finding systematic misalignment in points from PTX files, when angles/range calculated.

Please post discussions that do not fit into any other category.
Post Reply
psiplace
I have made 10-20 posts
I have made 10-20 posts
Posts: 19
Joined: Thu Apr 09, 2009 11:44 pm
Full Name: simon place
Company Details: withheld
Company Position Title: withheld
Country: UK
Linkedin Profile: No
Has thanked: 6 times
Contact:

finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by psiplace »

i am trying to develop some code to find solid geometry in point clouds, so lighting/collision stuff works.

for this i really need to get the angles and ranges, rather than the Cartesian locations, and am recalculating these from some old PTX files.

what i find confuses me, and is happening i various example files from various sources.

this is what i get for continuous surfaces....(as expected a regular grid)
Screenshot_2020-03-19 WebGL Pointcloud Viewer.png
(range is setting the grey colour)

but when a large change in range, at an edge, the angles change depend on range.
Screenshot_2020-03-19 WebGL Pointcloud Viewer(1).png
(the nearer, black squares are shifted to the side, but are exactly aligned up/down)

i guess this could be due to a calculation mistake, but its hard to understand since the points themselves produce fine looking images and the misalignment is really small and dependent on the range, its like there is a parallax, as if the scanner location in the file isn't quite right.

i tested and this effect is significantly larger than a rounding error.

if anyone has an idea whats happening/ seen it before, i would be very interested.

i found out several of the PTX files were taken with a Scanstation2 (others may have been but i don't know.)
You do not have the required permissions to view the files attached to this post.
User avatar
Scott.Warren
V.I.P Member
V.I.P Member
Posts: 204
Joined: Tue Jan 30, 2018 3:14 pm
Full Name: Scott Warren
Company Details: WSP USA
Company Position Title: Project Manager - Geomatics Engineer
Country: USA
Linkedin Profile: Yes
Location: Merrimack, NH, USA
Has thanked: 18 times
Been thanked: 47 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by Scott.Warren »

Read your post, kinda confused, so here are some thoughts/questions. Forgive me if I missed your intention:

Angle and ranges of a group of point taken, in my mind, would be different locally if there was more than 1 scan location capturing that group of points, which can be the case with PTX.

Have you thought of using point normals for your calculations? They are not included in a PTX file, but can be calculated in Cyclone and some other software packages. These point normals would give you a localized plane for every point in the cloud. Food for thought. Being able to sample data and only select points based on their normals may help find "solid" objects for your collision, but hey, i have no clue what software you are using.

You are using PTX files you say you don't have a whole lot of info for. So if the PTX file had inherent error in the registration process (putting all the scans together properly), how would you know? There are many different techniques for registration. Its a good scanner, but getting up in years.

We would need hard numbers of how far off things are before we could reasonably suggest a reason. An 1/2" to some is a lot of error, to others its acceptable.
psiplace
I have made 10-20 posts
I have made 10-20 posts
Posts: 19
Joined: Thu Apr 09, 2009 11:44 pm
Full Name: simon place
Company Details: withheld
Company Position Title: withheld
Country: UK
Linkedin Profile: No
Has thanked: 6 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by psiplace »

> Angle and ranges of a group of point taken, in my mind, would be different locally if there was more than 1 scan location capturing that group of points, which can be the case with PTX.

sorry, forgot to say, i'm dealing with each scan in the PTX separately.

> Have you thought of using point normals for your calculations? They are not included in a PTX file, but can be calculated in Cyclone and some other software packages. These point normals would give you a localized plane for every point in the cloud. Food for thought. Being able to sample data and only select points based on their normals may help find "solid" objects for your collision, but hey, i have no clue what software you are using.

as i understand it 'point normals' require some algorithm to make a determination as to when points are actually on a continuous surface, (probably improved/requiring all points/scans simultaneously) what i'm trying is what i think might be a new method, it requires going back to individual scans and processing them in a new way, normals could come out at the end but not as an intermediate step. as a side effect of going 'raw' there's not much software to help.

> You are using PTX files you say you don't have a whole lot of info for. So if the PTX file had inherent error in the registration process (putting all the scans together properly), how would you know? There are many different techniques for registration. Its a good scanner, but getting up in years.

this is about individual scans, AFAIK any registration would effect the points and origin identically and so the, relative, alignment, on the spherical grid, shouldn't be effected. which it isn't except for this 'parallax' effect.

i guess if the registration was only computing the transform to the nearest 1mm, say, that would explain it, but then i should see misalignments with random displacements, scan to scan, when they all appear the same.

i'll try to find out, where i can, what registration software was used.

seems like i might need to try to recalculate the origin, assuming vertical/horizontal alignment in the scan.

> We would need hard numbers of how far off things are before we could reasonably suggest a reason. An 1/2" to some is a lot of error, to others its acceptable.

very roughly, an error/offset in the position of the origin, relative to the points, of about 1mm would explain this. however it becomes noticeable through projection, (magnifying the effect by the ratio of the distances) a single scan with some elements 10x closer than others, will then produce a relative offset of something of the order of 1cm. from the scans i have this is clearly much more than the absolute error of the scans. maybe x10.
jedfrechette
V.I.P Member
V.I.P Member
Posts: 973
Joined: Mon Jan 04, 2010 7:51 pm
Full Name: Jed Frechette
Company Details: Lidar Guys
Company Position Title: CEO and Lidar Supervisor
Country: USA
Linkedin Profile: Yes
Location: Albuquerque, NM
Has thanked: 13 times
Been thanked: 52 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by jedfrechette »

From your images I can't quite tell what I'm looking at. Could you maybe annotate them to indicate the offset you'r trying to describe? If you can share any of the files I'd be interested to take a look.

In principle, any registrations shouldn't affect this. The registration transforms should just be stored in the ptx headers, so if you ignore them the coordinates should be relative to the scan origin (0,0,0). Of course in practice, there's no guarantee that's true. For example, we've got tools that will bake the transforms into the point coordinates and set the header transforms to identity. I wouldn't be surprised at all if other similar code is out there in the wild.
Jed
User avatar
Scott.Warren
V.I.P Member
V.I.P Member
Posts: 204
Joined: Tue Jan 30, 2018 3:14 pm
Full Name: Scott Warren
Company Details: WSP USA
Company Position Title: Project Manager - Geomatics Engineer
Country: USA
Linkedin Profile: Yes
Location: Merrimack, NH, USA
Has thanked: 18 times
Been thanked: 47 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by Scott.Warren »

also, the SS2 isn't quoted as being a 1mm instrument. If your expectation that all the data should marry to 1mm, don't be surprised to see deviations of >5+mm. Its is quoted at ~6mm 3D accuracy for a single measurement. Will usually be better at shorter ranges, but just be mindful of this.
psiplace
I have made 10-20 posts
I have made 10-20 posts
Posts: 19
Joined: Thu Apr 09, 2009 11:44 pm
Full Name: simon place
Company Details: withheld
Company Position Title: withheld
Country: UK
Linkedin Profile: No
Has thanked: 6 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by psiplace »

> From your images I can't quite tell what I'm looking at. Could you maybe annotate them to indicate the offset you'r trying to describe? If you can share any of the files I'd be interested to take a look.

both are snapshots from an rendering using squares for points, located in an x-y plane using a scans angles, (theta, phi in spherical coords) z (range) sets a squares greyness, (not its position) darker is nearer,

first image shows a sample with a roughly perpendicular plane, (scan points roughly same range) so same colour, and show very regular grid, as expected.

second image shows a step change in range, black much closer, it shows these points displaced by the same amount to the side, so the horizontal angles (vertical passes) are not aligned.

> In principle, any registrations shouldn't affect this.

agreed

> The registration transforms should just be stored in the ptx headers, so if you ignore them the coordinates should be relative to the scan origin (0,0,0). Of course in practice, there's no guarantee that's true. For example, we've got tools that will bake the transforms into the point coordinates and set the header transforms to identity. I wouldn't be surprised at all if other similar code is out there in the wild.

the files i have all have the transforms in, and when just viewing the transformed points everything looks right, separate scans seem well registered. for this, if the origin were 'baked in' to the coords, i'd loose the ability to regenerate the angles. just like pts/asc files.
psiplace
I have made 10-20 posts
I have made 10-20 posts
Posts: 19
Joined: Thu Apr 09, 2009 11:44 pm
Full Name: simon place
Company Details: withheld
Company Position Title: withheld
Country: UK
Linkedin Profile: No
Has thanked: 6 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by psiplace »

> also, the SS2 isn't quoted as being a 1mm instrument. If your expectation that all the data should marry to 1mm, don't be surprised to see deviations of >5+mm. Its is quoted at ~6mm 3D accuracy for a single measurement. Will usually be better at shorter ranges, but just be mindful of this.


yes, but why would the angle have values that vary depending on range?

surely each vertical scan would have the same angle, presumably from an encoder, and so should be unaffected by the laser range value.

mechanically its just not possible, something has to be happening in the calculation/software.
psiplace
I have made 10-20 posts
I have made 10-20 posts
Posts: 19
Joined: Thu Apr 09, 2009 11:44 pm
Full Name: simon place
Company Details: withheld
Company Position Title: withheld
Country: UK
Linkedin Profile: No
Has thanked: 6 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by psiplace »

i downloaded this; https://iweb.dl.sourceforge.net/project ... a/pump.ptx
Screenshot_2020-03-19 WebGL Pointcloud Viewer(2).png

mostly aligned, (file has much wider horizontal, than vertical, angle steps) except very bad here...

Screenshot_2020-03-19 Screenshot_2020-03-19 WebGL Pointcloud Viewer(2) png (PNG Image, 1272 × 727 pixels) - Scaled (57%).png
seems to depend on horizontal spacing? (curiouser and curiouser)


this is the same spot in 3d and coloured.. shows this again is where there's a step change in range.

Screenshot_2020-03-19 Screenshot_2020-03-19 WebGL Pointcloud Viewer(3) png (PNG Image, 254 × 310 pixels).png
You do not have the required permissions to view the files attached to this post.
Tookie
V.I.P Member
V.I.P Member
Posts: 108
Joined: Thu Sep 21, 2017 9:43 pm
Full Name: Iain Took
Company Details: WMP
Company Position Title: Dogsbody
Country: United Kingdom
Linkedin Profile: No
Has thanked: 2 times
Been thanked: 6 times

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by Tookie »

So a wall behind a pipe run?

Could be ranging error made worse by close proximity to the pipe? The laser maybe landing on part of the pipe and the wall at the same time?
psiplace
I have made 10-20 posts
I have made 10-20 posts
Posts: 19
Joined: Thu Apr 09, 2009 11:44 pm
Full Name: simon place
Company Details: withheld
Company Position Title: withheld
Country: UK
Linkedin Profile: No
Has thanked: 6 times
Contact:

Re: finding systematic misalignment in points from PTX files, when angles/range calculated.

Post by psiplace »

> So a wall behind a pipe run?

well another pipe but yes.

> Could be ranging error made worse by close proximity to the pipe? The laser maybe landing on part of the pipe and the wall at the same time?

i did wonder if maybe some smart software might be registering a double hit and returning the strongest but offset,(by a proportion of the x-spacing) since it knows that is a 'better guess' location, but the offset is the same way, (and an amount that seems in proportion to the distance to the far object) on the other side of the pipe and its always horizontal.
Post Reply

Return to “Any Other Issues”