Fast and Efficient Point Cloud Classification

[Lurker]

Hello, I use Pix4dMapper pro for drone post processing

 

but for area with vegetation or structural object, we need classify the point cloud data to get the bare land data, and right now I'm unsatisfied with existing point cloud classification in Pix4dmapper pro

 

don't have time to try One Button or Agisoft, but anyone here have the experience classifying point cloud, fast and efficient? 

 

prefer a tools that almost automatically classify point cloud data between object and bare land, I'm too lazy to do manual things    

[osurv]

Hi Lurker,

 

There are a few other softwares that will classify bare earth, but none will do a 100% efficient job automatically without some manual editing afterwards.

 

Eg:

Carlson Point Cloud

Global Mapper with LiDAR module

Terrasolid Terrascan

 

My recomendation would be Terrascan. It is the industry standard for Aerial LiDAR processing and does a fast and efficient job on auto classifying ground points. It also allows you to alter the parameters depending on the terrain and objects that you have in your dataset.

 

Osurv

[Lurker]

ah thanx for the feedback, yeah I know that, there is no 100% automatic for point cloud classification, but at least 90% or more, and next 10%, we do manual for refining it. 

 

In pix4dmapper , there are 3 variable for point cloud classification, :

 

1. Min object length

2. Max Object length

3. Min Object height

 

and somehow, I feel this is far from complete 

 

I will take a look for Terracan

[Lurker]

update , seems terrascan is a plugin for microstation, ah no standalone software for this ?

[osurv]

Correct, Terrascan uses Microstation Platform. But there is good reason for this - the Vortex point cloud engine is one of the most efficient for dealing with large point clouds.

 

Global Mapper / LiDAR Module is standalone. There is also VRMesh but don't know how effective this is.

 

You might be dissapointed if you want 90% auto classiifcation, especially with photo produced point clouds. Most softwares have a hard time dealing with this sort of data.

 

OSurv

[Lurker]

ah so every software have pretty much same quality result when running automatic point cloud classification?

[darksabersan]

90% is maybe a bit optimistic and anyway, whatever software you are going to use, some readjustment has to be performed behind.

I've tested in the past the version 6 of VRMesh and it was quite hard to do. Not sure about the new release and I can just guess that they have bring improvement.

 

Personally I love Global Mapper.

 

 

darksabersan.

[Lurker]

ahahaha 90% just my crude estimation, at least Im not completely manual when dealing with point cloud classification.   

 

BTW anyone here already compare the result for three UAV software mentioned above, especially in point cloud classification? Pix4D, One Button and Agisoft

 

I try search on their knowledge base. pix4d is the easiest, Agisoft have a complex setting, but I cant find bout one button

[Lurker]

osurv, 

just download terrascan, is it possible to install it on to Bentley Map ? or i should using ordinary bentley microstation?

[Lurker]

update :

 

I just got nice discussion related to this :

 

 

Processing/classifying Point Cloud data requires knowing how to use the software and experience regardless of the software being used. Vegetation is why people use LiDAR because you will likely get points on the ground. LiDAR is an active sensor which emits a pulse and then receives it with the time and angle used to determine where the point is in real space. Creating point clouds from photos is not LiDAR that is why vegetation is a problem and why Photogrammetry (an older technology) is being supplanted by LiDAR. Using photos requires the same point on the "ground" to appear in at least two photos shot from different angles so that a stereo image can be created to determine the Z value of the point. Orthorectification of a single photo is required to determine the point's X and Y position.

Multiple returns do provide more information that can be derived from the data but is not required to determine the ground surface. Even with multiple returns the last return is not always on the actual ground. So it is not as simple as saying if I have all last returns I have the ground surface, You would have every solid object in that surface as well including roof tops, water towers, heavy vegetation, etc.

Point cloud classification software regardless of the vendor takes an investment of time to learn how to use them. No vendor has the "one button push" and produce a ground surface solution. They all have some semi automatic tools to help classify points but they all require manual cleanup for good results.   

 

 

source :

http://diydrones.com/forum/topics/uav-generated-point-cloud-classification?id=705844%3ATopic%3A1879523&page=3#comments

one of member propose this step :

 

 

Filter ground points in Photoscan then export point cloud (ground only) to LAS format.

Then from Whitebox GAT tools run the next sequence Bare-Earth DEM (usually with 0.5m resolution) > Remove Off-Terrain Objects > Fill Missing Data Holes.

 

with some other propose :

 

1. global mapper

2. Recap 360

3. Terrascan

4. CANUPO

 

I think i will try Terrascan, Global Mapper and CANUPO

 

[Art409]

there is a couple of tools really interestng: Lidar analyst and feature analyst from Overwatch technologies (http://www.vls-inc.com/)

So far I can´t test it, but seems one of the best of the market.

[Lurker]

Hello Art409, 

the most disturbing part is that I think UAV using camera is different than LIDAR

I dont know that all lidar software can make a better classification in my case.

[osurv]

osurv, 

just download terrascan, is it possible to install it on to Bentley Map ? or i should using ordinary bentley microstation?

Lurker,

 

Sorry for the late reply... yes, I think you will need the ordinary Microstation, but if you already have Map, maybe you can try first...

 

I just remember one other classification software that is free called FUSION and created by the US Forestry Dept (I think..). I have never used it and may not be quite user friendly.

 

That quote from the DIY Drone forum is 100% correct. In my Company, experience tells us auto procesing of LiDAR point cloud data gives us about a 70-80% correct bare earth classification. Auto processing of UAV photo point cloud gives us around 30-40% correct classification of similar terrain.

[Lurker]

That quote from the DIY Drone forum is 100% correct. In my Company, experience tells us auto procesing of LiDAR point cloud data gives us about a 70-80% correct bare earth classification. Auto processing of UAV photo point cloud gives us around 30-40% correct classification of similar terrain.

 

I just finished classifying my cloud point data with pix4d and global mapper, and yes you are correct, the result is poor

even when I combined it with manual classification, the DTM resulted is bad, 

 

thanx for the info, I will take a look FUSION

[Art409]

Hi Lurker! I'm sorry too for the late reply.

As you I trying to classify UAV point clouds, and yes, Lidar derivates are different.

Lidar got echo numer (or full wave proccesing), and Image derivates got 3 (or more) color bands.

 

My main goal is extract buiding from cloud points, but I'm sure that I'm far behind your steps.

I got soid theroetical background, but I almost got not time for test,

(and I don't plan to trask/deliver anything 'till I'm not sure of the process)

 

Please take a look to one or both "Lidar analyst" and "feature analyst" 

Looks promising

 

Art