Amateur video broadcasting

jnebrera

Hi all,

I'm trying to build an amateur video broadcasting solution for kid sport games. This is more or less the idea:

I want to have at least 2 independent cameras, each pointing to a different angle of the field. The cameras need to be good quality, maybe not 4K, but for sure desirable and at least Full HD at 30fps

As an "autonomous" producer, I want to use DL to detect the camera that is seeing the ball, and broadcast that specific signal (to Facebook live for example), discarding the other.

Of course, it would be ideal to do this using the cheapest hardware, and also low power / reliable one.

Now, to make things a bit more complicated, the is a desirable improvement on top of this. I would like to overlay the score of the game on top of the final video. To do so, I pull the score from either the field scoreboard or from a remote API. This result could be "plugged" either at the camera level, in such a way that the camera produces the final frame with the scoreboard already present (thus both cameras will do the same) or at the host level in case this is not possible, which would mean encoding the video at the host level, which is not desirable.

What combination of hardware do you suggest?

Thank you

Brandon

jnebrera

Great questions. So this is something that megaAI can probably help with, particularly knowing where the action is (based on players, ball, or maybe total motion) and then using only this stream for live streaming.

And in the coming say 4-5 months megaAI will be even better suited for this sort of application. Being able to do more than what you're thinking here

SmartMotion (here)
Digital PTZ Support (here)
Pipeline Builder Gen2 (here)
Text Overlay Prior to encoding (here) (Just added this one based on your post... I've been meaning to write this one up and this was enough impetus!)

So combining 1 and 2, which you'd be able to piece together with 3 you could also do automatic smart zooming. So for example running motion detection and then zooming in automatically to the area that is of most interest.

So in terms of hardware to use, you could use a megaAI with a Raspberry Pi and say a PoE HAT (or battery) for easier powering/data.

We also have a PoE variant of megaAI in the works (the hardware is verified, now writing the firmware support) which could be used w/out any host as it will support RTSP.

Thoughts?

Thanks,
Brandon

jnebrera

Hi Brandon,
Thanks for you reply. Indeed, some of the stuff you are suggesting would be great as an "extension" of the project at a later stage.
Here are the alternatives I have found so far, prices kind of very approximate, just orders of magnitude:
MEGAAI

2x MegaAI cameras => 2x199 = 400
Dual Camera Splitter => 1x60 = 60
Rapsberry Pi Zero W => 1x10 = 10

So that's a total of 470

JETSON NANO

1x Jetson Nano Board => 1x110 = 110
2x High Quality Cameras => 2x60 = 120
Dual Camera Splitter => 1x60 = 60

For a total of 290

CORAL

1x Coral AI Board => 1x130 = 130
2x High Quality Cameras => 2x60 = 120
Dual Camera Splitter => 1x60 = 60

For a total of 310

LINDENIS V5

1x Lindenis V5 Board => 1x85 = 85
2x High Quality Cameras => 2x60 = 120
1x Wifi Module => 1x15 = 15

For a total of 220

So as you can see, the fact that I'm seeking for a two camera solution penalizes MegaAI technology. Of course, you have to consider size and power usage, but you get the point. Both Coral and Jetson are in a sweet spot, and the Lindenis solution seems quite interesting if you only consider pricing but risky by itself, as I don't see too much of a "popularity" factor and thus support from other projects and such.

Brandon

jnebrera

Very interesting. I hadn't seen that AllWinner chip before (here). These are generally extremely inexpensive but a bit higher power and physically larger compared to say NXP... but they're just so inexpensive.

So a small thing on our price comparison: You don't need the dual-camera splitter because megaAI is USB-based. You can also get OAK-1 from our KickStarter campaign now for $99 each so then the total for our solution is $210... actually the cheapest on the list! (At a total of $210 including the Pi Zero)

Also the ESP32 variant (here) and the PoE variant (here) may be of interest as they don't require any other hardware (i.e. no Pi or host/etc.)

Thoughts?

Thanks,
Brandon

jnebrera

BTW, you can add Tinker boards R and T from Asus in the listing.

Both include some kind of NPU and both have 2 slots for camera, so you don't need the splitter and can save that money.

I don't have pricing info at this time, but is a board from a big player, thus quite a good reference

jnebrera

Another SBC appearance https://www.khadas.com/product-page/vim3 for 99$, seems they do support directly 2 cameras

jnebrera

Ok, so I found some pricing info. Tinker Edge T (based on Coral) is around 133. Tinker Edge R is about 200.

jnebrera

Hi again, some updates:

Khadas VIM3 board (100) which states it does support dual camera, doesn't at least at this moment
Coral Board, they don't warrant it will work with the splitter, so don't know if it will be a good idea
I just noticed the latest Dev board from NVidia (B01) does have two slots for the camera, so it won't need the splitter (but funny enough they can still be used for a total of 4 cameras !!!). This will reduce the cost of the Jetson solution to 230
The same carrier board can support the more powerful Xavier SOM, in case you might want to do extra juice in a future (for an extra 300 of course)
I have found this Orange Pi 4B to be the cheapest (70) and a total of 200 with two cameras, as doesn't need splitter, but reviews are quite bad on the software side, seeming impossible to use the GPU and the NPU at the same time and with confidence
The guys from Coral are quite responsive. Docs are quite good too.
Just the opposite with Lindenis, no reply whatsoever.

As for your suggestions, yes, the Campaign prices are quite lower, but those are temporal. Later purchases would be more expensive. As for using the ESP32 variant, I would need some punch to do the overlay, getting the game results from external API, etc. I don't think they will have enough punch. In your case, a Pi would be more than enough, and I see you don't need the splitter, so the total would be 410

To be honest, at this stage and due to the characteristics of my idea, the winning combo is the new Jetson Nano Dev kit, as already supports 2 cameras, does have great support from community and software side and has the capability, be it needed in the future, to upgrade to a Xavier based beast.

But again, this is for now, I do have to do some further research yet.

Also, I do have quite a bit of ideas to do with these toys, maybe your alternative is better for a different project.

Brandon

Thanks! All makes sense. So another option I meant to mention is the BW1098FFC:

https://github.com/luxonis/depthai-hardware/blob/master/BW1098FFC_DepthAI_USB3/README.md#camera-usage

It's open source hardware so the 3x FFC connections there could be changed to support 3x Pi HQ cameras for example.

Or if you want to use off-the-shelf hardware for prototyping, the Pi HQ adapter board (here) can be used with this.

Cheers,
Brandon