Large scene or landscape capture

[Virtualhc]

Hello,

I'm a graduate student and for my thesis I've been testing using Photoscan. I've done a few successful small to large object scans. I'm working with VR so I've been intrigued by the idea of capturing large interior or exterior scenes and using them as VR environments with either the Oculus Rift or HTC VIve. I came across an example of someone using Photoscan and the Vive to capture full rooms but I was particularly wondering about the technique or workflow to capture a large landscape like the one he shows towards the end of this video:

https://youtu.be/gGbLKkWZCgY?t=1m42s

He has a series of blog postings where he briefly describes the process but without much detail. He does include a screenshot of Photoscan as the software he used.

So I was wondering if anyone here would have some workflow recommendations as to how to achieve something like this, particularly how to go about camera placement. In the video you can see that he got not only the large landscape but also the rocks so he probably did chunk processing?

Thanks!

[GPC]

I would guess the person did it with a small UAV to capture aerial photographs.. I think that's the way most people are using the software.

What software have you used to create the VR? this is a new area for me and wanting to go from Agisoft into a VR model.

[bigben]

Travel distance in the video isn't very far so it may not be a large scene... may be a spherical panorama for the distant landscape/sky

[Alexey Pasumansky]

Hello Virtualhc,

In the original post there's a screenshot with the camera locations (the first screenshot actually):
http://steamcommunity.com/games/250820/announcements/detail/117448248511523471
However, usually we do not recommend to take multiple images from the same camera position, unless nodal head is used.

Additionally, I can suggest to take images from the highest possible altitude (for example, mounting it on the rod or pole) and minimize the sky in the image frame space, as it is useless for the reconstruction. The idea of putting camera higher is simple: low camera position will result in a lot of sky and a lot of ground surface shot with extremely oblique angles (almost no matching points can be found on the ground in such approach).

[bigben]

Looking at those camera positions that looks similar to a project I'm working on at the moment. A full frame fisheye on a 20mp camera (or larger) would capture that scene relatively quickly.  I've used a painter's pole but it's a bit cumbersome in rougher terrain, or even a monopod supported on a belt around your waist to give a camera height of ~3m works pretty well.

2 cameras on a pole ~1m apart (vertically) would be even better.  This is a small test I did from a single position (handheld, no pole).
https://skfb.ly/ILRV

The nice thing with this camera/lens combination is that it's easy to get high res models (closer with more pics) at the same time as capturing a large scene.