Utilizing AI Kit

In this tutorial, we'll walk you through setting up your AI Kit and running example applications for object detection and human pose estimation using the Raspberry Pi 5 (RPi5).

PreviousTeleoperation NextSerial Commands

Last updated 9 hours ago

Utilizing AI Kit

In this tutorial, we'll walk you through setting up your AI Kit and running example applications for object detection and human pose estimation using the Raspberry Pi 5 (RPi5).

Prerequisites

Before you begin, ensure the following:

You have a Raspberry Pi 5 (RPi5) with the OS installed and booted successfully.
Your RPi5 is connected to your Wi-Fi network.
VNC is enabled, and you're using a VNC client (like TigerVNC) to access the RPi5 remotely.

For help with installing the OS or enabling VNC, refer to the section of this documentation.

About the AI Kit Installation

If you purchased a complete Hackerbot with an AI Kit (AI, AI Pro, or AI Elite Edition), the AI Kit comes pre-installed. If you purchased the AI Kit separately, follow the hardware installation instructions .

USB Camera Requirement

Make sure a USB camera is connected to the RPi5. Most Hackerbot kits come with a global shutter camera suitable for the examples in this tutorial.

Step 1: Connect and Launch Terminal

Connect to your RPi5 using TigerVNC. Once connected, open a terminal window. All following steps assume you're working from this terminal.

Step 2: Update the System and Firmware

It's important to run the latest software to ensure hardware compatibility and receive security updates.

Update OS Packages

Run the following to update all software packages to their latest versions:

sudo apt update && sudo apt full-upgrade

Check Your Bootloader Firmware

The firmware controls how the RPi5 interfaces with connected hardware.

sudo rpi-eeprom-update

Check the output. If the date listed is earlier than December 6, 2023, you will need to update your firmware.

Update the Bootloader

Launch the configuration tool:

sudo raspi-config

Navigate to: Advanced Options > Bootloader Version
Select Latest
Press Finish or Esc to exit the config tool

Then, apply the firmware update:

sudo rpi-eeprom-update -a

Reboot the system to apply changes:

sudo reboot

After rebooting, reconnect using TigerVNC and open a new terminal.

Step 3: Enable PCIe Gen 3.0

The AI Kit uses the PCIe interface to communicate at high speed with the RPi5. Enabling Gen 3.0 mode allows full bandwidth operation.

Run the following:

sudo raspi-config

Go to Advanced Options
Select PCIe Speed
Choose Yes to enable PCIe Gen 3.0
Exit and reboot:

sudo reboot

Note: This reboot may take a few minutes. Be patient and reconnect with TigerVNC once it's ready.

Step 4: Install Hailo Software and Dependencies

The Hailo AI accelerator requires software support to work with the RPi5.

Install the complete package:

sudo apt install hailo-all

This may take several minutes. Once complete, reboot again:

sudo reboot

After reboot, open a new terminal and verify that the device is detected:

hailortcli fw-control identify

Expected output (details may vary):

Executing on device: 0000:01:00.0
Identifying board
Control Protocol Version: 2
Firmware Version: 4.19.0
Board Name: Hailo-8
Device Architecture: HAILO8L
...

If you receive output like the above, your Hailo hardware is successfully recognized.

Step 5: Install Example Applications

Clone the official example repository from Hailo and run the installer script:

cd ~/Documents/
git clone https://github.com/hailo-ai/hailo-rpi5-examples.git
cd hailo-rpi5-examples
./install.sh

This installs scripts and models that demonstrate real-time AI inference using the Hailo chip.

Step 6: Run Example Pipelines

Each time you reboot the RPi5 or change Python environments, you'll need to source the environment setup script:

cd ~/Documents/hailo-rpi5-examples
source setup_env.sh

Now you can run the example applications. Make sure your USB camera is connected and recognized as /dev/video0.

Object Detection

python basic_pipelines/detection.py --input /dev/video0

This runs a real-time object detector and displays bounding boxes for detected objects.

Human Pose Estimation

python basic_pipelines/pose_estimation.py --input /dev/video0

This tracks and overlays human skeletons on detected individuals.

Instance Segmentation

python basic_pipelines/instance_segmentation.py --input /dev/video0

This performs segmentation by coloring each detected object with a unique mask.

Next Steps

Interested in training on your own dataset? Check out the retraining guide from Hailo:

Reference Links