Tips for Speeding up Processing of UAV photos

[vicviod]

Hello all,
I'm currently working on a project where speed of processing is a big priority. With that in mind, do you guys have any tips for speeding up processing of photos taken by a UAV? To be more specific,

• Are there certain attributes of photos that make them quicker to process? For example, if pitch and roll are extremely close to zero, will this make an appreciable difference in processing speed? Are there any other tips I should keep in mind while taking data to help speed up processing?
• Are there certain settings that will make the processing photos go faster (without sacrificing too much quality?
• How much does the number of images affect processing time? For example, is processing time O(n) complexity (where n is the number of photos), or O(n^2), or something else? Should I make as much an effort as possible to trim out photos with too much overlap?
• Is processing the data in multiple smaller chunks (for example, subdividing the region on the map) preferable to doing it all in one shot?
• Does certain data in the reference make a big difference to have? Currently I have long, lat, altitude, yaw, pitch, and roll
• Do you guys have any other tips?

This is my current process:

• Trim photos
• Align photos (using reference preselection)
• Build the dense point cloud (on medium quality)
• Generate mesh (arbitrary surface type)
• Build UV (generic mapping mode)
• Build Texture (with mosaic blending)
• Build orthomosaic (using model data and mosaic blending)
• Export orthomosaic in tif format

Knowing all this, is there anything I can do with regard to data collection, data trimming, settings modification, or anything else that will make a difference with processing time?

Thanks guys!

[Alexey Pasumansky]

Hello vicviod,

What do you mean by "trimming photos"?

For alignment you can enable both generic and reference preselection and use keypoint/tiepoint limits that are close to the defaults (40,000 and 4,000 correspondingly).
After alignment I can suggest to run the optimization and then reset region.

In case you are processing some landscape/field areas, you can use Height Field surface type on the Build Mesh step, or even generate DEM from the dense cloud and use it as a source for the orthomosaic generation. Building texture is not necessary for the orthomosaic generation.

[vicviod]

Thank you so much for the tips, Alexey!

By trimming photos I mean eliminating unnecessary images. Because we are dealing with 30 fps video footage, there is a good deal of unnecessary overlap and a perfectly good picture is created by taking 1/10 of the total number of images.

What do you mean by "run the optimization and reset region"?

If I produce the orthomosaic from the DEM (which I would produce from the dense point cloud) how will this orthomosaic be different from the one I am currently producing (using he mesh)?


 

[Alexey Pasumansky]

Hello vicviod,

Under Optimization I mena Optimize Cameras button on the Reference pane that refines the exterior and interior orientation of the cameras based on the reference data weights and values.

Reset Region recalculates the position, size and orientation of the bounding box that limits the reconstruction area. For the georeferenced chunks it makes the colored side of the bounding box parallel to the ground (XY) plane.

[Rony fly]

If it's just to have DEM and orthomosaic, no need to generate meshes and textures

This is my current process:

    Trim photos
    Align photos (using reference preselection)
    Build the dense point cloud (on medium quality)
    Build DEM (with densepoint  cloud)
    Build orthomosaic (using DEM)
    Export orthomosaic in tif format

Longer :

I create two pieces, one georeferenced, and one without. Then after refining the errors of the first model, I align the second on the first.

Alexey? Does it ok as a routine?

[Rony fly]

correction :

 like you,  i trim Sky zone or remote zone photos in first step, for exemple
but not optimized after alignment , because often error factor