Multiple skeleton tracking question

Hello, in the eco system of gamma and its nuggets, on wich devices would you go ?
And wich are the pro and the cons ?
Thanks !

I still prefer Kinect ONE (2) and a mini pc with data steamed to a master pc. Cheap, and doesn’t need a big pc to work,(latte panda2+ works pretty well) you can use multiples easily to cover larger areas. Everything else seems to use #ML #AI and doesn’t seem to do that much better a job.

Hi cat ! I really prefer to go an hardware solution not being outdated. What are the recommandations in case of reliablity ?

@karistouf I think we’re in a funky moment for tracking. One problem is the availability of Kinect devices (the Azure Kinect was recently discontinued), while the replacement isn’t ready for to use at the moment.

We asked Nuitrack and they recommend the Orbbec Astra+ to work with their software. The VL.Nuitrack nuget isn’t compatible with its latest software, however and needs updating.

So if you want to do skeleton tracking with VL at the moment, the *best option probably is a Kinect of some kind.

*totally my own limited opinion

Thankyou @Hadasi. Actually the question is more to prepare the tools, its for a project i will begin to work on in june. So i presume Nuitrack package will be up to date ? Yours

There is no ‘hardware solution’ apart from the orbec (visible light only) with a built in pc and Ethernet output, the ‘hack’ with a mini pc is peasants option to do that on the cheap. The trouble I have with k4 etc is you need a beefy PC, to make use of the them, or use a mini pc per k4 with more than the minimum specs, which are higher than you need with the k2. Given that you can pick up k2 cheaply, you can stock up spares for a long term install and hack the power supply and usb3 cable to rid you of the propietry Kinect for windows box. I’d love to find a better solution, I still can’t quite believe the k4 didn’t include an onboard risc compute module to allow multiple sensors per pc. To me that’s a massive gap in the market.

Thank you dear cat for this very valuable feedback !

Haven’t used them myself but aren’t Stereolabs Zed cams also an option?
The VL.Devices.ZED nuget is currently being worked on by @Rayment I think.

Depending on your usecase VL.MediaPipe might also be an option.


thank you joreg its going far more better that mocap4all.about installation i m missing something:

you’re mostly missing the correct way of installing nugets. it is: nuget install vl.mediapipe -pre as documented here. please continue any mediapipe discussion over at VL.MediaPipe.

thank you joreg, helas this version doesnt get my camera devices, tried on two computer. No camera detected, no calculation

Mediapipe Pose doesn’t support multiple people.

it does actually, see: Pose landmark detection guide  |  MediaPipe  |  Google for Developers i’ve tested 2 here.

ok then the info on github is outdated.

i tried adding opencv, wich show well my camera. on first computer i have 3 cameras, and on second only one. Same results on the two.