Agisoft Metashape

Agisoft Metashape => General => Topic started by: bgt051@mun.ca on August 31, 2020, 09:33:58 PM

Title: DEM Accuracy Issues
Post by: bgt051@mun.ca on August 31, 2020, 09:33:58 PM
Hi all,

I am having issues with my DEM's. I've read a few other posts somewhat similar to my situation but being very new to this field and software I am still confused.


I have collected aerial imagery data repeatedly from the exact same coastal beach location because I want to observe the elevation changes over time. Through flight planning software I've collected hundreds of vertical photos, the first survey flown at 60m and the rest at 40m. However, after following the workflow process to build DEMs for each visit to the location I find some differences between them to be greater than 30m (~100'). Locations on the DEM that I know to be impossible to have changed at all over my study period show changes of over 30cm (1').


I used 13 GCP over a span of <0.1 kmĀ² (25 acres), two of which I unchecked during processing in Metashape for the purpose of accuracy checkpoints. I have re-checked my GNSS data and have not observed any major errors with any of my GCPs.

I have noticed that water behaves very funny. For images I captured that was all water, Metashape fails to align those photos and remained unselected for the rest of the workflow. However, for the photos that are directly along the coastline and are able to align, does the water in these images create interpolation issues for the remained of the map?

Any help/suggestions would be greatly appreciated. I will be glad to elaborate/clarify anything to help lead to a solution.


Thanks,

Title: Re: DEM Accuracy Issues
Post by: dpitman on September 01, 2020, 04:12:51 AM
In general, SfM engines do not like water. The surface can change from one image to the next in the set and therefore provide little to no features to compare.

That does not entirely explain why you would get such divergent elevation data while using sufficient GCPs.  Did you specify a confidence report when you built the dense cloud?  If so, what does it show.
Title: Re: DEM Accuracy Issues
Post by: bgt051@mun.ca on September 02, 2020, 05:16:29 PM
Hi dpitman,

So I masked out any trace of water at all and it has improved significantly. My data now ranges in 8m difference rather than 30m. However, 8m is still quite a lot of change.

By confidence report do you mean the Processing Report? In the Processing Report, my Z error is 20 cm (<1').

Thanks,
Title: Re: DEM Accuracy Issues
Post by: dpitman on September 03, 2020, 07:10:46 AM
What Agisoft calls a "confidence" report is a visual overlap report.  You can get the same thing by going to FILE > EXPORT > GENERATE REPORT.  One of the pages will show you how good of a job you did getting sufficient overlap of images. (see attachments)

Did you follow this guided workflow to build your model : https://www.dropbox.com/s/k1yfx44z7zisoel/orthophoto-dem-generation-with-gcps.pdf?dl=0

Pay special attention to where you tell Metashape that the Markers are very accurate and the photo meta data is not very accurate prior to optimization.
Title: Re: DEM Accuracy Issues
Post by: bgt051@mun.ca on September 07, 2020, 05:19:32 PM
Dpitman,

That was indeed the workflow I followed! See my confidence report attached. As you see, my survey area is very narrow which may be problematic.


Shadows I have found to be another issue. I had to run my flight according to the ocean tide cycle rather than Sun conditions. After masking out water and shadows I get much more reasonable numbers!

Thanks,