Interior Photography Best Practise?

[jknox]

Up until now all my photogrammetry work has involved objects, tomorrow that will change!  I have a heritage roofless interior to scan and would like any tips on best practice.  I've read conflicting ideas about this; firstly to move around the perimeter of the interior pointing towards the centre, and secondly, to keep a set distance from the perimeter (shooting out).  Any ideas on this would be appreciated.
Thanks J

[bisenberger]

Chapter 2. Capturing Photos in the User Manual has a schematic that shows the preferred method for taking photos of interiors.

[Alexey Pasumansky]

Hello jknox,

Here are some recommendations, but they could be applied not only for interiors:

- provide good lighting conditions for the interior (indirect diffuse light is preferred), we suggest that windows should be covered, since they produce high contrast and can not be reconstructed anyway due to transparency;
- avoid blinking and mirror surfaces - they are problematic for reconstruction (almost impossible) and may induce false matches into the camera alignment, transparent objects can not be also reconstructed;
- PhotoScan works well with finely textured objects, whereas untextured, plain surfaces without any features (like metal/plastic objects or object parts) could not be reconstructed correctly, so you may encounter problems with untextured white walls;
- another type of problematic object type is thin objects like pipes;
- use minimal possible ISO value to prevent additional noise on photos;
- provide sufficient focal depth to capture details of scene, blurred images will be almost useless, probably you should use tripod;
- provide sufficient overlap between images but do not take photos from the same location just turning camera around - it's much better to step aside;
- note that areas that are not seen from at least two photos will not be reconstructed correctly;
- using fixed lens is preferred, for zoom lens we recommend to use only minimal or maximal positions and avoid intermediate focal length values.

[osima]

Hi Alexey,

I was wondering about one of the points on your list of good practices in image collection: Why is it not advised to take photos from one position by turning around (panorama like)?
What is the reasoning behind that? It is exactly oposite to what is adviced for multi-view stereo matching in here:
http://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-5-W1/251/2013/isprsarchives-XL-5-W1-251-2013.pdf

best regards,
Aleksandra

[Alexey Pasumansky]

Hello Aleksandra,

I've looked through the paper and it appears that problems of small angles between cameras as well as big stereobase are discussed in Section 2. As far as I understand image acquisition method proposed in the article they suggest to take non-overlapping images from the same camera position, since depth estimation requires sufficient stereobase.
In practice it is almost impossible to avoid significant overlap for photos taken from the same camera position, so we recommend to take each photo from different point.