3
votes

I'd like to be able to modify the raw depth stream data (and/or colour) before it's used to generate skeleton data, to see if i can get better results from skeleton tracking in the environment i'm working with. From what I've read/attempted I don't think it's possible, but would love to be proven wrong.

Can anyone suggest any other method i where might succeed, short of writing my own skeleton tracking algorithms? Is it possible to extract data from the .xed files recorded in Kinect Studio and modify them before replaying?

NB: Using Kinect for Windows, C# SDK 1.6. Cheers!

1
Could you elaborate a little on what kind of preprocessing you want to do on the depth data? (Like distortion correction for example or I understood it completely wrong?) - Sassa
Yes, i want to do a transformation on the depths based on the angles of the kinect and certain items in the frame. - user1902847
If you want disparity image (one step before depth data) you will probably have to work with libfreenect, instead of the MS SDK. But either way, why can't you just keep the depth bits and do whatever you want with them? - Sassa
Yes, I can take the depth bits and do what I want with them, but how can i push that modified stream of data back into the skeleton engine? - user1902847
I see, I thought you wanted to write your own skeletal tracking either way. I really don't believe that the MS drivers allow you to do what you ask for at this point. I would probably suggest either writing your own skeletal tracking, or searching for any already implemented open source skeletal tracking algorithms. Would it be also possible to transform directly skeletal coordinates instead of the depth frame? - Sassa

1 Answers

0
votes

Unfortunately, the SDK doesn't allow access to the data streams before generating the skeletal data. You would need to look into a different package to do that- check out PrimeSense NiTE 2... but if you gotta stick with the MS SDK you will have to implement your own skeletal tracking or tracking correction.