Autoclassify Ground Points option

DionybellDionybell Global Mapper UserPosts: 33Trusted User
edited December 2013 in Elevation Data
When choosing the "Minimun height departure from local mean for non-ground point". Can someone explain me what this means? What do I have to consider for choosing the correct one?

I'm trying to use this feature for Lidar data on a terrain with medium to high vegetation and buildings.

I dont like the results when using the default 0.3. The results don't seem accurate.

Thanks in advance.

Comments

  • global_mapperglobal_mapper Administrator Posts: 17,238
    edited October 2013
    That value has to do with what deviation will look like non-ground, it's exact meaning is somewhat complex. But in general if you have high vegetation and buildings, try using a larger value for that. The default is best for low-to-medium vegetation. For flatter areas you would use smaller values.

    Also keep in mind automatic classification will always have some mis-classifications, no automatic algorithms get better than 85-90% accuracy.

    Thanks,

    Mike
    Global Mapper Guru
    geohelp@bluemarblegeo.com
    Blue Marble Geographics for Coordinate Conversion, Image Reprojection and Vector Translation
  • DionybellDionybell Global Mapper User Posts: 33Trusted User
    edited October 2013
    Thanks a lot.

    Another thing. I'm trying to use the feature to classify a 205 million point laz file. The program goes "not responding" and the proccess won't even start. Any way I can solve that?

    Thanks in advance.
  • global_mapperglobal_mapper Administrator Posts: 17,238
    edited October 2013
    I'm afraid you would need to restrict the operation to a smaller area, the process would be extraordinarily slow when trying to automatically classify that large of an area at once. You should be able to specify the bounds of a smaller area on the dialog. I might also add some way to automatically split the operation into smaller pieces.

    Thanks,

    Mike
    Global Mapper Guru
    geohelp@bluemarblegeo.com
    Global Mapper
  • DionybellDionybell Global Mapper User Posts: 33Trusted User
    edited October 2013
    I can split the Las file in some parts. How many lidar points can Global Mapper 15 handle so Autoclassify Ground Points feature wouldn't be so slow?

    Thanks in advance, you have been of great help.
  • jh93989jh93989 Global Mapper User Posts: 41Trusted User
    edited December 2013
    Dionybell wrote: »
    I can split the Las file in some parts. How many lidar points can Global Mapper 15 handle so Autoclassify Ground Points feature wouldn't be so slow?

    Thanks in advance, you have been of great help.


    I cant answer that question directly but 20million points seems to be the sweet spot for the different classification programs I use. You could probably get away with 3k by 3k tiles depending on your point density.

    Also, whats with the spam posts?
  • global_mapperglobal_mapper Administrator Posts: 17,238
    edited December 2013
    Sorry I missed the original question about the number of points to classify to get better speed. I would try and break it into chunks of a couple million or so as the time per point will get slower as the whole area gets larger, but it's not as non-linear as some other operations.

    The spam posts are a little strange, they don't actually contain any links and often sound almost like someone being grateful, but they are too generic and strange. I think they are posting to probe the site and become a trusted user, then they would later follow up with real spam. In any case I've deleted those two posts and banned the users.

    Thanks,

    Mike
    Global Mapper Guru
    geohelp@bluemarblegeo.com
    Blue Marble Geographics for Coordinate Conversion, Image Reprojection and Vector Translation
  • geo72geo72 Global Mapper User Posts: 26Trusted User
    edited December 2013
    Hello Dionybell!

    I use the autoclassify-broud-option in GM quite often the following way:

    1) split up big pointclouds in tiles of about 5 million points (100m x 100m in my case) with the exportfunction of gm (create tiles when exporting)
    2) take one representative tile and test different parameters with the autoclassify ground-points option. (my favorite binsizes are 50cm and 100cm, for the min height dev I test 5cm, 10cm, 15cm,...,50cm - often 25cm is the best in my case)
    3) export the autoclassified-test-tile and import it again but only the ground points and IMPORTANT: with the option "treat heights as deaths"
    4) autoclassify the imported test-tile a SECOND time with the binsize like before and min height dev 5cm.

    So the surface gets autoclassified "from both sides", which give often better results and a smoother surface-pointcloud.

    5) If the results are ok I do the same with the other tiles and merge the all together by exporting into one big "ground pointcloud".

    Hope it helps and sorry for my poor english...

    Oliver
    TRB426
  • geo72geo72 Global Mapper User Posts: 26Trusted User
    edited December 2013
    forgot: when autoclassifying the points the second time (in step 4) check the option "reset all classified points".

    Thanks!

    Oliver
  • jh93989jh93989 Global Mapper User Posts: 41Trusted User
    edited December 2013
    geo72 wrote: »

    Hope it helps and sorry for my poor english...

    Oliver

    Oliver, thank you for the great post, no problems understanding you English over here. I will have to give your method a go; we have been using lastools for our classification but I had been looking to give Global Mapper another try.
  • manojmanoj Posts: 3
     Please Post here some document explaining what are the options available for LIDAR data classification and what it would do actually technically!!
  • bmg_bobbmg_bob Global Mapper Programmer Posts: 2,061
    edited February 2016
    manoj said:
     Please Post here some document explaining what are the options available for LIDAR data classification and what it would do actually technically!!

    You can find full product documentation, and access to recorded webinars and other training resources here.  Information about the LiDAR Toolbar should be of particular interest.
  • manojmanoj Posts: 3
    Hello, i have tried the automatic ground classification using LIDAR module in GM V.17! The input data was 14ppm,,After doing so many iterations i felt base bin size 2 seems okay,Finally,there were two stages of output acquired

    1.The Embankments missed ground points but the terrain quality was better(base bin size 2 as point spacing's & min height as 0.3 meters)

    2.The Embankments were well fit with ground points but the terrain quality is poor with major ground spikes (base bin size 2 as point spacing's & min height as 0.5 meters)

    max height delta ---60
    Expected terrain slope----5 for both the cases!!
    Someone please help me to get most appropriate result.
     
    Regards,
    Manoj

  • manojmanoj Posts: 3
     Is there any way we could retrieve the las files into each single tile which were imported as a batch file!
  • TRB426TRB426 Global Mapper User Posts: 14
    geo72 said:
    Hello Dionybell!

    I use the autoclassify-broud-option in GM quite often the following way:

    1) split up big pointclouds in tiles of about 5 million points (100m x 100m in my case) with the exportfunction of gm (create tiles when exporting)
    2) take one representative tile and test different parameters with the autoclassify ground-points option. (my favorite binsizes are 50cm and 100cm, for the min height dev I test 5cm, 10cm, 15cm,...,50cm - often 25cm is the best in my case)
    3) export the autoclassified-test-tile and import it again but only the ground points and IMPORTANT: with the option "treat heights as deaths"
    4) autoclassify the imported test-tile a SECOND time with the binsize like before and min height dev 5cm.

    So the surface gets autoclassified "from both sides", which give often better results and a smoother surface-pointcloud.

    5) If the results are ok I do the same with the other tiles and merge the all together by exporting into one big "ground pointcloud".

    Hope it helps and sorry for my poor english...

    Oliver
    Hi Oliver,

    I gave your method a try and it worked very well for me.  I wanted to pass along a shortcut I discovered: 

    Instead of exporting the point cloud and then re-importing it upside down( "treat heights as depths"), you can just invert the point cloud by scaling it by -1. Right click on the layer name, select <Options>, tab "Alter Elevation Values", change Scale Factor from 1 to -1 .  You can then run the autoclassify routine and then just change the scale factor back to 1 to turn right side up again.  

    Cheers! 
    Tim
  • JoskoJosko Posts: 1
    Hi guys,

    anyone did auto-classify ground points in steep terrain like cliffs? Whatever parameter I set I cannot get those cliff points to ground. Manual classification will take forever. If anyone has solved, please share. Thank you!

    Cheers
    Josko
  • JRedellJRedell Posts: 1
    I've been having a heck of a time with classifying data on steep slopes, especially when there's vegetation.  One of the biggest challenges I face right now is dealing with vegetated slopes along roadways honestly.  The sideslopes and ditches are huge time/budget killers for me.
Sign In or Register to comment.