Getting Started with Vision AI Kit 6490 / AI Demo

1. INTRO

About the board:

The Vision AI-KIT 6490 features an energy-efficient, multi-camera, SMARC 2.1.1 compute module, based on the Qualcomm QCS6490 SOC device.

(Learn more here: https://www.tria-technologies.com/product/vision-ai-kit-6490)

This board is excellent for multimedia and robotics applications. It can support up to 4 concurrent MIPI-CSI cameras, MIPI-DSI and MiniDP display outputs.

The kit includes:

• Carrier Board, SMARC compute module, and heatsink
• CSI IMX577 12MP camera
• 9V power supply

Prerequisites

To get started with Vision AI Kit 6490 you'll need:

• A Linux or Mac OS host PC (I'm using Ubuntu 22.04)
• A USB-C cable to connect the kit to your host PC
• A USB-TTL cable (I'm using an MCU-Link)

Optionally, you'll need:

• a display
• a keyboard/mouse
• An ethernet cable that's connected to a router or switch
• More cameras :) It supports up to 4 MIPI cameras

2. HARDWARE SETUP

To setup initial testing, refer to the notes below which index the tags on this board's photo above.

6. USB-C power connector: Once the above steps completed, connect the provided 9V, 3A “PD” power supply. (Note! This board will not function with a 5V supply!).

7) POWER button.

8) RESET and USER_B buttons: Difficult to access, avoid use of these buttons on the Rev.1 board.

3. LINUX HOST SETUP

We'll be using a Linux host in this tutorial to configure and flash the board.

You can also flash the board using Windows. Instructions for flashing using Windows can be found in the Demo section below.

Qualcomm loves their acronyms. Here is a short list that you can refer to:

Abbreviations and acronyms

QSC == Qualcomm Software Center

QPM == Qualcomm Package Manager

QUD == Qualcomm USB driver

PCAT == Qualcomm Product Configuration Assistant Tool

CDT == Configuration Data Table — it's a proprietary Qualcomm data structure that stores hardware-specific configuration for a board.

user@HOST:~$  :This command is run on the Linux HOST
sh-5.1#       :This command is run on the Vision-AI Kit 
$             :This command is run on the Windows HOST

In order to install user software from Qualcomm, you'll need a Qualcomm account. Register one here.

First install the Qualcomm Package Manager (QPM)

Install the Qualcomm Software Center (QSC) from the Qualcomm Package Manager (QPM).

After the QSC is installed, it will suggest some packages to install like the QUD and PCAT. Install these as well.

You can user the QSC to see what boards are connected, as well as install other software.

4. FLASH THE BOARD

A. Flashing Preparation

You can find more detailed instructions for flashing the board on Qualcomm's Linux Build Guide documentation.

1. Update the udev Rules:

Configure the udev USB rules for the Qualcomm manufacturing vendor ID 05c6 on the Linux host:

• Navigate to the udev USB rules directory and list the contents:

user@HOST:~$ cd /etc/udev/rules.d
$ ls

• If the 51-qcom-usb.rules file is not present, add the file with:

user@HOST:~$ sudo vi 51-qcom-usb.rules

• Add the following content to the file:

SUBSYSTEMS=="usb", ATTRS{idVendor}=="05c6", ATTRS{idProduct}=="9008", MODE="0666", GROUP="plugdev"

• If the file exists, then check for the previous content:

user@HOST:~$ cat 51-qcom-usb.rules

• Restart udev:

user@HOST:~$ sudo systemctl restart udev

• Now use the Qualcomm Device Loader (QDL) (instructions below) or Qualcomm Product Configuration Assistant Tool (PCAT) commandline tool to reflash the board.

Flash the board with the QDL Tool

• Download the QDL tool and unzip the contents of the downloaded folder. Qualcomm Linux 1.4 requires QDL version 2.3.1 or higher.
• To provide executable permission, inside the folder that matches your host OS (e.g. QDL_Linux_x64) run the following command:

user@HOST:~$ chmod +x <qdl_tool_path>/qdl_2.3.1/QDL_Linux_x64/qdl

Flashing CDT

Skip CDT Flashing steps if UFS flash memory on the target board has already been provisioned and it’s CDT configured for BSP 1.3.

• Ensure the Device is in EDL Mode:The device must be in Emergency Download (EDL) mode before proceeding with CDT flashing.
• Stop ModemManager Tool (for Linux distributions with systemd):

The ModemManager tool may interfere with the flashing process. Stop the tool using the following command:

user@HOST:~$ sudo systemctl stop ModemManager

Create a Download Directory:

user@HOST:~$ mkdir <cdt_download_path>
user@HOST:~$cd <cdt_download_path>

Download the CDT Binary:

user@HOST:~$ wget https://artifacts.codelinaro.org/artifactory/codelinaro-le/Qualcomm_Linux/QCS6490/cdt/rb3gen2-vision-kit.zip
unzip rb3gen2-vision-kit.zip

Navigate to the Download Directory:

user@HOST:~$ cd <cdt_download_path>

Flash the CDT:

• Connect a USB cable from the OTG/DL USB-C connector to the test computer
• Set SW1.1 = ON, power-up the board, wait 2 seconds, then
• Set SW1.1 = OFF, the board should now be in EDL mode

Use the Qualcomm Device Loader (QDL) tool to flash the CDT binary. Ensure you have the QDL tool downloaded and the correct path to the tool is specified in the command.

user@HOST:~$ <qdl_tool_path>/qdl_2.3.1/QDL_Linux_x64/qdl prog_firehose_ddr.elf rawprogram3.xml patch3.xml

Linux BSP Image Flashing Prerequisites

• Stop ModemManager: If you need ModemManager, you can restart it after flashing is complete.
• Ensure Device is in EDL Mode:Ensure the device is in EDL (Emergency Download) mode.

Install Required Modules: Verify that make and gcc are installed and available:

user@HOST:~$ make --version && gcc --version

If not installed, run:

user@HOST:~$ sudo apt update
user@HOST:~$ sudo apt install build-essential
user@HOST:~$ sudo apt-get install git libxml2-dev libusb-1.0-0-dev pkg-config

• Download the provided Tria custom QC Linux BSP 1.3 image qcom-multimedia-image-hdmi-test.zip and unzip the contents of the downloaded folder.

user@HOST:~$ unzip qcom-multimedia-image-hdmi-test.zip

B. Flash the Images, Boot and Connect to the board

How to Flash/Reflash the QCS6490 SMARC

• Connect a USB cable from the OTG/DL USB-C connector to the test computer
• Set SW1.1 = ON, power-up the board, wait 2 seconds, then
• Set SW1.1 = OFF, the board should now be in EDL mode

Navigate to the Directory where images are:

user@HOST:~$ cd <extracted zip directory path>/qcom-multimedia-image-hdmi-test

Replace <extracted zip directory path> with the actual path where the zip file was extracted.

Flash the Images:

user@HOST:~$ <qdl_tool_path>/qdl_2.3.1/QDL_Linux_x64/qdl prog_firehose_ddr.elf rawprogram*.xml patch*.xml

Ensure you use the correct qdl_tool_path based on your host machine architecture.

Boot the board

Set SW1.1 = OFF then power-cycle the board

Normal boot from UFS

Press + hold to power-up the carrier board and compute module.

After the image is flashed and booted you can connect to the board via Serial-UART or ADB interface.

Connect to the board via ADB

After the image is flashed restart the board and connect to it via ADB

Setup

1. Install ADB on the Linux Host

Run the following command to install the necessary tools:

user@HOST:~$ sudo apt install git android-tools-adb android-tools-fastboot wget

2. Verify ADB Version

Ensure the ADB version is correctly installed by running:

user@HOST:~$ adb --version
Android Debug Bridge version 1.0.39 Version 1:8.1.0+r23-5ubuntu2

3. Connect the Device

Using a USB Type-C cable, connect the Vision AI-Kit 6490 device to the host system.

4. List ADB Devices

Run the following command to list the connected ADB devices:

user@HOST:~$ adb devices
List of devices attached
68f592a         device

5. Enter the ADB Shell

To enter the shell, use the following command:

user@HOST:~$ adb shell
sh-5.1#

By following these steps, you should be able to establish an ADB connection with your QCS6490 device. If you have any issues, verify the connection and the ADB installation as outlined above.

Connect to the board via UART

Console Debug Headers

1) SER0 UART (on 12-pin header) - Connect pins 3, 4, 5 to TXD, RXD, GND of an external USB-to-serial cable adapter. eg. Inexpensive NXP MCU-LINK adapter, or other 3.3V compatible USB-serial adapter. Set port settings of your serial console to 115200 8N1. (This is the Linux console debug interface, the signals are level-shifted to 3.3V levels)

For a console interface, use of an inexpensive third-party USB-Serial adapter is recommended, eg. Use of VCOM port of an NXP MCU-LINK debugger is shown in the photo below.

Using Minicom (or other terminal application), monitor the Linux boot messages, then login using:

UserName = root

Password = oelinux123

Connecting to a WiFi Network

To connect to a WiFi network using the nmcli command, follow these steps:

• Connect to the WiFi Network:

user@HOST:~$ adb shell
sh-5.1#
sh-5.1# nmcli dev wifi connect QualcommWiFi password 1234567890
Device ‘wlan0’ successfully activated with ‘d7b990bd-3b77-4b13-b239-b706553abaf8’.

• Check the WLAN Connection Status and IP Address:

sh-5.1# ifconfig wlan0

• Ensure Connection is Active:

sh-5.1# ping google.com

Switch to a Different WiFi Network

If you are already connected to a network and need to reconnect to another network:

• Disconnect from Current Network:

sh-5.1# nmcli c down QualcommWiFi
Connection ‘QualcommWiFi’ successfully deactivated (D-Bus active path: /org/freedesktop/NetworkManager/ActiveConnection/1)

• Check Disconnect Status:

sh-5.1# nmcli -p device

Establish an SSH Connection

Ensure network connectivity is established before connecting to SSH.

• Find the IP Addressfor WiFi:

From the console CLI interface, enter ifconfig to list the board’s IP address for its Ethernet connection (assigned by the network router’s DHCP service):

sh-5.1# ifconfig wlan0

• Establish SSH Connection:

sh-5.1# ssh root@ip-address

Example:

root@qcs6490-rb3gen2-vision-kit:~# ssh root@192.168.0.222

Enter SSH Password:oelinux123

5. RUN THE DEMOS:

What makes this board so impressive is Qualcomm's software. There are dozens of high performance models that have been optimized to run on the QCS6490 available for download on the Qualcomm AI Hub.

Tria's Vision AI Kit demo for Embedded World 2025 showcases 4 of these models running concurrently.

AI Demo System Setup

2. CSI1 22-pin MIPI camera connector (4-lane, SMARC-compliant pinout):Note pin1 markers, then connect FPC ribbon cable from supplied Arducam IMX577 MIPI-CSI camera to CS1

5. MiniDP connector: Connector to HDMI display using miniDP to HDMI cable. Display: On Rev.1 hardware, only the MiniDP display output should initially be used. Use of DP1 for this interface means MIPI-DSI cannot be used at same time as MiniDP.

Demo Overview

The demo uses live video from two USB cameras, to display menu UI-selected concurrent AI inferencing models, while also charting live profiling info (resource-utilization and thermal measurements from QCS6490 CPU, GPU and LPDDR5 memory), all output via graphical UI onto a 4K HDMI screen.

Hardware Requirements

Note: the demo requires a 4K monitor

• VisionAI-KIT 6490 (plus SMARC heatsink)
• Computer mouse (wireless)
• 2x USB cameras, 720p or better.
• 4K HDMI monitor or TV (27-inch or larger recommended)
• miniDP to HDMI converter (active)
• HDMI to HDMI cable
• PD 9V DC power supply (65W PD supply preferred)

Software Requirements

The following software install is done on Windows 10 or 11:

• Install Qualcomm Software Center
• Qualcomm Profiler: Use Qualcomm Software Center to download and install Qualcomm Profiler (v2.25.1.6) onto a Windows PC
• Qualcomm USB Driver: Use Qualcomm Software Center to download and install the Qualcomm USB Driver (required by PCAT)
• PCAT: Use Qualcomm Software Center to download and install PCAT. Add PCAT.exe's parent folder to Path.
• Linux BSP1.3 image (includes gstreamer, psutil, pip3, etc) Custom TRIA Linux BSP 1.3 image 3/06 qcom-multimedia-image-hdmi-test.zip
• target-le installer: Install Qualcomm Profiler library APIs from the Windows PC to target board, by running InstallerLE.exe with the board powered-on and connected via USB ADB interface: C:\Program Files (x86)\Qualcomm\Shared\QualcommProfiler\API\target-le\InstallerLE.exe
• Download the Vision AI-KIT demo installer QCS6490-Vision-AI-Demo-Installer.run

Software Install Procedure

1) Skip over the Provision UFS if UFS flash memory on the target board has already been provisioned. Skip over CDT Flashing steps if it’s CDT has already been configured for BSP 1.3.

2) Download the provided Tria custom QC Linux BSP 1.3 image qcom-multimedia-image-hdmi-test.zip

3) Download the UFS (provision.zip) and CDT (rb3gen2-vision-kit.zip) files for QCS6490 from CodeLinaro.org

3) Make a staging folder C:\FLASH_qcs6490. Unzip qcom-multimedia-image-hdmi-test.zip, provision.zip, and rb3gen2-vision-kit.zip here.

Now that that has been set up, run the following commands in Windows Powershell on your Windows PC:

4) Find the device serial number. This will be used in the rest of the PCAT commands:

$ cd C:\FLASH_qcs6490 
$ pcat -devices

1. Provision UFS:

Put the device in EDL mode by making sure switches 1 and 3 are ON and the rest are OFF.

$ PCAT -PLUGIN SD -DEVICE <SERIAL NUMBER> -DEVICEPROG C:\FLASH_qcs6490\provision\prog_firehose_ddr.elf -MEMORYTYPE UFS -UFSPROV TRUE -UFSPROVXML C:\FLASH_qcs6490\provision\provision_1_3.xml

2. Flash CDT

$ PCAT -PLUGIN SD -DEVICE <SERIAL NUMBER> -BUILD "C:\FLASH_qcs6490\rb3gen2-vision-kit" -MEMORYTYPE UFS -FLAVOR asic

3. Flash Image

$ PCAT -PLUGIN SD -DEVICE <SERIAL NUMBER> -BUILD "C:\FLASH_qcs6490\qcom-multimedia-image-hdmi-test" -MEMORYTYPE UFS -FLAVOR asic

4. Install Qualcomm Profiler & Run Qualcomm Profiler target-le installer

Use Qualcomm Software Center to install Qualcomm Profiler on Windows PC

Once Qualcomm Profiler is installed on Windows PC, then run InstallerLE.exe (Vision AI-KIT board must be powered-on and connected to the PC via USB ADB interface).

• Select your device and select "Auto Configure."

Run Qualcomm Profiler target-le installer:

$ cd C:\Program Files (x86)\Qualcomm\Shared\QualcommProfiler\API\target-le\InstallerLE.exe

(Note! It takes a couple of minutes to complete the installation of the Qualcomm Profiler library files…) Download, adb push, change permissions, then launch Vision AI-KIT demo installer, on the target board…

This installs Qualcomm Profiler on the Vision AI Kit 6490 so it can provide system profiling to the application.

5. Run Vision AI Demo installer

$ adb push "C:\FLASH_qcs6490\QCS6490-Vision-AI-Demo-Installer.run" /tmp 
$ adb shell
sh-5.1# chmod +x /tmp/QCS6490-Vision-AI-Demo-Installer.run
sh-5.1# ./QCS6490-Vision-AI-Demo-Installer.run

Running this installer installs all of the necessary files under the /opt folder… (Changes will have also been made to weston.ini located in /etc/xdg/weston)

Running the Demo Application

Reboot the board. You will notice that the appearance of Wayland has changed, and there is now a Tria Launcher icon at the top left of the screen.

Launch the demo application using a single mouse click on the TRIA launcher icon:

Camera 1 and Camera 2 drop-down menus should now be visible, just below the title bar, and the Profiler measurement values and animated charts will be visible at the bottom of the screen (Use the CLOSE button at bottom right corner of this panel, to exit the INFO help screens mode)

Now select AI models to exercise, using mouse-pointer on the Camera 1 and Camera 2 drop-down menus…

Test the two Cameras

Vision AI Kit 6490 on Hackster Videos

http://avnet.me/qcs6490-demo-video2

What's next ??

Customizing the AI-demo!!