"As an edge AI system, the Vision Box AI is a versatile platform for the development and use of image processing and AI applications," explains Carsten Strampe, Managing Director of Imago Technologies. "It is suitable for customers who require high computing power, long-term availability and a compact design. The ability to use GPGPU-based systems in complex industrial environments gives users the opportunity to break new ground with embedded computing."

The fanless vision computer is based on the GPGPU 'Jetson AGX Orin'. It has up to twelve ARM CPU cores with access to a maximum of 64 GB RAM for standard applications. Tasks can be distributed and sufficient memory is available for image data, even if this is recorded by high-resolution cameras. The GPGPU achieves an AI-relevant computing performance of 275 TOPS (Tera Operations per Second, 8-bit integer) with low energy consumption.

The edge AI solution 'Vision Box AI' enables high computing power and low power consumption.

© Imago Technologies

The powerful GPU accelerator used in the 'Vision Box AI' enables a wide range of applications: it supports AI applications with a high process cycle or more complex networks. The 'Vision Box AI' can execute both the learning program and the inference program of an AI solution. The GPGPU (General Purpose Computation on Graphics Processing Unit) computer offers sufficient computing power for applications such as hyperspectral imaging or light field cameras that require a CUDA-based accelerator. "The 'Vision Box AI' is suitable for many customer groups in various industries," explains Strampe. "In the food and beverage industry, it can enable new quality control methods based on AI and the analysis of light in the non-visible spectrum (NIR). In the production, packaging and logistics industries, there are many new applications that require high computing power from the CPU and GPU. In addition to AI applications, multi-camera systems can also use the GPU as an accelerator."

The 'Vision Box AI' combines the GPGPU architecture with the interfaces required for industrial image processing. Standard interfaces such as real-time-capable digital I/Os, an RS-422 interface for connecting incremental encoders, Ethernet, USB and an optional monitor output are available. In addition to CameraLink, the Vision Box AI also offers up to four GigE interfaces for connecting standard GigE cameras.