First of all, I'm intrigued as to why you chose the Kinect sensor for your application. Is there a specific aspect that you're looking for that needs to be captured on camera? Or is it strictly depth analysis? If only depth analysis, why didn't you go for an infrared transmitter/receiver circuit or ultrasonic sensor like the
PING?
Second, why not divide your textile surface into small planes, and then mesh them together? The smaller your planes, the less the chance for marginal errors.
As for identifying if a point belongs to a plane or not, why not use measure the distance from the point to the plane and see if it does not exceed epsilon (or some other constant you specify)?