5 Raspberry Pi AI Kit Projects For Every User Level

Raspberry Pi offers lots of different add-ons and HATs (Hardware Attached on Top) for their $50 computer, including a desktop kit, a Build HAT for connecting to Lego motors, and a TV HAT for receiving digital television streams. You can also buy third-party Raspberry Pi kits that enable you to build anything from a camera to a handheld games console. There's also a HAT for incorporating AI into your Raspberry Pi projects, which was released in June 2024; we previously covered how it works and what things you can do with it if you're planning to build AI projects.

Since the Raspberry Pi AI Kit was released, it's been updated with a newer model, the Raspberry Pi AI HAT+, released in October 2024. The AI HAT+ is available in two versions. You can choose between 13 TOPS (tera-operations per second) with a Hailo-8L chip for $70 — which is the same as you got with the AI Kit — or 26 TOPS with a Hailo-8 chip, which costs $110 and gives you twice as much oomph. The kits use a Hailo AI acceleration module for object detection, image segmentation, and pose estimation.

Raspberry Pi fans have built on this functionality to create a range of projects, including traffic monitors, cluttered desk alerts, and a warning system in case your boss is lurking behind you. Some of the projects below use the original AI kit but don't worry, they are fully compatible with each other. You can use the AI HAT+ for any project that would have used the AI Kit. For all the projects below, you'll need a Raspberry Pi 5 and camera, as well as a Raspberry Pi AI kit or AI HAT+.

VEEB projects created a simple Raspberry Pi AI device that lets you know if there's somebody behind you while you're at your desk. They were inspired by the monitor mirror of the 00s. The full instructions are provided on their GitHub page. If you want to create your own Peeper Pam, you'll need a Raspberry Pi 5, a camera model, an analog voltmeter (5V), a Pico W, two resistors (220 Ohm and 1K Ohm), a MOSFET (metal-oxide-semiconductor field-effect transistor) and a red LED.

This gadget works by using the AI kit's detection capabilities and setting the computer vision analysis so that only a specified type of object (in this case, people) will activate the meter. You can see in the video that Peeper Pam ignores the chair, table, and plant in the frame. The meter goes from 0 (no people) to 1 (certain that there's a person) but allows for an area of uncertainty in between. If the dial is hovering between 0 and 1, then Peeper Pam is 50% sure there's a person behind you.

You can also use the AI kit's computer vision to count how many objects it can see and perform an action if it exceeds a certain number. Australian company Core Electronics shows in this video how to set up an alarm to let you know when you've got too many empty cups sitting on your desk, which is handy for those of us who get too caught up in what we're doing to notice that our workspace has become completely overrun and a trip to the kitchen is called for. The instructions for the counting objects alert start at 13:58, but it's worth watching the whole video, as Jaryd at Core Electronics gives an excellent, easy-to-understand run down of all the Raspberry Pi objection detection features. There are also projects to make a proximity alarm if someone gets too close to a dangerous item (like Jaryd's resin printer) and a person detector similar to Peeper Pam.

The full instructions are on the Core Electronics website, and there's also a forum where Jaryd helps people who are having any issues setting it up. Jaryd's cup detector activates when there are more than three cups and deactivates when the number of cups in view goes back down to less than three. You can adapt the code to trigger at different numbers and look for other types of objects, too. As Jaryd says, "You can go ahead and set this to three cups or 10 horses or 50 cars or whatever your project calls for."

Here's another video from Jaryd at Core Electronics. This time, he's demonstrating things you can do with the Raspberry Pi AI HAT+'s pose estimation capabilities. Pose estimation identifies key points on your body, including your eyes, ears, shoulders, elbows, wrists, knees, and ankles. By identifying these angle points, it can track the shape and movement of objects. By combining this information, the software can tell whether someone is, for example, walking or sitting and can figure out what direction they're facing. It plots the coordinates on X and Y coordinates to see where all the key points are in relation to each other. Jaryd covers some slightly more advanced projects using pose estimation, like using your arm movements to control a robot arm (made out of blu-tack and Legos) and a Fruit Ninja-style game where you slice fruit with your wrists.

However, the first project on the video is both simple and impressively effective. It's designed to allow you to pause or restart a YouTube video while you're away from your desk by lifting your arms in the air. This works by setting the camera and AI setup to detect when your left wrist and right wrist are both above your nose. When it determines that your wrists are higher up in the Y axis than your face, it pauses YouTube. It's a useful little demonstration because it uses a pose you're unlikely to assume while doing other things. Obviously, if you spend a lot of time lifting boxes down from high shelves, you could use a different body pose to operate the controls. As before, Core Electronics provides full instructions on its website and a forum to discuss any problems you might be having.

This traffic monitoring project uses an app called Viam, which is a data capture and configuration software that connects to real-world devices like cameras, sensors, and motors. Viam is a paid product, but you can sign up and get a free month's worth of data without providing credit card details. These instructions show how you can use the Raspberry Pi AI Kit or AI HAT+ to make a live traffic monitoring dashboard. Both the 13 TOPS and 26 TOPS versions will work. You'll need to be able to set your camera up somewhere with a good view of the traffic outside for this to work.

The project uses Raspberry Pi's AI object detection functionality, which you'll need to configure so that it only looks for relevant objects, like cars, buses, and people. Then, you can use Viam's tele-operations to produce a real-time graph. This graph will update regularly and let you see whether the number of vehicles and people outside your house is increasing or decreasing. All the data is captured by Viam, and you can change the configuration settings if you want to play around with the stats.

This unstructured data processing demo project shows how to automatically store images from your camera in different places, including your Slack account, Milvus, an open-source vector database, and MinIO, an open-source object storage system. The video and instructions were created by Tim Spann, who worked for Zilliz, the company behind Milvus, so much of it is focused on vector database usage.

However, for beginners, the Slack feature is the most fun. Using Spann's code, you can set up your Raspberry Pi AI HAT+ to check when your camera detects something and then send a message to your Slack account. You could use it like a security camera. Perhaps somebody has been stealing your pens. With this setup, you could receive a Slack notification as soon as one of your pens gets moved.

This project uses the original Raspberry Pi AI Kit, so it only requires 13 TOPS and uses one of the included Hailo AI Python examples as a starting point. If you're interested in storing your images on a vector database, Spann also shows how you can automatically attach metadata like the type of object and confidence percentage to your files. This enables you to search the images using the metadata, so, for example, you could look for all the pictures with people in them.