Yesterday I published an article about the world’s first AI Edge camera powered by Raspberry Pi 3B+ and two Myriad X processors. This is a compact, ultra-low-power high-performance IP camera capable of running multiple deep learning models in parallel in real-time.
Raspberry Pi is a fantastic single board computer (SBC) but sometimes one might prefer an SBC with an Intel processor rather than ARM. One of the best SBCs in this category is UP Squared.
In this post, I am going to show you how you can turn a UP Squared into an extremely potent AI Edge Camera. While the Raspberry Pi AI Edge Camera has many benefits, the camera described in this article has its own strengths too. Since this camera is based on UP Squared with an Intel Atom processor, it is possible to use both Intel CPU and GPU to run deep learning inference. This means that you can run two models in parallel without any additional hardware. I highly recommend additional Myriad X VPUs to make this camera into an AI powerhouse! In my case, I have plugged in one AI Core X mini PCIe module featuring one Myriad X processor. In addition, I have plugged in two Intel Neural Compute Sticks 2.
I have used the following hardware:
- Dummy camera
- 2x USB Extenders
- 2x Intel Neural Compute Stick 2
- UP Squared
- UP HD Camera
- UP PoE Module
- UP Core X
This camera is very spacious so the whole project was a lot less challenging than the Raspberry Pi AI Edge Camera project. The only thing you need to do is to remove the top lid, unscrew the back cover and remove the dummy camera tray.
Remove everything from this tray so it looks like this.
Drill a hole through the middle so that the UP HD Camera lens can fit through. I did not have a drill but had a soldering station so I burnt one through 🙂 The picture below shows the fake lens.
This picture shows the UP HD Camera lens pushed through the newly made hole.
You will be left with the UP HD Camera body that you will need to screw in to the lens.
One you plug everything in it should look something like this. In this picture you do not yet see the PoE module connected but check this article to see how it should be done.
Then I used a thermal glue gun to fix the board to the metal tray.
Once the board was fixed the tray could be slid back to the camera.
And this is all there is to it. Now you have a super powerful AI Edge Camera capable of running multiple deep learning models (or other stuff) in parallel and in real-time (~40FPS).
I hope you find this blog useful and if you need any help with the code, which I am not publishing here, get in touch!