Notifications
Clear all
Topic starter
2025-02-17 8:50 pm
Check out two artificial intelligence cameras for the Raspberry Pi - the Raspberry Pi AI Camera and the ArduCam PiNSIGHT.
Today, we will examine an exciting new category of edge computing devices: the AI Camera. These cameras process an AI model at the source and deliver AI metadata to the Raspberry Pi host. They can also be used as conventional cameras.
Today, we have two cameras to play with. The first is the Raspberry Pi's AI Camera module. Although it resembles the standard Pi camera, it has an intelligent sensor from Sony and an RP2040 microcontroller to manage data transfer.
The second unit is the ArduCam PiNSIGHT. This complete autofocus camera interfaces using a USB-C connector. You can mount a Raspberry Pi board directly on the camera, and it has a standard ¼-inch tripod thread for easy mounting.
In this video, we’ll set up and test both cameras and see how they perform. We’ll also discuss the merits and drawbacks of each camera to help you decide which would be better suited to your next intelligent project.
Here is the Table of Contents for today's video:
00:00 - Introduction
01:50 - AI Cameras
05:54 - Raspberry Pi AI Camera
10:16 - Raspberry Pi AI Camera Setup
13:22 - Demo 1 - rpicam-apps
15:41 - Demo 2 - Pose estimation
16:41 - Demo 3 - Picamera2
19:52 - ArduCam PiNSIGHT AI Camera
22:37 - ArduCam PiNSIGHT Setup
25:13 - Demo 1 - depthai demo
27:18 - Luxonis Repository & OCR
31:28 - Camera Comparisons
36:57 - Conclusion
These cameras are sure to inspire some incredible projects. By doing most of the AI work on the camera, your Raspberry Pi board is free to perform other tasks. You can also use them with a lesser-powered Raspberry Pi board, like the Pi Zero 2 W.
I hope you enjoy the video!
Bill
"Never trust a computer you can’t throw out a window." — Steve Wozniak
2025-02-17 11:29 pm
I've got the rpi AI cam and the provided examples that are available show what can be done as illustrated in Bills video, but to get your own AI recognitions learnt and loaded does take quit a bit of learning, and in the short time I have had the camera I have not got that far. Its the sort of project where one needs to spend a fair bit of time to get to understand it all, and as usual Bill ( @dronebot-workshop ) has given an excellent intro which I certainly would have benefited from when I had my initial play a couple of months ago.
In case its of any use to anyone I did make a simple 3D printed stand for the rpi Cam and I give the stl files , though I expect theres a number of such examples to be found with a google. My version has a base with a slot into which I slot a rectangular camera mount. Its been useful to hold the camera whilst getting started with it. The stl files are attached as is a small pic of what it looks like.
THRandell and DroneBot Workshop reacted
Topic starter
2025-02-18 1:14 am
but to get your own AI recognitions learnt and loaded does take quit a bit of learning, and in the short time I have had the camera I have not got that far.
@byron Were you using the tools provided by Sony AITRIOS to build your own models? That would be my next step, as I have a robotics application that I think the RasPi AI Camera would be perfect for.
I'd be interested in hearing about your experiences with it. And thanks for the nice words about the video!
😎
Bill
"Never trust a computer you can’t throw out a window." — Steve Wozniak
2025-02-18 10:03 am
@dronebot-workshop Hi Bill, I did make a tentative start to read the docs on the Sony site -
https://developer.aitrios.sony-semicon.com/en/raspberrypi-ai-camera
but reading was as far a I got. Its probably also worth noting the Rpi forum for this camera:
https://forums.raspberrypi.com/viewforum.php?f=169
I got distracted from the AI camera with too many things on the go at the time as I got a new mac mini to set up plus some waveshare gps rtk boards, but now the camera is back on my desk I will refresh my memory on exactly where I got to. I have not actually trained it to recognise any of my own objects as yet. I remember being impressed when the rpicam-hello demo program was running and recognising objects like my monitor an keyboard even though they were at somewhat oblique angles to the cam, and then I stuck just my finger into the camera view and it recognised it as a person. I did 3D print an orange disk to make a start on seeing how I could get this simple object to be pick out by the camera but right now its just serving as a handy coaster for my coffee cup. But now is probably the time to have a proper go at seeing how far I can get and I will let you know of any progress I make.