend AI workflows, for example, by integrating the models of their choice. Microservices can also allow for easier integrating of new services with older systems, including point-of-sale systems.
Retail store analysis AI workflow lets developers create complete retail vision AI applications for store analytics by creating customized dashboards. Developers can use and customize the workflow to suit a variety of data analytics, including queue analytics such as shopper occupancy, shopper count, the duration of a stay and the trajectory of the customer’s journey lines crossings, proximity, and areas of importance.
These attributes are easily modified to contain specific information regarding use cases specific to the individual stores. Stores can use this information to enhance staffing, improve store layout and merchandising, and enhance customer experiences to increase sales.