Large sets of imagery

[fpbv]

Hi

I am a cartographer engineer here in Brazil and we are testing Photoscan. We got some pretty good results and we are considering to buy it. Nowadays we are doing some projects with some large sets of imagery, i.e., more than 3 or 4 thousand photos. I would like to know how Photoscan can handle this, and if we could have a custom feature to deal to this large sets.
I hope to hear from you soon.

[Dmitry Semyonov]

Hello,

Current version of PhotoScan should be able to process up to 500 - 700 aerial photos (using 64bit Windows with 12Gb - 24Gb RAM) in a reasonable time.

Larger data sets may be processed in parts, so that multiple orthophoto/dem maps are generated. It should be possible to merge these parts in the external software.

We are also planning some improvements to the 3d reconstruction engine which should make possible to process larger data sets in a single pass.

With best regards,
Dmitry Semyonov
AgiSoft LLC

[fpbv]

This is nice to know.
When you talk about reasonable time how long this is it? 5 or 6 hours or more?

Thanks in advance.

[Dmitry Semyonov]

Hello,

The time will depend on the kind of imagery. Processing photo sets with oblique viewing angles usually takes longer time, than photos with vertical viewing directions.

Also using high end GPU will help to build geometry faster.

We were able to process photo sets consisting of 500 - 700 photos with vertical viewing directions in 5 - 8 hours using the following hardware:

Intel Core i7 920 2.67GHz CPU
NVidia GeForce 480 GPU
24Gb RAM

The following parameters were used:

Reconstruction method: height-field
Geometry quality: medium

With best regards,
Dmitry Semyonov
AgiSoft LLC

[Greg]

Hi,

A technique I've wondered about but not tried would be to shrink the photos down to a megapixel or less to produce a rough model, then do full resolution chunks of parts of your area which you could align to the rough model. Then finally merge full res chunks without rough chunk?

Regards,

Greg