Temporal comparison in an urban environment - best way to align two sets?

[lemieszek]

Hi all,

I'm comparing two different picture sets from an urban environment that are 1 year apart of each other. It's a huge area (600 hectares). The idea is to extract from this information such as new built area, vegetation suppression and growth, etc.

I've processed them separately. I have about 60 GCPs and used them for reference (disabled the drone pics) and they're very decently aligned when I look at an orthomosaic. But the DEM is slightly off and I'm wondering if there's a better way around this.

The attached pic shows a simple raster calculation (A-B) and the result is an increase (blue), decrease (red) or no change (yellow) in height. Some stuff show up nicely: there's a new building (blue) built around an area where soil was removed (red). A few tree suppressions (red blotches). Some vegetation growth (blue shades around existing trees).

But there are a few problems: the overlap is slightly off by around half a meter on x and y (blue and red "ghosting" around buildings that didn't change), and the model is sloped to the right (red becomes stronger to the right of the image. This image is just a fraction of the whole area, but it's like this all over (going up and down by around half a meter or a little more than that).

The CloudCompare workflow doesn't work for me (each point cloud is 28 gigs and the error is larger then actual changes that I would like to capture (i.e. vegetation growth). So I need to find a way to line them up a little better.

I've tried processing them all as one chunk and then manually splitting them into different chunks to take one DEM out of each, but when I do that the alignment process resets. Maybe I'm doing that wrong. I've been adding a few more GCPs, but I feel I'm chasing my own shadow through this approach. I also tried processing as two chunks and aligning them after, but that doesn't seem to change much at all (maybe I'm doing that wrong too).

Any ideas?

Thanks!

[lemieszek]

Ran some tests on small scale that show some promise: I aligned all photos together (on high - maybe could have gone ultra high). Cleared the tie points with aggressive "gradual selection" (reprojection error and reconstruction uncertainty) so as to get rid of bogus tie points due to actual differences between the different sets. Then duplicated the chunk and on one I deleted photos from 2020, on the other I deleted the ones from 2021. Am now building the dense cloud on highest. Will report back.

If anyone has comments on this workflow, they're welcome.

[macsurveyr]

I don't think you should delete photos from the duplicated chunks. It would be much better to disable the photos you don't want to use in each of the duplicated chunks. Any products going forward will only use the enabled photos in each chunk.

Tom

[lemieszek]

I don't think you should delete photos from the duplicated chunks. It would be much better to disable the photos you don't want to use in each of the duplicated chunks. Any products going forward will only use the enabled photos in each chunk.

Tom

Delete isn't the right word - I removed them. Isn't disabling them essentially the same thing in terms of data? Though disabling allows me to reenable them later if I need to (which if all goes well I won't...)?

[lemieszek]

I'm also getting this awesome looking but terrible dense cloud. The bleeds around the edges are fine since I'm removing that at some point, but the entire map is covered in blotches of red (low confidence points). Is there a way to remove those while keeping the low confidence points within the model itself (which are sometimes useful)?

[macsurveyr]

Delete isn't the right word - I removed them. Isn't disabling them essentially the same thing in terms of data? Though disabling allows me to reenable them later if I need to (which if all goes well I won't...)?
[/quote]

Removing photos is not the same as disabling them. Removing photos removes tie points and can affect the results of the alignment and optimization and will affect future products. The images should be disabled.

Tom

[lemieszek]

Removing photos is not the same as disabling them. Removing photos removes tie points and can affect the results of the alignment and optimization and will affect future products. The images should be disabled.

Tom

Thank you! I'm processing the dense cloud now and will report back when it's done.

[andyroo]

...the entire map is covered in blotches of red (low confidence points). Is there a way to remove those while keeping the low confidence points within the model itself (which are sometimes useful)?

I typically filter the dense cloud to clean it a bit by classifying low-confidence points (0-2, sometimes more) as something (I usually do "low noise"). This allows you to use other tools to remove them or do further filtering. You could always just delete them, but why remove if you can classify? :-)

[lemieszek]

So here's the result from the method mentioned above - duplicating the chunk and disabling half the images on each. Incomparably better than the original (there's some fall off at the lower right corner, but that's the edge of the map anyway.

Thanks for the help.