Raspberry Pi AI Kit: artificial intelligence available as an add-on module

Raspberry Pi AI Kit: artificial intelligence available as an add-on module

On June 4, 2024, the Raspberry Foundation announced the availability of an optional kit intended for owners of the Raspberry Pi 5 board. This is Raspberry Pi AI Kita solution that takes AI processing to a much higher level.

Developed in collaboration with Hailothe kit offers an accessible way to harness the abilities of inference high performance and maximum energy efficiency in a wide range of AI applications. No data is shared online because the activity of computing occurs using exclusively the local resources.

What Raspberry Pi AI Kit is and how it works

Raspberry Pi AI Kit – marketed at a price of 70-80 euros – includes two main components, often used with Raspberry Pi devices. The first is a card HAT+ (Hardware Attached on Top) with M.2 connector for PCI Express devices, which includes mounting holes for different M.2 sizes (2230 and 2242). The second, the real protagonist of the offer, is an M.2 module that hosts a Hailo-8L AI accelerator. The latter must be connected to the HAT+ card using the PCIe connector.

Installed on a Raspberry Pi 5, the kit allows you to quickly develop complex AI applications. For example, you can activate skills artificial vision operating in real time with low latency and reduced power requirements.

State-of-the-art neural networks for object detectionsemantic and instance segmentation, pose estimation, and facial recognition (to name just a few examples) run entirely on the Hailo-8L coprocessor, leaving the Raspberry Pi 5’s CPU free for other processing.

Raspberry Pi artificial intelligence kit on Raspberry Pi 5

Main characteristics

The Raspberry Pi AI Kit offers performance equal to 13 TOPS (Tera Operations per Second) via a single-lane PCIe 3.0 connection running at 8Gbps. Suffice it to say that the NPU (Neural Processing Unit) of a processor Intel Core Ultra last generation (Meteor Lake) does not exceed 11 TOPS.

The system proposed by Raspberry, in fact, integrates with the other components offered by the foundation and is compatible with a wide range of cameras. The Hailo-8L accelerator can drive multiple neural networks on a single camera or enable the simultaneous use of a single neural network with two cameras.

Software and integration

One of the main challenges in creating applications Of artificial vision AI-based is the software complexity of integrating the camera subsystem with the AI ​​framework. Raspberry engineers have worked to simplify this process as much as possible.

The rpicam-apps application suite includes a template post-processing for integrating real-time inference onto the camera pipeline. Using Hailo Tappas post-processing libraries, you can fine-tune advanced AI-powered applications with just a few hundred lines of C++ code.

Picamera2 is a framework developed for managing cameras and cameras on Raspberry Pi devices, designed to be used with the Python library. The Raspberry team anticipates that Picamera2 will also soon allow you to take advantage of the same levels of integration.

How to try the Hailo-8L AI accelerator

L’software installation it’s very simple: just load some packages via the package manager aptreboot and get started with the AI ​​demos provided by the Raspberry Pi itself.

The first command to execute is sudo apt install hailo-all. This way you install the driver and the Hailo firmwareil middleware HelloRTthe core post-processing libraries Hi Tappasthe software rpicam-apps.

After reboot (sudo reboot), you can type the following statement to verify that everything is working correctly:

hailortcli fw-control identify

The return of a output “speaking”, confirms that the AI ​​Kit and its software dependencies are configured as required. Again, typing the following starts the camera connected to the Raspberry Pi. A preview window, which remains displayed for 10 seconds, is further confirmation that everything is ready:

rpicam-hello -t 10s

The following command allows you to clone the GitHub project which allows you to manage post-processing:

git clone --depth 1 ~/rpicam-apps

Now you can start object recognition, pose estimation, image segmentation and more. Some examples of the results that can be obtained are visible in the videos published on YouTube by Raspberry Pi. The instructions for proceeding are given in this official guide.

Leave a Reply

Your email address will not be published. Required fields are marked *