Introduction
Smart retail solutions come in many forms, but all share a relatively similar objective – to provide a more positive experience for customers. A popular convenience store recruited AAEON to help them create an AI inference-based automated self-checkout that could reduce the time customers spent queuing and deliver a more convenient and streamlined way to shop.
There were several advanced functions and non-standard features that the convenience store chain required of the application, and so they tasked AAEON with developing a tailor-made embedded solution, knowing that AAEON had the technical knowhow and design flexibility required to deliver it.
When the Prerequisite is Excellence
Because the convenience store chain had a stellar reputation as a brand that delivered both quality and consistency across all aspects of its business model, they needed its self-checkout to be exceptionally fast, but also accurate. As such, they required a product that could host multiple 3D cameras with high-bandwidth data transmission to obtain images of products passed through the checkout kiosk from multiple angles, without any latency.
In addition to this, the cameras chosen would need to pass the client’s rigorous testing criteria, including being able to withstand being continuously turned on and off over a period of seven days while maintaining a frame drop rate of below 0.1%, thus ensuring they would be able to obtain stable, high-resolution image data for AI inferencing purposes.
To obtain the most accurate image classification possible, the cameras selected needed to provide the application’s inference engine with an extremely detailed imaging dataset, and so the customer requested a system that could support cameras with extremely high bandwidth. It was decided that 3D cameras were the most suitable option for the client’s system, given both RBG and depth data streams could be transmitted and processed simultaneously, resulting in an increase in the total volume of data that the AI model could draw upon for product recognition purposes.
The high bandwidth offered by 3D cameras further solidified this decision, given they could ensure fast, latency-free data transmission while maintaining high frame rates. When combined, the high and wide bandwidth that 3D cameras could offer the application made them the perfect solution for multi-dimensional data capture and real-time image processing.
Due to the nature of the application, it goes without saying that the client required a product with the capacity to run complex AI inference models, able to perform object classification tasks in real time, identifying products as they were captured by the aforementioned 3D cameras. Once again, the focus of client was quality, and so only a product with the highest possible AI engine would be suitable.
While the speed, stability, and accuracy of the product’s image acquisition and AI inferencing capabilities were the most important of the client’s prerequisites, the product they selected would also need to have suitable environmental durability in order to operate in a retail environment.
Unique requirements required an innovative solution, and so the project led AAEON to develop the BOXER-8642AI, a Fanless Embedded AI System powered by the NVIDIA® Jetson AGX Orin™. The client was aware of AAEON’s reputation as a company that approaches its customer’s needs with flexibility, technical expertise, and a willingness to go the extra mile to make sure applications are market-ready. As such, it came as no surprise that the BOXER-8642AI precisely addressed each of the client’s requirements to optimize the application’s success.
Application Architecture
