Raspberry Pi’s Camera Module and AI Kit, with their powerful AI accelerator modules, offer exciting opportunities in computer vision and machine learning. The combination of the Raspberry Pi platform’s versatility and robust AI capabilities unlocks new potential for creating innovative smart projects. From creative experiments to practical solutions such as smart pill dispensers, makers are leveraging these tools to push the boundaries of AI. Below, we explore several standout projects that showcase this potential, aiming to inspire your next endeavor.
Monster AI Pi PC By Jeff Geerling
Jeff Geerling’s ambitious project leverages the Raspberry Pi AI Kit to build a Monster AI PC equipped with eight neural processors. This system delivers an astounding 55 trillion operations per second (TOPS), outperforming processors from AMD, Qualcomm, and Apple Silicon.
Central to the setup is the AI Kit’s Hailo-8L chip, integrated into a 12× PCIe slot card via a PEX 8619 switch. The card connects to a Raspberry Pi 5 through a Pineboards uPCIty Lite HAT, supplemented by an additional power supply for the demanding processors.
The system utilizes Edge Impulse computer vision to monitor and count vehicles in real-time, using a wide-lens camera for broader coverage. Training a YOLOv5 machine learning model was essential to achieve accuracy, and detailed instructions for model training are available on the project page.
Accelerating MediaPipe Models By Mario Bergeron
Mario Bergeron tested Google’s MediaPipe framework, known for building machine learning pipelines for video and image analysis, on the Raspberry Pi AI Kit. His project demonstrates how the AI Kit’s Hailo-8L module outperforms standard TensorFlow Lite models on the Raspberry Pi 5.
The results show up to a 5.8× speed improvement, achieving 26–28 frames per second for single-hand detection and 22–25 frames per second for two-hand detection. Detailed insights and a Python demo application are available on the project page.
Shakhizat Nurgaliyev Created Safety Helmet Detection System
This project focuses on improving workplace safety by identifying whether workers are wearing helmets on construction sites created by Shakhizat Nurgaliyev using the YOLOv8 machine learning model with the Raspberry Pi AI Kit, the system achieves high-speed inference at 30 frames per second.
The project highlights firmware and driver optimizations for efficient operation. A comprehensive guide on the project page details the installation process and includes a dataset of 5,000 annotated images to assist in training similar models.
Peeper Pam AKA The Boss Detector By Martin Spendiff
Using AI-powered computer vision, this project can identify objects in real-time via a live camera feed. Created by Martin Spendiff and Vanessa Bradley, the system detects human presence, allowing users to determine if someone, such as a boss, is approaching from behind.
The setup involves two components – a Raspberry Pi 5 with a Camera Module and AI Kit handles image recognition and serves as a web server, while a Raspberry Pi Pico W, paired with a voltmeter, receives signals via web sockets. The voltmeter’s needle indicates the AI’s confidence level in identifying the object.
The two-part configuration also allows flexible camera placement, enabling monitoring of areas that are otherwise hard to view. The project’s GitHub repository offers code adaptations for additional uses, such as deterring unwanted visitors, including pigeons in window boxes.
Naveen’s New Car Detection and Tracking System
As a proof of concept, Japanese maker Naveen developed an automated solution for identifying and monitoring vehicles at toll plazas. This system ensures an accurate tally of vehicles entering and exiting the area, demonstrating the efficiency of AI inferencing on the Raspberry Pi AI Kit.
These projects highlight the power and versatility of Raspberry Pi’s AI capabilities, offering a glimpse into the endless possibilities for innovation. Whether you are an experienced maker or a curious beginner, these examples provide inspiration and practical knowledge to embark on your own AI-driven projects.