Any Automatic Registration options for SLAM scans (GeoSLAM)
-
- I have made 10-20 posts
- Posts: 10
- Joined: Tue Feb 18, 2020 2:29 pm
- 4
- Full Name: DMag
- Company Details: Scan3D
- Company Position Title: Cofounder
- Country: us
- Has thanked: 6 times
- Been thanked: 1 time
Any Automatic Registration options for SLAM scans (GeoSLAM)
Hi Everybody,
I have worked for a company using GeoSLAM scanners for the past 5 years (Zeb1 and now Revo RT) and admittedly fell into the old trap of just doing what we know and may have fallen a little behind some of the advances that this industry is constantly producing.
I am looking to the community to see if anybody is aware of new software, or recent advances in the GeoSLAM HUB, or other methods that people may use to more automatically register SLAM scans, the way that something like Cyclone works with the Leica scanners for example.
We were trained in the merge feature of HUB last year, so we are aware of that method, but as of the training last year and all the info I can find online, it is still a manual process of getting them relatively aligned and then reprocessing the scans together. Has GeoSLAM made any advances in the HUB software that can do this more automatically?
Are there other programs out there than can help expedite this registration process for SLAM scans? I should point out I have tried finding answers online but I think we SLAMmers are still too fringe for the internet to care - everything that came back was just for automatically registering terrestrial scans. Speaking of which, I have been given the (possibly false) impression that Autodesk ReCap Pro can do some automatic registration of SLAM scans but it requires at least one "structured" scan to start that process and we just don't have that capability at this time. Has anybody ever tried that and does it work or is it just a misunderstanding? Also, in years past, we tried some of the tools in CloudCompare to help automatically register, but they weren't terribly helpful, so we ended up just doing manual static alignments.
Lastly, this conversation caught on fire in our company because of a recent project we did, and I will be making a separate post about that to discuss workflows for scanning hotels and apartment buildings, so keep your eyes open if you're interested.
Thank you for any and all insights you may have (even if the answer is "no," at least we will know where things stand).
All the best,
Doug
I have worked for a company using GeoSLAM scanners for the past 5 years (Zeb1 and now Revo RT) and admittedly fell into the old trap of just doing what we know and may have fallen a little behind some of the advances that this industry is constantly producing.
I am looking to the community to see if anybody is aware of new software, or recent advances in the GeoSLAM HUB, or other methods that people may use to more automatically register SLAM scans, the way that something like Cyclone works with the Leica scanners for example.
We were trained in the merge feature of HUB last year, so we are aware of that method, but as of the training last year and all the info I can find online, it is still a manual process of getting them relatively aligned and then reprocessing the scans together. Has GeoSLAM made any advances in the HUB software that can do this more automatically?
Are there other programs out there than can help expedite this registration process for SLAM scans? I should point out I have tried finding answers online but I think we SLAMmers are still too fringe for the internet to care - everything that came back was just for automatically registering terrestrial scans. Speaking of which, I have been given the (possibly false) impression that Autodesk ReCap Pro can do some automatic registration of SLAM scans but it requires at least one "structured" scan to start that process and we just don't have that capability at this time. Has anybody ever tried that and does it work or is it just a misunderstanding? Also, in years past, we tried some of the tools in CloudCompare to help automatically register, but they weren't terribly helpful, so we ended up just doing manual static alignments.
Lastly, this conversation caught on fire in our company because of a recent project we did, and I will be making a separate post about that to discuss workflows for scanning hotels and apartment buildings, so keep your eyes open if you're interested.
Thank you for any and all insights you may have (even if the answer is "no," at least we will know where things stand).
All the best,
Doug
- kpob46
- I have made 100> posts
- Posts: 102
- Joined: Sat Aug 16, 2014 5:43 pm
- 9
- Full Name: Kyle OBrien
- Company Details: Nutrien
- Company Position Title: Mine Technologist
- Country: Canada
- Location: Saskatoon, SK
- Has thanked: 18 times
- Been thanked: 23 times
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Hi Doug,
I also agree that a cloud to cloud method isn't the most reliable way for registering GeoSLAM scan data. I have found success by using three dimensional targets (spheres) to successfully register separate scans together. Flat targets are not desirable since the surface noise from SLAM scanners on a flat plane makes the target extraction inaccurate. The three dimensional targets helps distribute that error much better.
I also agree that a cloud to cloud method isn't the most reliable way for registering GeoSLAM scan data. I have found success by using three dimensional targets (spheres) to successfully register separate scans together. Flat targets are not desirable since the surface noise from SLAM scanners on a flat plane makes the target extraction inaccurate. The three dimensional targets helps distribute that error much better.
-
- I have made 10-20 posts
- Posts: 10
- Joined: Tue Feb 18, 2020 2:29 pm
- 4
- Full Name: DMag
- Company Details: Scan3D
- Company Position Title: Cofounder
- Country: us
- Has thanked: 6 times
- Been thanked: 1 time
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Hi Kyle,
Thank you so much for your reply - it is nice to hear from others with similar issues. I am curious about your suggestion for using the sphere targets. What software and or methods do you use to integrate these targets into your registration process? Are you somehow leveraging the spheres in the HUB merging process? Can you describe your workflow a little bit? Thank you so much.
I'm excited to hear you may have a solution for this!
Thank you Kyle.
Best,
Doug
- kpob46
- I have made 100> posts
- Posts: 102
- Joined: Sat Aug 16, 2014 5:43 pm
- 9
- Full Name: Kyle OBrien
- Company Details: Nutrien
- Company Position Title: Mine Technologist
- Country: Canada
- Location: Saskatoon, SK
- Has thanked: 18 times
- Been thanked: 23 times
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
I actually do not use the hub at all. I prefer to have the actual points and go from there. I typically use 3DReshaper or as its called now, Leica Cyclone 3D. From there, I can extract the spherical targets and register separate points clouds together. I have also used RiSCAN to import the data and perform the same work flow. I don't have an issue with C2C registration with FARO or Riegl scans, but I find the surface noise with SLAM scanners not precise enough for my workflows for C2C. I believe you can use CloudCompare with the same workflow.
- tadol
- V.I.P Member
- Posts: 126
- Joined: Wed Dec 14, 2011 12:32 am
- 12
- Full Name: Tad Laird
- Company Details: Koppa Target Spheres
- Company Position Title: Owner
- Country: USA
- Linkedin Profile: No
- Location: Oakland, California, USA
- Has thanked: 2 times
- Been thanked: 4 times
- Contact:
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Excuse me if I am mis-reading this, but are you just trying to say that you like my balls?
I got no problems with that at all - I like them too!
But that might get you an extra discount -
Tad
I got no problems with that at all - I like them too!
But that might get you an extra discount -
Tad
http://www.KoppaTargets.com
Inexpensive Laser Scanning Targets & Accessories
for Architecture, Engineering, & Forensics
"We're right on targets!"
Inexpensive Laser Scanning Targets & Accessories
for Architecture, Engineering, & Forensics
"We're right on targets!"
-
- I have made 10-20 posts
- Posts: 10
- Joined: Tue Feb 18, 2020 2:29 pm
- 4
- Full Name: DMag
- Company Details: Scan3D
- Company Position Title: Cofounder
- Country: us
- Has thanked: 6 times
- Been thanked: 1 time
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Great tips Kyle; thank you for sharing some ideas to pursue. Looking forward to trying these out.kpob46 wrote: ↑Wed Jun 10, 2020 8:20 pm I actually do not use the hub at all. I prefer to have the actual points and go from there. I typically use 3DReshaper or as its called now, Leica Cyclone 3D. From there, I can extract the spherical targets and register separate points clouds together. I have also used RiSCAN to import the data and perform the same work flow. I don't have an issue with C2C registration with FARO or Riegl scans, but I find the surface noise with SLAM scanners not precise enough for my workflows for C2C. I believe you can use CloudCompare with the same workflow.
All the best,
Doug
-
- I have made 10-20 posts
- Posts: 10
- Joined: Thu Jun 29, 2017 1:18 pm
- 6
- Full Name: Charlie Cropp
- Company Details: Correvate UK
- Company Position Title: Inbound Sales Manager
- Country: UK
- Linkedin Profile: Yes
- Has thanked: 6 times
- Been thanked: 1 time
- Contact:
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Hi Doug,
We're having some pretty good successes with our VERCATOR Cloud Registration service. Yes we were born out of the static market but have an increasing user base within the mobile/robotic/SLAM market.
Have a look at www.vercator.com and signup for a free trial!
BR, Charlie
We're having some pretty good successes with our VERCATOR Cloud Registration service. Yes we were born out of the static market but have an increasing user base within the mobile/robotic/SLAM market.
Have a look at www.vercator.com and signup for a free trial!
BR, Charlie
- Daniel Wujanz
- V.I.P Member
- Posts: 227
- Joined: Fri Aug 24, 2018 11:26 am
- 5
- Full Name: Daniel Wujanz
- Company Details: technet GmbH
- Company Position Title: 3D Laser Scanning Specialist
- Country: Germany
- Linkedin Profile: Yes
- Location: Berlin
- Has thanked: 206 times
- Been thanked: 188 times
- Contact:
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Dear Doug,
first of all I should point out that "ordinary" registration algorithms will be able to give you some results.
BUT: the problem that you have in processing kinematic scans is slightly different compared to registering static scans. In simple terms you're determining heaps of small registrations while you're ambling through the object space with your kinematic scanner. That means that your network stability decreases over time comparable to long, uncontrolled traverses with static scanners (or any other geodetic instrument) since the registration error accumulates. As a consequence, the inner geometry of the point cloud will start to swing like a cows udder and typically does not reflect the geometry of your object as you would expect.
Hence, your aim is to a take large hammer to straighten up your point cloud - which is referred to as homogenisation.
So, which options do you have? Again, our best friend is redundancy:
- Add loops - as all kinematic scanner manufacturers suggest
- add ground control points
- add static scans
NavVis and Optech (I simply don't know about GeoSLAM - can somebody help me out here?) appear to have quite good solutions for this problem yet are not able (to the best of my knowledge - please let me know if I'm wrong) to process kinematic & static scans and other geodetic sources in one common block adjustment. The crux is the following: you can only improve your point cloud in areas where redundant information is given.
We're currently working on a prototype that performs a combined block adjustment at which you can chuck data from different sources. But again, it will take some time until one reaches a stage where you can market something. This week we were testing different sources and played around with different mobile mapping systems and will start moving towards smaller handheld systems such as DotProduct and / or GeoSLAM.
I've attached an image that helps you to get an idea of the problem. What you're seeing is a horizontal section through a small tunnel which was scanned six times with a mobile mapping system over the course of a day (in order to have notable GNSS-variations). In addition, the tunnel was scanned with a static scanner. The largest deviations that you can see in the left part of the image are around 12 cm. The right part shows the intersected point cloud after homogenisation with our prototype.
Cheers
Daniel
first of all I should point out that "ordinary" registration algorithms will be able to give you some results.
BUT: the problem that you have in processing kinematic scans is slightly different compared to registering static scans. In simple terms you're determining heaps of small registrations while you're ambling through the object space with your kinematic scanner. That means that your network stability decreases over time comparable to long, uncontrolled traverses with static scanners (or any other geodetic instrument) since the registration error accumulates. As a consequence, the inner geometry of the point cloud will start to swing like a cows udder and typically does not reflect the geometry of your object as you would expect.
Hence, your aim is to a take large hammer to straighten up your point cloud - which is referred to as homogenisation.
So, which options do you have? Again, our best friend is redundancy:
- Add loops - as all kinematic scanner manufacturers suggest
- add ground control points
- add static scans
NavVis and Optech (I simply don't know about GeoSLAM - can somebody help me out here?) appear to have quite good solutions for this problem yet are not able (to the best of my knowledge - please let me know if I'm wrong) to process kinematic & static scans and other geodetic sources in one common block adjustment. The crux is the following: you can only improve your point cloud in areas where redundant information is given.
We're currently working on a prototype that performs a combined block adjustment at which you can chuck data from different sources. But again, it will take some time until one reaches a stage where you can market something. This week we were testing different sources and played around with different mobile mapping systems and will start moving towards smaller handheld systems such as DotProduct and / or GeoSLAM.
I've attached an image that helps you to get an idea of the problem. What you're seeing is a horizontal section through a small tunnel which was scanned six times with a mobile mapping system over the course of a day (in order to have notable GNSS-variations). In addition, the tunnel was scanned with a static scanner. The largest deviations that you can see in the left part of the image are around 12 cm. The right part shows the intersected point cloud after homogenisation with our prototype.
Cheers
Daniel
You do not have the required permissions to view the files attached to this post.
-
- V.I.P Member
- Posts: 246
- Joined: Sat Jul 07, 2018 9:19 pm
- 5
- Full Name: Andrew
- Company Details: NDC Surveys
- Company Position Title: Surveyor
- Country: Uk
- Linkedin Profile: No
- Has thanked: 56 times
- Been thanked: 22 times
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
wow - great milk references!!Daniel Wujanz wrote: ↑Thu Jun 11, 2020 2:24 pm
inner geometry of the point cloud will start to swing like a cows udder and typically does not reflect the geometry of your object as you would expect.
Hence, your aim is to a take large hammer to straighten up your point cloud - which is referred to as homogenisation.
-
- I have made 10-20 posts
- Posts: 10
- Joined: Tue Feb 18, 2020 2:29 pm
- 4
- Full Name: DMag
- Company Details: Scan3D
- Company Position Title: Cofounder
- Country: us
- Has thanked: 6 times
- Been thanked: 1 time
Re: Any Automatic Registration options for SLAM scans (GeoSLAM)
Thank you Daniel,Daniel Wujanz wrote: ↑Thu Jun 11, 2020 2:24 pm BUT: the problem that you have in processing kinematic scans is slightly different compared to registering static scans. In simple terms you're determining heaps of small registrations while you're ambling through the object space with your kinematic scanner. That means that your network stability decreases over time comparable to long, uncontrolled traverses with static scanners (or any other geodetic instrument) since the registration error accumulates. As a consequence, the inner geometry of the point cloud will start to swing like a cows udder and typically does not reflect the geometry of your object as you would expect.
That has always been my understanding of the problem as well - I guess I was curious to see if great minds like yourself had come up with any solutions over the recent years. It can be hard for some people to understand that these systems have inherent limitations like you described. Like I had said in my OP, even if the answer is "no," or "not yet," at least we know where things stand. But it is great seeing the kind of progress you were describing in your tests!
Thank you for the great insight and also keep up the exciting work!
Best,
Doug