Introduction

[Brian Logan]

I wanted to introduce myself. I am looking for a hobby project to do some systems level programming with Rust. I think Moonfire-NVR is a perfect balance as I was disappointed in the Linux alternatives to BlueIris and I'd like to get rid of my Windows NVR as it takes up considerable space and power, requires a GPU for performance, and produces extra noise and heat. 

I am able to run the same amount of cameras on a 2GB LXC Rocky Container with a similar level of streaming performance. 

I have 11 years of Sysadmin experience, and some hobbyist experience with Python, Javascript, and Golang. 

I saw that Microsoft announced Project Volterra and I think getting Linux ARM and Moonfire-NVR running on that device and taking advantage of the Neural Compute chip for doing AI detection would be pretty neat as a long term hobby goal.

I haven't experimented much with AI as well and think this would be a useful and productive way to learn while providing myself an actual funcitonal use case at my home that I can "experience" the end product as a user. 

But! I'm super excited to start interacting with the others here and see where I could try and be of most use. 

[John L. Poole]

Hi Brian,

Thank you for the introduction.  I'm unfamiliar with Project Volterra, but quickly looked and saw:

Microsoft hasn't provided any info on pricing either, but we can assume it'll come in under $1000.
source: https://www.windowscentral.com/hardware/laptops/surface/project-volterra-everything-you-need-to-know#section-project-volterra-price-availability

You might want to consider the Raspberry Pi "pimped out" with a a Western Digital Purple hard drive (USB hard disk array?) and a root system on a flash drive.  See https://github.com/scottlamb/moonfire-nvr/wiki/Hardware-Recommendations

I've been exploring the Google Coral Development Board (these can still be obtained), but ran into what I felt was a poorly staged system for testing with an older version of gstreamer.  I did some trials using the Google Jupyter Notebook at https://colab.research.google.com and then ran my own Jupyter in Ubuntu sitting on an AMD Bulldozer (circa 2013)  with no GPU and that taught me that a GPU/or like is essential for any AI work.  Then I turned focus on the Coral Dev Board and conlcluded the Coral processor offers at a very cheap price the horsepower I would need for edge detection of my video streams.  Google also makes a USB Plugin with the Coral Board, the problem is they are hard to find during this days of chip shortages.  I found Farnell in the UK (https://uk.farnell.com/pi3g/g950-01456-01/usb-accelerator-raspberry-pi/dp/3583033  -- note, the US site doesn't have the price/offering that the UK site has, so do not assume they are equal in the ability to procure and provide) offered a wait-list for the USB Coral and I am expecting a shipment to go out this month, June.  I'd encourage you to consider the USB Coral alternative for your AI because it would be something others could use, I don't know how many other people are going to want to commit near $1,000 for a new hardware offered by Microsoft.

My goal is to use the Coral USB on a Gentoo Linux based system with the high watermark of gstreamer.  I have a Rapsberry Pi4 with a Western Digital Purple and have 11 Reolink cameras, all at highest resolutions plus secondary feeds, running.  I really should write a page with my specifications... and writing this response will bring it to mind as a near-term TODO.

Cheers,

John

--
You received this message because you are subscribed to the Google Groups "moonfire-nvr-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to moonfire-nvr-us...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/moonfire-nvr-users/ec08e40b-73c6-406f-b29b-2f05e46e0547n%40googlegroups.com.

--

John Laurence Poole
1566 Court ST NE
Salem OR 97301-4241
707-812-1323

[Scott Lamb]

Hi Brian,

Welcome!

I hadn't seen Project Volterra. Looks like a nice device. fwiw, while this has been a pretty tough year for hardware purchases, if you want already-being-sold ARM/computer vision hardware in the same league, it's worth looking at the nVidia Jetson series. As far as cheaper stuff, I'm keeping an eye on the upcoming Rockchip RK3588-based boards. Even the Raspberry Pi4 is more capable than one might think when combined with a Coral USB Accelerator as John mentioned, although I'm not sure what the limit is in terms of how much H.264 decoding the Pi4 can do.

I'd also love to have analytics features in the project. I think that can be subdivided into a lot of parts (storage schema to represent the results, API to transfer them around, frontend work to display them well, and of course the actual analytics themselves). I think prototyping a processor that just churns through a .mp4 file and writes results in a text file or an overlay might be a fun way to get started. It might help us decide what a storage schema should look like.

There's also various smaller open bugs and missing features. Some on the Rust side. Many on the TypeScript/React frontend side, if you also are interested in doing stuff there. I'm happy to help you find something approachable.

Best regards,

Scott

To view this discussion on the web visit https://groups.google.com/d/msgid/moonfire-nvr-users/dec19441-e056-bd12-50d5-5cd2a33a869f%40gmail.com.

--

Scott Lamb <https://www.slamb.org/>