Spot the difference between “pie” and “pi” with the new Raspberry Pi AI Camera
Oct 6, 2024
David Prochnow
Our resident “how-to” project editor, David Prochnow, lives on the Gulf Coast of the United States in Ocean Springs, Mississippi. He brings his expertise at making our photography projects accessible to everyone, from a lengthy stint acting as the Contributing How-To Editor with Popular Science magazine. While you don’t have to actually build each of his projects, reading about these adventures will contribute to your continued overall appreciation of do-it-yourself photography. A collection of David’s best Popular Science projects can be found in the book, “The Big Book of Hacks,” Edited by Doug Cantor.
Share:
Everywhere you turn today, it’s “AI this” and “AI that,” in the news. If you’ve had a hankering to try your hand at “AI something,” then Raspberry Pi Limited has come to your rescue. The new Raspberry Pi AI Camera provides you with an experimental platform for discovering the fundamentals of image recognition processing. Don’t worry, though, Raspberry Pi Ltd provides you with enough hand-holding guidance that you’ll be expertly spotting furniture and human beings in no time. This new AI camera is not to be confused with the $65 or so Arducam, which also works with a raspberry Pi.
In the case of AI Raspberry Pi-style, however, no programming is necessary for implementing a couple of ready-to-go examples. Using a 12-megapixel CMOS Sony IMX500 Intelligent Vision Sensor that provides a 76-degree field of view, this manually-focused camera module shares the same size and shape as the Raspberry Pi Camera Module 3. This similarity enables the new AI camera to be a drop-in replacement on any Raspberry Pi 4 Model B and Raspberry Pi 5 single-board computer. In fact, suitable interface cables for both computers are included with the Raspberry Pi AI Camera.
Oh, and those previously mentioned examples, are painstakingly explained on the Raspberry Pi Documentation page. Basically, you are given two neural network models. One example provides object detection, while the other example demonstrates pose estimation.
In either video or still image format, a floating bounding box is displayed around the target subject of each example. Once you’ve successfully implemented these two examples, you can experiment with image classification and object segmentation examples in the Picamera2 GitHub repository .
While these are significant technical achievements for the tiny Raspberry Pi AI Camera, the Raspberry Pi computer, itself, isn’t labored with any complex tasks. Basically, the Sony IMX500 sensor does all of the heavy AI lifting in the Raspberry Pi AI Camera. All of the major neural network AI work is performed inside the sensor. This off-loading to the sensor allows the Raspberry Pi computer to perform other user-defined tasks related to the received image data. So let the DIY AI revolution begin!
You can pre-order your Raspberry Pi AI Camera today for $70 from Sparkfun .
In the end, remember, if it’s too good to be true, then it must be AI. I’d love to see this new camera incorporated into Raspberry Pi cameras like PiBZ or the Pi Zero camera.
Enjoy.
David Prochnow
Our resident “how-to” project editor, David Prochnow, lives on the Gulf Coast of the United States in Ocean Springs, Mississippi. He brings his expertise at making our photography projects accessible to everyone, from a lengthy stint acting as the Contributing How-To Editor with Popular Science magazine. While you don’t have to actually build each of his projects, reading about these adventures will contribute to your continued overall appreciation of do-it-yourself photography. A collection of David’s best Popular Science projects can be found in the book, “The Big Book of Hacks,” Edited by Doug Cantor.
Join the Discussion
DIYP Comment Policy
Be nice, be on-topic, no personal information or flames.