Depth camera hunt

Good day my dears,

I am currently trying to decide which depth cameras to get for upcoming projects and could use your experience in the field.

I am interested in:

  • IR image support
  • Decent quality RGB image
  • Decent quality depth image
  • Skeleton tracking
  • General object tracking/recognition

In general you can imagine the use cases: tracking objects via markers, tracking objects using AI/CV, skeleton tracking and pointcloud reconstruction.

So far I was thinking about the Orbbec Femto sensors but their native SDK does not do Skeleton tracking and instead relies on a wrapper they built for the Azure Kinect SDK, which is no longer being developed and feels a bit off.

I heard good things about the Zed camera but after asking around a bit more, it sounds like their Skeleton Tracking is based on an AI interpretation of the RGB stream which can get confused when you have projections in the scene’s background (a very likely use case in our field). Not great…

There is also Nuitrack which provides its own layer of Skeleton tracking and features for quite a few sensors out there but I was hoping to go native where possible.

Finally there is the Intel Realsense which I have not yet read much about.

What sensors have you worked with in the past and what are their pros and cons? What would you recommend to use these days? What would you NOT recommend?

Thanks in advance for any information you can share!

Cheers!

I think realsense are depreciated now too. Depth cams are out of fashion because Ai is the ‘future’. I still use kinect2 with the 12v hack, cheap, low over head, but with limited range etc. Oak-D OAK-D Pro – Luxonis look interesting, but I have one (not pro) and so far havent found a use for it, and its Ai again and using python, but with stereo vision, the pro can do kinect dot projection. All the ai skeletons seem quite laggy to me, so the kinect still wins…

1 Like
2 Likes

I’ll check it out when I have some time, but tbh, the skeleton tracking was pretty low frame rate anyway, the depth disparity was better I seem to remember.

This will also be available in VL.PythonNET soon:

2 Likes