3D/Depth cam: Oak-D lite AF vs realsense vs kinect

Hi TD community,
i’m on mac os, so bye bye kinect, isn’t it? kinect 2 too?
Well i’m hesitating between buying :slight_smile:

  • an Oak-D lite AF Cam ( hope we can control the AF remotely or lock it though ) edit: this model has no IR sensor, so won’t work in the dark
  • or a realsense…
    I saw Kinect has stopped Azure lately…
    or any other one?
    could you share your Feedback on those depthcams on macOS?

thanks a lot!

Well MacOS currently isn’t the ideal OS for working with depth sensors…

Hi and thanks for the synthétic reply,
I guess I ll have to switch on PC,
And therefore Nvidia AI recognition will help to avoid using Kinect for depth map anyway…
Best regards

Hi @nettoyeur,
what do you advice as a PC configuration (barbone style I ve heard about geekom as a nice and cheap solution ) , to get Kinect V2 ( aka kinect One) depth analysis and sending it over NDI video to the network and make the depth map processed by TD on another plateform?
@jacqueshoepffner would you confirm this as a nice solution? it would allow to use kinect with more distance and devote a machine to depth map aquisition and basic adjustements
Oak-d seems a bit not really handy for now …

Hello Ben,
For me thats not a easy way, because NDI is 8 bits, compressed RGB image.
It depends on what information you need from depth camera.
– Skeleton, you can transform information (25x3) into 32 bits image with Chop to Top Top and then pack a 4x75 8 bits image with pack Top but you need not compressed image
– Depth map, you need a 16 bits monochrome uncompressed image
– point cloud, you need a 32 bits RGBA uncompressed image (as EXR)
For the moment, AI depth recognition is not on par with depth camera.
Concerning Kinect azure, Microsoft has stopped production but Orbbec is manufacturing the same model.
OAK-D is not exrensively tested at the moment but seems interesting.
Jacques

thanks a lot for the complete reply @jacqueshoepffner !
indeed “touchin/out top” could do better than ndi, but i’m afraid of the brandwith it would take especially for pointcloud, do you have any idea?
i guess skeleton analysis could be done on kinect devoted machine and send datas via touch in/out chops,
have you tried comparing kinect V2 and orbbec ( this seems to have mac support btw ) ?
Therefore i’d go through an USB /RJ extender i guess…
I’m longing AI depth recognition as i already have a good router and multiple phones that bring me HD through NDI HX Cam app with really low latency
Regards, Ben

Im running macosventura on a macstudio m2, and still looking for a solution to use any depth camera and SDK for real time body tracking; I’ll be happy to get a simple silhuete so i can stream it via ndi to Resolume or Mad Mapper. Any suggestions??’