Optimizing AI Performance for IP Cameras with Google Coral Mini

Optimizing AI Performance for IP Cameras with Google Coral Mini

I have Frigate installed in Unraid, which is an amazing tool to connect IP cameras in your home, enabling 24/7 capture and analysis of footage for people detection. (Plus, I integrated it with Double-Take and CompreFace for facial recognition, but this will be discussed in a different post). You don’t need Unraid for this, but I find it convenient to centralize all such applications in one home server.

Initial Setup and Challenges

When I first started with Unraid, I realized that the CPU was struggling with the AI processing required for person recognition in the images. It was inefficient, so I quickly switched to a GPU. I used an Nvidia gaming card, and while the performance was good, the power consumption and heat generation were significant. The GPU consumed an additional 70-90W, even idling at no less than 30W, which was far from efficient.

The Switch to Google Coral Mini

Based on recommendations in the Frigate documentation, I switched to a Google Coral Mini. The official Google Coral website didn’t ship to my country, so I used AliExpress to get it. The Coral Mini comes with various connectors, and while the USB stick is popular, I opted for a model that connects to an M.2 connector or PCIe slot.

I ordered the Google Coral Mini PCIe M.2 Accelerator TPU B+M Key from AliExpress. Make sure to choose the longer B+M version, not the A+E version. Additionally, to connect it to a PCIe slot in my computer, I also got this M.2 NVMe SSD to PCIe x1 Adapter Card. I chose the PCIe x1 over the x4 version.

Setup and Configuration

This setup worked wonderfully. The Coral Mini was easily detected by Unraid and Frigate after installing the necessary drivers in Unraid, which you can get by installing the “Coral Accelerator Module Drivers” from the Unraid Apps tab. In my Frigate configuration, I ensured the following section was included:

detectors:
   coral:
    type: edgetpu

Performance and Efficiency

The Coral Mini uses only 2W of power and is entirely passive, meaning no noisy GPU fans or excessive heat. It outperforms the GPU in terms of speed and efficiency. Moreover, the Coral Mini is much more cost-effective than a GPU, making it the clear choice for AI processing tasks in a home server setup.

Summary

Switching to the Google Coral Mini significantly improved the efficiency and performance of my Frigate setup on Unraid. It drastically reduced power consumption and eliminated the noise and heat issues associated with using a GPU. This setup is not only more efficient but also more cost-effective, providing a perfect solution for AI processing in home automation.

Where To Buy


Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply