working with multitemporal data

[wiesehahn]

I wonder what are the possibilities in lidR to work with point clouds from more than one point in time? Are there examples?

Recently I received ALS data of a larger area where there was data from a couple of years ago for the first time. And revisited surveys are getting more and more I guess.

My first thought was that I could maybe load both data sets in the same LAScatalog and work with it, but I got Be careful, some tiles seem to overlap each other. lidR may return incorrect outputs with edge artifacts when processing this catalog. which brought me to this discussion (https://gis.stackexchange.com/questions/345919/overlapping-las-tiles). So it seems working with multiple point clouds for the same area might not be intended in lidR and I guess is also not easy in general as point clouds may vary between ettributes etc. Anyway I think there is a lot of potential to work with mutliteporal point clouds.

Possible use cases I was thinking of include:

• multitemporal change detection, either
  • directly on the point cloud (as already dicussed for single LAS files in Set operations with lidR? #305)
  • based on intermediate products such as e.g. the difference of CHM from points of time A and points of time B
• create metrics based on the densified point cloud of A and B together where only subtle changes are expected (e.g. terrain)

For single LAS files this should be possible I guess, but what about larger areas in a LAScatalog?

[Jean-Romain]

create metrics based on the densified point cloud of A and B together where only subtle changes are expected (e.g. terrain)

You can read all the files as a single collection. You have a message for being careful, but it is only a message to trigger your attention. Process the collection with a chunk size != 0 and it will work. However pay attention to your data quality. It is unlikely that both dataset are perfectly aligned especially in Z.

based on intermediate products such as e.g. the difference of CHM from points of time A and points of time B

You can do that by splitting your files in two independent groups and make two CHMs

However there is no way to have two LAScatalog and generate the same chunks for both and load two LAS at the same extents to compare the two collection chunk by chunk. That might be the object of a feature request. I can imagine something like

multi_ctg = readMultiLAScatalog("folder1", "folder2")
opt_chunk_size(multi_ctg) = 1000

compare =  function(las1, las2) { ... }
catalog_compare(multi_ctg, compare)

[KBodolai]

I like the idea, unfortunately the data I'm looking to process is some of the EBA lidar transects, which contains hundreds of acquisitions, of which a few will overlap. For some of them (e.g. transects 0076 and 0720) it may be impossible to find such invariant features.

My initial idea was to try something similar for the whole DTM, given that overall, the lowest differences across the transects will be achieved when the heights are aligned (under some very mild assumptions). I also like the idea to do it more generally, where the displacement could be in any (x,y,z) dimension. Perhaps a similar thing can be made with shifting the DTM in X/Y and selecting the one which minimises the difference.

[Jean-Romain]

If you want me to develop custom tools for you specific need this now my full time job. See my website https://www.r-lidar.com/services. I'm currently integrating some point cloud registration algorithm in lasR as well as affine matrix transformation so I may already have a code base for your need.

I don't think lidR has what you need right now but I'm not sure to fully understand your requirements.

[KBodolai]

Hi @Jean-Romain, apologies if I came across as asking you to develop something specific, I was just wondering if you had come across this problem before. I'm currently testing a couple of registration algorithms that seem to perform well on this specific dataset, but it got me thinking about the problem more generally.

Great to know that you're working on this full time now though! I think I'll have to process quite a lot of ALS data in the near future, so I may get in touch soon :)

[Jean-Romain]

No offense. I was just informing you of my new status. If you are not interested it is not a problem. My software remain free and open source. If you are potentially interested I get a potential new a client. That's it.

I have never come across registration problem. That is why I never implemented anything. But you are lucky I will probably come into registration problem soon so I started developing tools. But the free and open source tools are coming "when it's ready" i.e. I don't know when 😉

[KBodolai]

Excellent! Let me know if you can use a hand with that. I'm already writing a small benchmark of different approaches and it could at least serve as a small example in the documentation.