DTM from photogrammetry point cloud

[Annette Dietmaier]

I am working with a point cloud that was derived by photogrammetry in the boreal forest of northern Alberta, Canada. 

I have tried out different parameter combinations with lasground_new and am mainly looking for feedback on my "calibration". This is how I used lasground:

lasground_new -i %FILES%\noise\*.laz ^

-step 10 ^

-bulge 0.5 ^

-spike 0.1 ^

-offset 0.1 ^

-all_returns ^

-extra_coarse ^

-compute_height ^

-olaz ^

-drop_classification 7 ^

-odir %FILES%\ground 

From checking my results visually, this is the "best" combination of parameters I could achieve. However, I would love to hear any feedback from whoever has succesfully classified a photogrammetry point cloud of a forested area into ground/non ground before. 

 This is a screenshot of my result. As you can see, there are some areas where the triangulation is resulting in big triangles (upper right area). Any hints on how to deal with that?

Thank you very much! 

Annette

[Mark Levitski]

I cannot believe that a photogrammetry point cloud could give you enough ground in a forest unless some ground points were shot with a traditional total station survey, which is tedious and expensive for a large tract of land. Nowadays UAV lidar that will penetrate canopy well is being used with great success. I know Martin has suggested some creative routines to get the best out of an SFM point cloud, but there are limits. I will be interested to follow this, thanks.

Mark Levitski

--
Download LAStools at
http://lastools.org
http://rapidlasso.com
Be social with LAStools at
http://facebook.com/LAStools
http://twitter.com/LAStools
http://linkedin.com/groups/LAStools-4408378
Manage your settings at
http://groups.google.com/group/lastools/subscribe

[Martin Isenburg]

Hello,

there are several tricks one could try. One particular recommendation would be to first classify a subset of "low" points using lasthin and then only run the ground classification on the subset. Your use of '-extra_coarse' (which gives a similar effect) suggests that this may be the right way to go. Here there are again many ways one could subset. I don't really have a lot of spare time to experiment on your data myself but in general it's a good idea to also post a link (a link, don't attach the data!!!) in compressed LAZ format (because if it's in bulky LAS I am less likely to attempt downloading it while - like now - in remote, Internet-band-width limited areas) with your inquiry.

Regards,

Martin @rapidlasso

[Sam Hackett]

HI Annette

Looks like an interesting task.  Definitely not what photogrammetry is suited for.  What accuracy are you looking for from your ground TIN? Does your dataset include clearings in the canopy where you can see the ground?   Why are you not keen on long triangles?  Long triangles suggest the ground algorithm is being conservative, which I would guess is the safest approach with photogrammetry in this environment.  I would be more concerned if you ended up with dense or consistent ground coverage, which would suggest you are being lied to.

Depending on the type of imagery and software to produce the densified point cloud you may also have low noise to deal with.  Low altitude small photos from UAV may give false low match points if a few images have poor registration in some areas.  Water can contribute with reflections etc.  Higher altitude med/large format imagery is less likely to have this.  Does the data include RGB and/or NIR?  These may help you have more confidence in the result, probably the latter more so, as the ground will all be shadows.

Sam

[Annette Dietmaier]

Hello everybody,

thank you for your replies! 

I have just started working with LAStools. Being a newbie and the only one at my university department who is working with LAStools, I am trying to figure things out step by step, so I appreciate your patience!

My task is to examine forest characteristics using both LiDAR and DAP data. For my thesis, I have to discuss the different advantages and disadvantages and the methods' capabilities. 

Everything is working out very well for my LiDAR data set. I am now trying to derive the same kind of products (DTM, CHM and DSM) from the photogrammetry data. I am aware that it will not produce the same results as the LiDAR, but I still want to make sure that I get the best result possible for a DAP dataset. 

My dataset includes clearings, roads, bogs and lakes. I chose "coarse" because the terrain is not very steep. Sam, please correct me if I'm wrong in this assumption. How would I go about choosing longer triangles? 

I have been given the DAP point cloud already assembled, so I (unfortunately) cannot do anything about the coregistration quality of the point cloud. 

Thank you so much for everybody's assistance!

Annette

[Martin Isenburg]

Hello,

we have a number of articles on how to use LAStools to get reasonable DTMs from dense-matching photogrammetry on our company blog that often provide the data and guide you through step by step:

https://rapidlasso.com/2017/07/04/removing-excessive-low-noise-from-dense-matching-point-clouds/

https://rapidlasso.com/2017/06/13/integrating-external-ground-points-in-forests-to-improve-dtm-from-dense-matching-photogrammetry/

https://rapidlasso.com/2016/11/22/creating-a-better-dtm-from-photogrammetic-points-by-avoiding-shadows/

https://rapidlasso.com/2015/10/07/removing-low-noise-from-semi-global-matching-points/

https://rapidlasso.com/2015/09/21/creating-dtms-from-dense-matched-points-of-uav-imagery-from-senseflys-ebee/

https://rapidlasso.com/2014/08/25/processing-large-dense-matching-point-clouds/

https://rapidlasso.com/2014/06/13/dtm-from-dense-matching-points/

Given the density of these point clouds I usually advice to run the entire ground finding pipeline only on a thinner subset of points and then "densify" it later by adding all the points that are within +/- 2 or +/- 5 cm of this thinner ground TIN to the final set of ground points.

Regards.

Martin

--