different camera heights --> resulting resolution?

[jan]

Hi I have taken >1000 pictures from a landscape with a helikite.
for most of the pictures, the camera was at altitudes between 50m and 150m.

Because photographs which were taken at lower camera elevations (e.g. 50m) have more spatial resolution than photo's produced from higher camera elevations (e.g. 150m), the question is how PhotoScan deals with photo's with different resolution and field of view, resulting from different camera heights?

Let's say for example that 900 pictures where taken at 50m height, and that those 900 pictures are enough to build the textured 3D model in PhotoScan. Let's further say that there are also 100 pictures that were taken at 150m height with lower resolution and higher field of view.

How will PhotoScan process this mix of 1000 (900+100) photos. Is it smart enough to leave the 100 low resolution photos out, because this will improve the resolution of the model? Or how will Photoscan deal with this situation if this is not the case, for (i) the alignment process, (ii) the geobuilding process, (iii) the texture building process and (iv) the orthophoto process?

This is important for me, because this may mean that photo's taken from a high elevation are best left out of the set of photos which is used as input in Photoscan. The problem here is that I don't know at forehand if the set of photo's is then sufficiently overlapping.

Kind regards,
Jan

[Diego]

Hi janq,

From experience I can tell you that if flight heights are very different it is necessary to work on different data set, the program in some cases is able to align the pictures but in very large data set and had problems, I have understood that  is developing a better way to align when you know the approximate coordinates of each camera position.  read here. http://www.agisoft.ru/forum/index.php?topic=54.msg282#msg282

Using GPS camera coordinates can be used to speed up alignment of large data sets, Hopefully this will be a reality soon. 

Best regards,

Diego

[gEEvEE]

Hi Jan,

I recently reconstructed a very large scene using Helikite photographs. I reconstructed the scene first with the high photograph only. Afterwards, I did the same with the low-altitude images. It is best to do so, as PS will mix the images and a less accurate model and texture will result (I tested this with RMSE, article is to be submitted soon I hope). I use PhotoShop afterwards to blend in the two different orthophotos.

Often, when the photographs are too low, the overlap is not enough for the alignment process to succeed. I then perform an alignment with all aerial images after which I disable the photos acquired from a higher altitude to build the 3D model and texture. There are some recent technical papers on this problem, but there is not yet a solution for this implemented in PS (as far as I know and see from my results).

Cheers,

Geert

[jan]

@Diego:
thx but I currently do not have GPS (and elevation) coordinates of camera positions.

@Geert:
I'd like to see that article once, interesting! After reading your answer: maybe the mixture of low- and high altitude photographs doesn't have a negative impact for the alignment process (the creation of the 3D point cloud), because the high elevation photos will only contribute to more points in the point cloud and not replace others (?). It is further maybe only the texture building process where it is best to leave out the higher altitude photos, at least if there is enough overlap left to cover the entire field. This would be less of a problem than selecting an appropriate set of photos at the beginning, as the texture building process is really fast, so that one doesn't loses a lot of time with choosing the best covering photoset with trial and error.

Is it correct to say that the geometry building process will be solely calculated from the point cloud generated by the alignment process, so that disabling photos in this geometry building process will not influence the outcome? 

[gEEvEE]

Indeed, use all photos for alignment (so those covering smaller areas can be positioned using the imagery from higher up). Afterwards, disable the highest photographs and start reconstructing. During the alignment, the positions of the cameras is calculated. Those positions are then used in the reconstruction step. So yes, it does not matter if you have a very dense point cloud or not, as the reconstruction algorithms do not work on feature points any more but on the pixel values itself. The feature points are needed to align the imagery.

Cheers,

Geert

[tkwasnitschka]

Hi all,
is it correct that in the current version 8.2.985 all loaded images are taken into account for alignment, regardless wether they are disabled? I usually disable every second image anyway, and the only difference I see in the viewport is the coloring of the cameras in different tones of blue.
Not that I mind this, I just want to clarify.
Cheers,
Tom

[Alexey Pasumansky]

Hello Tom,

You understand right, that for alignment procedure PhotoScan uses all photos in chunk (even disabled).
Disabled photos are not taken in account for geometry  and texture reconstruction.