Best fit of cylinders

Added by Vandenbroeck Yannick over 3 years ago

Hello all,

I have a question about the best fit of cylinders on Lidar data with a lot of noise.

I have a dataset, measured with a mobile laserscanner (ZEB1). When I compare the DBH with data of a fixed laserscan (Leica P40), the ZEB1 data consequently overestimates the DBH.
In fact, the ZEB data has a lot of noise (see picture included).

Is the modeled cylinder a best fit (mean of the data), or models it the inner/outer diameter of the trees? If it models the outer diameter, it would be a good explanation for the consequent overestimation.

Thanks in advance!

Boom_Snede_P40_ZEB_VGL.PNG (63.6 KB) Boom_Snede_P40_ZEB_VGL.PNG P40 (Left) vs. ZEB1 (right)

Replies (5)

RE: Best fit of cylinders - Added by Hackenberg Jan over 3 years ago

Hello Yannick

good and interesting question at first! I am Jan and joined the Computree recently. I am developing a new plugin for the platform, see my old open source stand alone software SimpleTree here. This one is though unsupported, as
I focus on Computree.
I am guessing that the plugin you use is relying on NLS fit, which in fact does not choose the median, it minimizes the sum of squares of the residuals. In general, NLS is known to produce
the best possible fit in a perfect world, i.e. no noise. But with noise RANSAC fit is supposed to be more robust. I am not sure, if this holds true for your data, as the noise here is quite regular.

If you can provide me with either:
- two isolated tree clouds (ZEB1 and Leica P40). I need to know the height above ground in addition, at which those clouds were cut. I can run my standalone product which relies on RANSAC today
and see if the results of the two different clouds are significantly different. Rather than DBH you will get complete tree models and total volume, the DBH is one additional parameter. I am not sure
if my plugin/standalone is able to model the complete trees, as there is some constrains on the quality of the cloud still. But I am hugely interested in this test, as I never encountered mobile scan data before.
- two isolated sub clouds just containing something like a 20 cm slice around the DBH (1,2m to 1,4m). For a baby step analysis this is even better. There is 6 different RANSAC routines available in PCL
and I can test them out quickly


RE: Best fit of cylinders - Added by Vandenbroeck Yannick over 3 years ago

Dear Jan,

Firstly I would like to thank you for your fast answer.

I’ve send you a personal e-mail with a subset of the pointclouds.

Kind regards.

RE: Best fit of cylinders - Added by Bauwens Sébastien over 3 years ago


I did the same work: comparing ZEB1 with FARO Focus 3D data. I used computree to get the diameters of trees and I have better results with ZEB1 than multiscan of my 8 plots! The least square fitting is then giving satfisfactory results.
I did a first presentation in forestsat 2014 on this topic (the FARO data were first giving betters results in 4 of my plots) and I am currently submitting a paper on this topic.
The forestSat presentation:
And attached a figure showing that the noise has a “gaussian shape”.

RE: Best fit of cylinders - Added by Vandenbroeck Yannick over 3 years ago

Hi Sébastien,

Thanks for your information.

Have you already a conclusion about which factors would cause better results with ZEB1 than Faro Focus 3D? Did you take the taped diameters as reference data in this case?
I’ve calculated following results: bias of 1,5 cm in planimetric position between both systems (multiscan P40 versus HMLS ZEB1), and 2,4 cm in calculated DBH (Computree, same parameters for both datasets). This values are the mean results of 25 trees (DBH > 10 cm). I’ve chosen to determine the Leica P40 multiscan as reference data in my research instead of manually measured diameters, seen the high accuracy of P40.

Kind regards.

RE: Best fit of cylinders - Added by Bauwens Sébastien over 3 years ago

Yep, I take the tape diameter as reference as comparing dbh estimates from 2 scanning methods will include the random and systemtic errors from the processing methods. On the other hand, tape measurement is not free of measurment errors, but It is still a reference.