AI Model Deployment Guide

1. Model Deployment Overview

This document describes how to deploy a YOLOv5 model to an RK1126B board and run a demo program on the board. This model is a simply trained sample model and does not guarantee actual detection accuracy.

2. Preparation

Prepare the board, connection cables, network connection, and the PC-side compilation environment.

Use a tool such as MobaXterm to operate the board through SSH remote login or serial debugging.

2.1 Hardware Preparation

Prepare the RV1126B board, a Type-C cable, and a LAN cable. Connect the Gigabit Ethernet port to a router or switch, and connect the serial interface if necessary.

Network connection:

Serial connection:

2.2 Development Environment Preparation

When performing this procedure for the first time, set up the RV1126B compilation environment according to the Getting Started Guide. In the PC-side Ubuntu environment, run the run script to enter the compilation environment.

cd ~/develop_environment
./run.sh 2204

When connecting to the board via wired LAN, confirm the board IP address. This manual requires the program to be executed on the board.

Board IP address: 192.168.11.85

load_rootfs 192.168.11.85

developer@easyeai-compiler-2204:/opt/linuxshare/work/rv1126b/jp/AI/demo/yolov5_detect_C_demo$ ls /mnt
bin boot dev etc freq_set_status home lib lost+found media mnt oem opt proc root run sbin srv sys tmp userdata usr var

If the board is not connected, mount the rootfs.img used for writing.

Firmware download link for writing:

https://dl.dragonwake.com/download/rv1126b/embedded/images/CSUN-RV1126B_gstreamer-Ubuntu22.04-firmware_20260110.7z

After extraction, the image file rockdev/rootfs.img is available.

developer@easyeai-compiler-2204:/opt/linuxshare/work/rv1126b/jp/embedded/images/CSUN-RV1126B_gstreamer-Ubuntu22.04-firmware_20260110/rockdev$ load_rootfs rootfs.img

3. Source Code Compilation

Download the YOLOv5 C Demo sample file, move it to the Ubuntu environment, and extract it. After extraction, enter the corresponding directory in the RK1126B compilation environment and run the build script.

YOLOv5 demo download link:

https://dl.dragonwake.com/download/rv1126b/AI/demo/yolov5_detect_C_demo.tar.bz2

cd /opt/linuxshare/work/rv1126b/jp/AI/demo
tar -xvf yolov5_detect_C_demo.tar.bz2
cd yolov5_detect_C_demo
./build.sh
cp yolov5_detect_demo_release/ /mnt/userdata/ -rf

Note

During compilation, the dependent libraries are located on the board OS side. Refer to the Getting Started Guide and mount rootfs.img to /mnt. If it is not mounted, a compilation error will occur.

4. Run the YOLOv5 Demo on the Board

Log in to the development board through serial or SSH, and move to the directory where the demo program is located. Grant execution permission and then run the demo program. After execution, the detection result image result.jpg is generated.

cd /userdata/yolov5_detect_demo_release/
chmod 777 yolov5_detect_demo
sudo ./yolov5_detect_demo

The following figure shows the execution result after logging in to the board via SSH using MobaXterm. The algorithm execution time is 49 ms.

Check the test result image in the RV1126B Docker development environment on PC Ubuntu.

1. Copy the test image from the board back to the Docker development environment.

cp /mnt/userdata/yolov5_detect_demo_release/result.jpg .

2. The test result is shown below.