BitFlow recently announced that it integrated the NVIDIA Jetson AGX Orin module, which can perform 275 trillion operations per second, with the “lightning-fast” data rates of its CoaXPress (CXP) frame grabbers.
The platform is not only cost-effective but also gives engineers the power to prototype complex machine vision and autonomous inspection applications, leveraging AI-accelerated image processing while, at the same time, supporting up to four CXP cameras and a multitude of concurrent AI application pipelines. This makes it possible for new applications to be more easily developed and move past “pass/fail” to tasks such as object detection and image classification and segmentation.
“Once proof-of-concept is established, a production model can move forward utilizing an Advantech AIR-030 AI Inference System Box featuring PCI Express x16 and based on the NVIDIA Jetson AGX Orin.” As a result, the time to market and related development costs are greatly reduced.
The BitFlow CXP frame grabbers are connected to the Jetson AGX Orin through a built-in x16 PCIe slot. Image data can be transferred at speeds of up to 50GB per second from CXP cameras to the NVIDIA Ampere GPU architecture. “BitFlow CXP frame grabbers DMA directly into the embedded GPU memory for image capture, pre-processing, and machine learning inference, shifting the load from the host computer to avoid CPU overhead.”
The CoaXPress interface makes it so that one cable can carry all data, triggering, control, and a maximum of 13W of power to connected cameras that are as far as 100 meters. The integration between the NVIDIA Jetson AGX Orin and the BItFlow frame grabbers is through the former company’s Linux AArch64 SDK. Because the SDK is universal, Bitflow’s entire line of CXP frame grabbers (Cyton and Claxon) are supported as well as the company’s Axion Camera Link line.