导言

FPGA 技术能够以低功耗和低延迟实现复杂的神经网络，同时连接大量外设并提供对工业应用非常重要的高水平鲁棒性，因此正在成为嵌入式人工智能应用领域的主要参与者。

客户挑战

In this case, Enclustra was their own customer. The challenge was to explore the potential of FPGAs in embedded AI applications and showcase it through a demo system.

解决方案

Based on the Mars XU3 module, featuring an AMD Zynq™ UltraScale+™ MPSoC device, mounted on the Mars ST3 base board, the application employs popular neural networks like resnet50 and SSD for image classification and real-time face detection, respectively. The images are captured with a standard USB camera, connected to the Mars ST3 base board. For higher performance a MIPI interface can be used, also available on the Mars ST3. The live image with added overlays can then be viewed on a DisplayPort-capable monitor. Moreover, adding actuators such as BLDC or stepper motors is a straightforward task using Enclustra’s Universal Drive Controller IP Core.

结果

Enclustra successfully deployed an AI-powered embedded real-time image processing application to run on Enclustra’s own SoC Module, which now serves as a demo system.

Keywords

AMD Zynq™ UltraScale+™ | VHDL | Mentor Graphics ModelSim® | DNNDK | C++ | Linux | Mars XU3 | Mars ST3 | resnet50 | SSD | USB | DisplayPort | MIPI