Reconstruction of endoscopy videos

[Michael Barbour]

Hi, 

I'm attempting to reconstruct a human airway from endoscopy videos. I understand this is a challenging problem for SfM, but I've had okay success with the trials I've done so far - there is a decent amount of texture (rings, arteries, etc) that have generated reasonable reconstructions thus far. 

I'm currently having issues with noise around the edges of the image after stereo processing. I'm using a mask during feature matching, but the noise that I see in the depth map is outside of this mask. I assume the mask I use during feature matching is not applied during stereo matching or during fusion? If I apply the mask myself to the depth images prior to fusion, what values should I apply to 'blank' the edges? zeros?

I also have a question about scaling the reconstructions to world coordinates. During the videos, the camera is passed through a cylinder of a known diameter. My plan is to include this cylinder in the reconstruction, and scale the full reconstructed model based on the physical diameter of this cylinder. Assuming the camera parameters are consistent for every reconstruction, this should work correct? Even if I have models that are disconnected? 

Best,

Mike