kinect 2 depth distortion

[john brown]

Hi there 

i was wondering if anyone has suggestions for correcting the depth distortion or stretchedness in the raw point clouds. it would be great to be able to correct using  a transform the point clouds into something that is more physically accurate. 

[Brekel]

When exporting geometry you can try to play with the "Stretch Threshold" parameter to filter out polygons that are stretched too much along the depth axis, assuming that is what you mean.

[john brown]

Hi Mr Brekel. 

I've been trying to match up some of the scans with footage taken from different angles on other cameras , so I've come across some inconsistency's  particularly in the depth axis. 

I was quite shocked at how long my chin and nose were when seeing them for the first time in the 3d viewer ;)  

I've come across this post with some people speculating what's the cause and how to fix it . http://www.depthbiomechanics.co.uk/?p=2659

It's quite an interesting problem - I can imagine in the future people will have to calculate lens distortion for your camera sensor  in the way you do now before camera matching or stabilising video footage .

Maybe the same algorithms could be used ?

with video you can hold up a chart with straight edges drawn on it and then distort the footage until you make some straight lines .

maybe you could hold some geometry in front of the kinect first and adjust the 3d scan to fit ?

Cheers 

John 

[Brekel]

Hi John,

That article in particular references Kinect v1 and similar structured light sensors.

The Kinect v2 is based on the time of flight principle which has very different noise and distortion characteristics for example due to the way light bounces around in corners.

Distortion compensation for depth sensors is an active research topic that hasn't been solved yet, especially not in the cheap consumer depth camera realms.

It is not something I'm actively working on but if things arise in this field I'm always interested in looking at them.

Greets,

[Brekel]

Also keep in mind that it is not as simple as a static geometric distortion.

Since the sensor senses depth by measuring reflected distortion will be dependent on things like reflectance properties of the measured surface.