Hi,

I tried out your software with a single kinect but I have limited space to perform so to maximize my performance space I sometimes get really close to the wall opposite my sensor. However when getting very near the wall the depth image seems to be filtered out even before I touch it. I guessed there must be some filtering plane in the software causing this and wondered if there is a setting somewhere I can adjust the value todial it in against the wall more or remove it completely and use as much floor as I actually have.

There is no special filtering plane in our software. Background filtering works like this: depth values in video are compared to depth values evaluated for the background. If two values are close enough - that point is considered as background.
Can you please provide some screenshots of your issue, or better share a videofile?

Actually I don't have the movie or software here but I can try to recreate when it is available again. The room is about 10 feet long from sensor against one wall to the wall opposite it. I evaluate the background with iPi Recorder then I go to the wall and do a T pose. If my back nearly touches the wall when I do this the depth looks fine in Ipi Recorder but when I import that movie into mocap studio the 3d points of the parts of me that are closest to the wall are missing. Usually my elbow region. If in the video recording I walk a few inches forward from the wall and redo the T pose all the points are captured. Since the depth map looks the same (no gaps) in the recorder software I assumed there was some sort of occlusion plane. I did try to save the movie once with depth compression method set to none thinking it was being subtracted their but that did not seem to help.

Somewhere I read you need to not be in the background when first starting the record process but I assume that was before the "evaluate" feature on the recorder? I was assuming that it only necessary to be out of frame during that so I have bee using the countdown cutting right to a T pose on actual record. Would waiting out of frame a few seconds also during record help maybe?

Yes, it's definitely the work of background filter. The depth values of some, especially thin, body parts are too close to the values of the wall (which is the background), and are cut-off. You can do nothing about it except finding larger room for making records.


Yes, it was actual only in v1, where background was post-evaluated in Studio.

Thank you vmaslov for all your help. Perhaps a feature request is in order? I think it should be possible to gather all the data points the make up the background point cloud and push them negative on the Z axis all relative to each other. An entry for background offset could be included in an advanced setting. Well this seems to be only a minor inconvenience though and maybe only affecting a small amount of users.

As an alternative do you know if a lens adapter like this: http://www.amazon.com/Zoom-Kinect-Xbox- ... oom+kinect would work with your software?

No, cause it greatly distorts image, and our software have no correction algorithms for this. As a result, the data quality is not suitable for accurate mocap.