NavVis M6 point cloud data quality

ZEB-REVO, ZEB-HORIZON, BLK2GO, NAVVIS
dhirota
V.I.P Member
V.I.P Member
Posts: 1109
Joined: Sun Nov 01, 2009 11:18 pm
Full Name: Dennis Hirota
Company Details: Sam O. Hirota Inc.
Company Position Title: President
Country: USA
Linkedin Profile: Yes
Location: Hawaii, USA
Has thanked: 5 times
Been thanked: 30 times

Re: NavVis M6 point cloud data quality

Post by dhirota »

Lieven

We post-process all our M6 sensor work at a 5mm grid. Many others for visualization or speed of processing may use 20 mm as sometimes suggested by NavVIS. They may have post-processed the sample at 10mm, which is the default setting. One needs to remember that the resolution difference between 5mm and 20mm is a factor of 64, depending on how you count. Also depending on the post-processing parameter settings, that I did not have the time to discuss on this thread, we use sharp edged corners.

Here is a closer look at a Recap point cloud of the Revit project.

SLICE.JPG

SLICE2_SMALL.jpg

Our Revit Team took it from here.
You do not have the required permissions to view the files attached to this post.
Last edited by dhirota on Thu Jan 30, 2020 9:33 pm, edited 1 time in total.
Dennis Hirota, PhD, PE, LPLS
[email protected]

dhirota
V.I.P Member
V.I.P Member
Posts: 1109
Joined: Sun Nov 01, 2009 11:18 pm
Full Name: Dennis Hirota
Company Details: Sam O. Hirota Inc.
Company Position Title: President
Country: USA
Linkedin Profile: Yes
Location: Hawaii, USA
Has thanked: 5 times
Been thanked: 30 times

Re: NavVis M6 point cloud data quality

Post by dhirota »

GeorgSchroth wrote:
Thu Jan 30, 2020 12:57 pm
smacl wrote:
Wed Jan 29, 2020 9:13 am
I'm guessing this is a combined scan + photogrammetry derived solution to give this balance, as I can't think how a scanner alone would manage this, with no visible profiling pattern or obvious noise.
By the way, I missed mentioning that this is indeed a laser scanner only point cloud :)
Quite some work went to achieving this through probabilistic filtering of the raw measurements.
Shane

I missed your question, but the answer is the same as Georg's, my point cloud at the beginning of this thread does not have any panorama images as well. I was using the new NavVIS extraction feature that allows the export/import to other URL's. Sometimes the color processing is good where it might fool one into thinking that a dense colored point cloud is an image.
Dennis Hirota, PhD, PE, LPLS
[email protected]

GeorgSchroth
I have made 30-40 posts
I have made 30-40 posts
Posts: 35
Joined: Fri Aug 30, 2019 11:05 am
Full Name: Georg Schroth
Company Details: NavVis
Company Position Title: Managing Director
Country: Germany
Linkedin Profile: Yes
Has thanked: 9 times
Been thanked: 8 times

Re: NavVis M6 point cloud data quality

Post by GeorgSchroth »

landmeterbeuckx wrote:
Thu Jan 30, 2020 3:37 pm
Nice one Dennis. I've been playing around a bit with the same data from Navvis as Shane. I've imported it in Pointcab but get wobbly walls and no straight corners.
How do you treat these datasets in Revit. Seems difficult to evaluate the data for consistency.
Hello Lieven,

Thanks for pointing that out. To see whats going on I opened the dataset in pointcab myself but only used a slice of 20cm thickness at a height of about 1.5m to get the two screenshots below. The first one is based on exactly the same data you used and the second is the less filtered version (we call it raw point cloud) with sharper edges. Both clouds are automatically generated with the M6.

Filtered point cloud option
PointCab_Filtered.PNG
Raw point cloud option
PointCab_RAW.PNG
The elevations in the middle of most walls are actually whiteboards of about 2cm thickness. Nevertheless the actual walls look straight in the raw point cloud. Also the corners are crisper and one can recognize the details of the door frame that Shane pointed out in his evaluation.
We will look further into the filtered cloud to make it even more precise.
You do not have the required permissions to view the files attached to this post.

User avatar
landmeterbeuckx
V.I.P Member
V.I.P Member
Posts: 939
Joined: Tue May 01, 2012 5:19 pm
Full Name: Lieven Beuckx
Company Details: Studiebureau Beuckx
Company Position Title: Owner
Country: Belgium
Linkedin Profile: Yes
Has thanked: 24 times
Been thanked: 55 times

Re: NavVis M6 point cloud data quality

Post by landmeterbeuckx »

GeorgSchroth wrote:
Fri Jan 31, 2020 5:27 pm
landmeterbeuckx wrote:
Thu Jan 30, 2020 3:37 pm
Nice one Dennis. I've been playing around a bit with the same data from Navvis as Shane. I've imported it in Pointcab but get wobbly walls and no straight corners.
How do you treat these datasets in Revit. Seems difficult to evaluate the data for consistency.
Hello Lieven,

Thanks for pointing that out. To see whats going on I opened the dataset in pointcab myself but only used a slice of 20cm thickness at a height of about 1.5m to get the two screenshots below. The first one is based on exactly the same data you used and the second is the less filtered version (we call it raw point cloud) with sharper edges. Both clouds are automatically generated with the M6.

Filtered point cloud option
PointCab_Filtered.PNG

Raw point cloud option
PointCab_RAW.PNG

The elevations in the middle of most walls are actually whiteboards of about 2cm thickness. Nevertheless the actual walls look straight in the raw point cloud. Also the corners are crisper and one can recognize the details of the door frame that Shane pointed out in his evaluation.
We will look further into the filtered cloud to make it even more precise.
The last one looks better indeed. Pointcab is my main start when checking pointclouds.
Why would you filter the original one when this looks better?
LSBbvba
Surveying services - 3D Laserscanning
Tel : +32477753126
www.lsbbvba.be
[email protected]

dhirota
V.I.P Member
V.I.P Member
Posts: 1109
Joined: Sun Nov 01, 2009 11:18 pm
Full Name: Dennis Hirota
Company Details: Sam O. Hirota Inc.
Company Position Title: President
Country: USA
Linkedin Profile: Yes
Location: Hawaii, USA
Has thanked: 5 times
Been thanked: 30 times

Re: NavVis M6 point cloud data quality

Post by dhirota »

landmeterbeuckx wrote:
Fri Jan 31, 2020 6:19 pm
GeorgSchroth wrote:
Fri Jan 31, 2020 5:27 pm
landmeterbeuckx wrote:
Thu Jan 30, 2020 3:37 pm
Nice one Dennis. I've been playing around a bit with the same data from Navvis as Shane. I've imported it in Pointcab but get wobbly walls and no straight corners.
How do you treat these datasets in Revit. Seems difficult to evaluate the data for consistency.
Hello Lieven,

Thanks for pointing that out. To see whats going on I opened the dataset in pointcab myself but only used a slice of 20cm thickness at a height of about 1.5m to get the two screenshots below. The first one is based on exactly the same data you used and the second is the less filtered version (we call it raw point cloud) with sharper edges. Both clouds are automatically generated with the M6.
....
We will look further into the filtered cloud to make it even more precise.
The last one looks better indeed. Pointcab is my main start when checking pointclouds.
Why would you filter the original one when this looks better?
Lieven

I will let Georg answer the question on why NavVIS filters the raw point clouds into a grided point cloud.

Your observation of the NavVIS raw point cloud looking better, maybe limited to a single application result with an unknown algorithm producing it. I have never used PointCab, so I maybe speaking out of turn, but the algorithm may produce a better looking line with a denser raw point cloud than a grided less dense object of information. Your familiarity with the software application should guide it's use for the application. Both the raw and grided point clouds are available automatically as indicated by Georg.

I have included some data of a historical high school building that we did in 2018 with an older M3 NavVIS system. It consisted of two stories, half under construction and half completed and occupied.

Screenshot from 2020-01-31 14-54-34.jpg

Screenshot from 2020-01-31 15-00-20.jpg


Raw point clouds=277.7 Mpts
5mm grided point clouds=179.7 Mpts
Images=3,242
Panorama images=538
Area=4 KSqM
Path of travel by M3=714 Meters
Total raw + processed information=87.87 GB
Web browser based on 5mm grided=29.29 GB

My version of the point cloud filtration for the NavVIS system is the reduction of noise; removal of moving objects; probably better color representation of the point clouds when compared to some of the other SLAM systems (I have not checked all in detail); signifcantly smaller information size for visualization.
You do not have the required permissions to view the files attached to this post.
Dennis Hirota, PhD, PE, LPLS
[email protected]

GeorgSchroth
I have made 30-40 posts
I have made 30-40 posts
Posts: 35
Joined: Fri Aug 30, 2019 11:05 am
Full Name: Georg Schroth
Company Details: NavVis
Company Position Title: Managing Director
Country: Germany
Linkedin Profile: Yes
Has thanked: 9 times
Been thanked: 8 times

Re: NavVis M6 point cloud data quality

Post by GeorgSchroth »

landmeterbeuckx wrote:
Fri Jan 31, 2020 6:19 pm

The last one looks better indeed. Pointcab is my main start when checking pointclouds.
Why would you filter the original one when this looks better?
Hello Lieven,

as Dennis pointed out we see different requirements for the output of the M6 so we decided to cover them with two point clouds as of now.
The raw point cloud is mostly for modelling and tools that work on point cloud sections. Tools like PointCab generate the section view by coloring pixels darker if more points are lying at the same vertical axis (like an x-ray image). Therefore walls appear dark and scattered noise/outliers are hardly visible since they are randomly distributed. Thus PointCab and the human eye are basically doing a filtering along the z-coordinate.

The "filtered" point cloud option is only a third in size (GB) and a lot better to understand and analyze in a perspective fly-through since outliers and noise are almost completely removed already. This comes at the cost of slightly rounded edges, but we have an algorithm in the pipeline that is doing the job even better.

User avatar
landmeterbeuckx
V.I.P Member
V.I.P Member
Posts: 939
Joined: Tue May 01, 2012 5:19 pm
Full Name: Lieven Beuckx
Company Details: Studiebureau Beuckx
Company Position Title: Owner
Country: Belgium
Linkedin Profile: Yes
Has thanked: 24 times
Been thanked: 55 times

Re: NavVis M6 point cloud data quality

Post by landmeterbeuckx »

GeorgSchroth wrote:
Mon Feb 03, 2020 8:49 am
landmeterbeuckx wrote:
Fri Jan 31, 2020 6:19 pm

The last one looks better indeed. Pointcab is my main start when checking pointclouds.
Why would you filter the original one when this looks better?
Hello Lieven,

as Dennis pointed out we see different requirements for the output of the M6 so we decided to cover them with two point clouds as of now.
The raw point cloud is mostly for modelling and tools that work on point cloud sections. Tools like PointCab generate the section view by coloring pixels darker if more points are lying at the same vertical axis (like an x-ray image). Therefore walls appear dark and scattered noise/outliers are hardly visible since they are randomly distributed. Thus PointCab and the human eye are basically doing a filtering along the z-coordinate.

The "filtered" point cloud option is only a third in size (GB) and a lot better to understand and analyze in a perspective fly-through since outliers and noise are almost completely removed already. This comes at the cost of slightly rounded edges, but we have an algorithm in the pipeline that is doing the job even better.
@GeorgSchroth
@dhirota

I have no unlimited software packages to my disposal for pointcloud processing. ;)

Thanks for the insights. Great to see there happening a lot in the scanning world, to difficult to follow-up on everything.
I must say the visuals look awesome!
LSBbvba
Surveying services - 3D Laserscanning
Tel : +32477753126
www.lsbbvba.be
[email protected]

User avatar
smacl
Global Moderator
Global Moderator
Posts: 595
Joined: Tue Jan 25, 2011 5:12 pm
Full Name: Shane MacLaughlin
Company Details: Atlas Computers Ltd
Company Position Title: Managing Director
Country: Ireland
Linkedin Profile: Yes
Location: Ireland
Has thanked: 58 times
Been thanked: 58 times
Contact:

Re: NavVis M6 point cloud data quality

Post by smacl »

GeorgSchroth wrote:
Mon Feb 03, 2020 8:49 am
The "filtered" point cloud option is only a third in size (GB) and a lot better to understand and analyze in a perspective fly-through since outliers and noise are almost completely removed already. This comes at the cost of slightly rounded edges, but we have an algorithm in the pipeline that is doing the job even better.
What you're doing here with the filtering is really what impresses me most with this solution. Removing two thirds of the data and maintaining an excellent representation of the scanned surface is very valuable given the data bloat we're currently seeing in the industry. Any plans to produce this as an add on for other manufacturers scanners? I can think of quite a lot of people who would jump on such a solution.

GeorgSchroth
I have made 30-40 posts
I have made 30-40 posts
Posts: 35
Joined: Fri Aug 30, 2019 11:05 am
Full Name: Georg Schroth
Company Details: NavVis
Company Position Title: Managing Director
Country: Germany
Linkedin Profile: Yes
Has thanked: 9 times
Been thanked: 8 times

Re: NavVis M6 point cloud data quality

Post by GeorgSchroth »

smacl wrote:
Mon Feb 03, 2020 11:47 am
What you're doing here with the filtering is really what impresses me most with this solution. Removing two thirds of the data and maintaining an excellent representation of the scanned surface is very valuable given the data bloat we're currently seeing in the industry. Any plans to produce this as an add on for other manufacturers scanners? I can think of quite a lot of people who would jump on such a solution.
Fully agree. Parts of our algorithms assume that it is a mobile scanner that observes objects from multiple different perspectives, but some could also be applied to TLS. We are thinking of providing a processing service as part of IndoorViewer in the future and the optimization of TLS clouds could be a great extension :)

dhirota
V.I.P Member
V.I.P Member
Posts: 1109
Joined: Sun Nov 01, 2009 11:18 pm
Full Name: Dennis Hirota
Company Details: Sam O. Hirota Inc.
Company Position Title: President
Country: USA
Linkedin Profile: Yes
Location: Hawaii, USA
Has thanked: 5 times
Been thanked: 30 times

Re: NavVis M6 point cloud data quality

Post by dhirota »

GeorgSchroth wrote:
Mon Feb 03, 2020 11:57 am
smacl wrote:
Mon Feb 03, 2020 11:47 am
What you're doing here with the filtering is really what impresses me most with this solution. Removing two thirds of the data and maintaining an excellent representation of the scanned surface is very valuable given the data bloat we're currently seeing in the industry. Any plans to produce this as an add on for other manufacturers scanners? I can think of quite a lot of people who would jump on such a solution.
Fully agree. Parts of our algorithms assume that it is a mobile scanner that observes objects from multiple different perspectives, but some could also be applied to TLS. We are thinking of providing a processing service as part of IndoorViewer in the future and the optimization of TLS clouds could be a great extension :)
Georg

We have a project that I displayed earlier in this thread, with scans by Z+F 5016 and Riegl VZ400i sensors with our M6. If I imported them as E57 scans, would they be filtered as the M6?
Dennis Hirota, PhD, PE, LPLS
[email protected]

Post Reply

Return to “Dynamic Mapping Scanners”