Decimation and "fixed length per point"

[travis_b]

If I understand correctly, there is an algorithm in place whereby Agisoft upon decimating the project will look at neighboring vertices to determine how many points should be retained - ie - more points in areas where large changes on the axis. For the purposes of aerial surveying, this would generally be with respect to large changes in elevation.

I have a contact who has requested a point cloud with one point per 20 cm (whether he is implying squared or cubed I am unsure - lets assume it is cubed). This is generally how he determines the level of accuracy that he would like to have for the given area.

In Agisoft I dictate the number of faces to retain when decimating the mesh, however, this is difficult to convert into a measured distance between pixels that may be requested. Is there some other way to communicate the level of accuracy of the point cloud? The number of faces will provide very different accuracy dependent upon the area included.

I do believe that the algorithm that Agisoft uses would produce a much better result with an average of every 20cm than a point every fixed 20cm. I am not arguing the method - just the communication of the results.

Any ideas or tips here would be appreciated.

Thanks.

[bigben]

Just generate a DEM instead of a mesh?

[Dave Martin]

Travis,

Would suggest you try generating the best DEM you can from PhotoScan and then use Kriging or a similar technique to interpolate a DEM with the required resolution.  I've been doing this (with LiDAR data) for years and find well-tuned Kriging is the best at preserving features; however if its "piles" you're surveying then Minimum Curvature might be a more appropriate technique. The principal package I use for this is "Surfer" from Golden Software; and if you have any pre-survey breaklines you can utilise these when carrying out the Kriging.

Dave

[Kiesel]

When you want to do it in the point cloud itselve you could use Meshlab "poisson disk sampling-filter" for that (even I know it is not a grid). For a tutorial see:
https://www.youtube.com/watch?v=lttK4uMAMFw

It is normally used for terestrical laser scanning which produces highres point clouds in the near of the scanner and lowres at longer distances.
(For use with photogrammetry you have to scale your point cloud at first)

Karsten