Embededded GPGPU Platform

Vision Box AI

  • Vision Box PC with integrated NVidia AGX Orin GPGPU
  • Appropriate for vision applications requiring a multi-core ARM CPU plus GPU accelerator, e.g. for AI
  • Interface support for Camera Link and other vision relevant technologies
  • Long-term availability, low power consumption, compact size

Today’s machine vision applications require a powerful CPU for traditional image processing algorithms. Combining this with suitable AI functions further improves the performance of an application. IMAGO´s “Vision Box AI” offers impressive technical features, all packed in a small form factor:

For standard applications, up to 12 ARM CPU cores with access up to 64 GB of RAM are available. Tasks can be distributed, and there is sufficient memory for image data, even when acquired by very high-resolution cameras.

For AI and other applications that require GPU computing power, up to 2048 CUDA cores are available on the computer. This computing power is equivalent to the performance of GPU cards installed in specially designed industrial PCs not so long ago.


A powerful CPU and GPU is not the only important factor for a successful image processing application – it also needs vision relevant interfaces. The Vision Box AI is a GPGPU computer that meets all requirements for practical use: It offers ports for GigE and Camera Link cameras, real-time capable IO (Real Time Communication Controller) as well as an encoder interface to support incremental encoders. Especially for line scan and hyperspectral applications, Camera Link still is a powerful real-time interface with an attractive price-performance.


The Vision Box AI presents its versatile functions in an industrial, fanless housing with low-power consumption.

Due to its powerful features, the Vision Box AI is an excellent choice for use in the following areas of application:

Embedded vision applications using IMAGO’s ViewIT software including the Halcon library inside.

Deep Learning programs: Because of its very powerful GPU, the Vision Box AI can run both the learning program and the inference program of an AI solution. You can develop your own AI solution or use off-the-shelf AI software such as “Oròbix’AI.go” supplied by our partner Oròbix.

Hyperspectral Imaging: This technology uses SWIR-based cameras that can acquire images in wavelength areas up to 1,700 nm that are far beyond the spectrum humans can perceive. Working in this wavelength range opens up whole new possibilities for detecting defects in materials and ensuring the quality of products. The volume of data captured with hyperspectral cameras significantly exceeds that of conventional image processing applications, therefore a CUDA-based accelerator is required to manage the amount of data collected and to calculate results with the corresponding algorithms. The Vision Box AI´s GPU accelerator offers the computing power needed to get the job done. Due to this feature, the Vision Box AI is also a powerful option for the food industry, where the ability to take more reliable decisions with the help of self-learning AI programs is becoming increasingly important.

Light-field cameras: This type of camera is mainly used in 3D applications and generates many images of the same object that need to be merged. The GPU accelerator used in the Vision Box AI can handle this task.

GPU accelerator enables AI-based image processing and other applications 
A powerful 12-core Linux computer enables you to develop your application by using a flexible operating system and CUDA operators running on the GPU.

For industrial equipment, availability is a precondition to success. Compared to IMAGO’s “AGE-X” Vision Box series using commercial graphics cards and off-the-shelf PCIe GPU accelerators with a short lifetime, the GPGPU integrated in the Vision Box AI is available for 10 years.

Developing an application with the Vision Box AI is easy: Just connect your camera, the sensors, the PC environment for Linux applications via the Ethernet interface, and you can start. All IOs are managed by IMAGO´s real-time communication controller, which is included in an SDK for Linux OS.

Vision interface support 
The first type of Vision Box AI serves up to four GigE interfaces for connecting standard GigE cameras. More GigE ports are not useful, as additional GPU power would be required for systems using more than four cameras.

The Camera Link interface has been established as a robust image processing interface for many years. It serves the highest demands in terms of fast image transfer. Camera Link provides the data in real time – an advantage for time-critical applications.

Real-Time IO: There are applications that require a hard real-time operating system, e.g. when peripherals have to generate output data in time, based on incoming information and a pre-set logic, to guarantee precise process timing. Especially for fast machines, IMAGO’s real-time communication controller and GPU processing are key features for secure process timing.

HDMI output: a display can be directly connected via a HDMI interface to manage your software.

Incremental encoder input: Many machine vision systems in industrial production lines are used at conveyor belts. To make sure the line scan or area scan cameras are perfectly synchronized with the conveyor belt speed, encoder signals need to be processed and aligned with the camera trigger to acquire high quality images. IMAGO´s real-time communication controller provides numerous logical functions to guarantee the reliability of the image processing systems used.

Long-term availability, low power consumption, compact size 
Long-term availability, low power consumption and compact size are important when designing image processing systems for industrial use. The Vision Box AI, as many other IMAGO products, is available for 10 years and uses less than 50W in operation. With a compact housing of just 230mm x 163mm x 67mm, the image processing computer Vision Box AI is suitable even in applications with limited space.

Food & Beverage
In the food and beverage industry, there are emerging solutions based on the principles of Deep Learning or non-visible spectrum (NIR) analysis. Since both applications require incredible computing power, Vision Box AI is the solution. It can be integrated into compact cabinets, and no additional fans are required.
Further sample applications

In pharmaceutical production processes, AI-based solutions combined with proven algorithms are a new trend. Thanks to Linux OS, the innovation requires less space and validation processes become easier.
Further sample applications

In manufacturing, there are many new trends that require CPU and GPU computing power. In addition to AI, for example, a multi-camera solution can use the GPU as an accelerator. Light-field camera solutions create a 3D image from multiple views and require GPU power. In addition to energy-saving solutions, reducing the space required in the control cabinet also becomes important.
Further sample applications

Packaging & Logistics
In the packaging and logistics industry, processes run very fast, with many cameras observing packaging that require additional GPU power as an accelerator. Systems need to become smarter to detect broken packages or non-compliant cartons, for example – a field for AI-driven analytics.
Further sample applications

For developers who are familiar with the Linux operating system, want to use libraries for classical machine vision and AI, Vision Box AI is a performance-strong solution.

Christoph Siemon
Sales Manager

Telephone: +49 6031-684 26 13

Sales Manager Fabian GarbsDipl.-Phys. Fabian Garbs
Sales Manager

Telephone: +49 6031-684 17 84