Embedded SBC plays AI reasoning

Jacktang

Embedded SBC plays AI reasoning [Copy link]

OPEN Al LAB, Arm China, and Rockchip jointly launched the EAIDK-310 artificial intelligence development kit single-board computer. This article uses the EAIDK-310 single-board computer to play AI reasoning.

Introduction:

EAIDK stands for Embedded Artificial Intelligence Development Kit. It is the world's first artificial intelligence development platform using the Arm architecture. It is specially designed for AI developers and is an artificial intelligence development kit for edge computing.

Hardware parameters:

EAIDK-310 is an ordinary embedded ARM single board computer using Rockchip's ARM SoC chip RK3228H. The detailed hardware parameters are as follows:

Software Platform:

The features of the EAIDK-310 development kit are almost all concentrated on the software aspect. The pre-installed system of the development kit integrates the embedded AI development platform AID, which includes three parts:
1. Embedded deep learning framework Tengine
2. Embedded computer vision acceleration library BladeCV
3. Heterogeneous computing library HCL.
The details are as follows:

From the parameter point of view, Tengine supports NPU units, and is the underlying hardware-level optimization, supporting multiple models.
BkadeCV is also the underlying hardware-level optimization, HCL heterogeneous computing acceleration.

The embedded AI development platform is a set of AI development platforms completely customized for ARM low-power SoC, with underlying hardware-level optimization, heterogeneous computing, multi-model support, and NPU support.

System Image

The EAIDK-310 development kit is pre-installed with the Fedora 28 system, which includes the above-mentioned embedded development kit, and provides system image downloads and burning tools. You can use it with confidence without worrying about the system crashing and being unable to recover, no backup system, and AI development platform installation.

---

Power on:

The EAIDK-310 development kit only needs to be connected to an HDMI display and an Ethernet cable, and then it can be powered on.

There are requirements for the power supply here. For example, the hardware parameter specifications above require a power supply of 5V2A. In actual use, at least the USB3.0 interface must be used for power supply. The power supply capacity of the USB2.0 interface is too small and cannot be turned on.

If everything goes well after booting up normally, the screen will first display the startup information:

Soon you will enter the system login interface:

Very sci-fi, very AI theme, EAIDK exclusive system theme, the username and password are both openailab, you can log in and use it normally, the system is fedora 28

Embedded AI experience:

1. Run a classifier first:

identify the cat as a fox, an ocelot, an Egyptian cat, chihuahua


clothes, hairstyles, and characters. The classification is quite accurate.

2. Connect the USB camera and get the video:
This application uses codec

3. Face recognition
EAIDK-310 development board needs to be connected to an HDMI monitor. It cannot be tested outdoors and can only be tested indoors. Test method:
1. Play the video on the computer iQiyi player
2. The USB camera of EAIDK-310 captures the computer screen and recognizes the face in the iQiyi video


The mtcnn used for face recognition is a multi-task convolutional neural network, mainly for face recognition. Feature point identification:
After the face is recognized, it is framed and 5 points of the eyes, nose and mouth are traced.

It takes 245ms to identify three people in a video, which can reach 4 fps

The camera takes a static picture, recognizes the face, and uses the detection model or mtcnn


to recognize a person in a static picture. It takes 170ms, which is about 6 fps.

---

Summarize:

The EAIDK-310 development board did not experience any freezes, crashes, or interruptions during the AI reasoning and recognition process, and achieved AI reasoning, object classification, face recognition, etc.:

From the frame rate, you may feel that it is a bit slow, but based on such hardware performance parameters, it is very satisfactory and unexpected to achieve this effect. The EAIDK-310 development kit is not about how good or powerful the hardware performance is, but more about feeling the power of this embedded AI software development platform, which can run in a low-power ARM SoC single-board computer, and achieve AI reasoning functions with lower energy consumption and lower hardware thresholds.