Can someone explain photogrammetry?

Write here about everything related to photogrammetry
Post Reply
User avatar
avogra
Site Admin
Posts: 95
Joined: 11 Nov 2016, 17:27
Location: Munich, Germany
Contact:

Can someone explain photogrammetry?

Post by avogra » 12 Jun 2017, 09:47

After replying to the question of 3dxcan, I realized that I have a very weak understanding of photogrammetry. Maybe someone can help me out?

My understanding of photogrammetry is, that basically you take several photographs of the surface to be reconstructed from different positions. The algorithm then tries to find matching features in those photographs. As each feature appears in several photographs, it's position in space can be reconstructed by triangulation. The position of the camera, which is needed for triangulation, is reconstructed from the photographs themselves, supported by target points in some systems.
This lets me assume, that perfectly uniform surfaces can't be reconstructed, because there are no particular features to be identified. Best objects would be heavily textured ones, like for example wood. At the same time, resolution would be limited by the granularity of the texture.

I think that photogrammetry is much better than that. So where is my mistake?

Curiousjeff
Posts: 116
Joined: 16 Nov 2016, 22:31

Re: Can someone explain photogrammetry?

Post by Curiousjeff » 14 Jun 2017, 14:57

I think you are correct.

Indeed, uniform surfaces are not ideal.

But when using RAW format, you can boost the contrast a lot, which can help the software to create a better model.

I want to also try:

Soft sandblasting of the model model
Speckle painting the model
Try polarizing filters on lamps and on camera

For the moment, on some model, I have obtained very good results and on others not so good.

As metionned in my other post, this can be a very fast solution for web/computer 3D display. I don't consider it highly precise like David.

It's just good to have various tools in the bag.

Jeff
P.S. You can try Photoscan Pro for free for 30 days. The standard version will provide you with exactly the same quality but without a built-in scale reference. I had very bad results using the same photos in Autodesk Remake.

3dxcan
Posts: 64
Joined: 02 Jun 2017, 17:21

Re: Can someone explain photogrammetry?

Post by 3dxcan » 17 Jun 2017, 18:06

avogra wrote:
12 Jun 2017, 09:47
After replying to the question of 3dxcan, I realized that I have a very weak understanding of photogrammetry. Maybe someone can help me out?

My understanding of photogrammetry is, that basically you take several photographs of the surface to be reconstructed from different positions. The algorithm then tries to find matching features in those photographs. As each feature appears in several photographs, it's position in space can be reconstructed by triangulation. The position of the camera, which is needed for triangulation, is reconstructed from the photographs themselves, supported by target points in some systems.
This lets me assume, that perfectly uniform surfaces can't be reconstructed, because there are no particular features to be identified. Best objects would be heavily textured ones, like for example wood. At the same time, resolution would be limited by the granularity of the texture.

I think that photogrammetry is much better than that. So where is my mistake?
My experience with photogrammetry comes from working with GOM Tritop System. it uses a set of coded targets (stickers with barcodes in round shapes) to match sequential pictures made around the object to construct its 3d model. then it can precisely measure coordinates of uncoded targets which sticks to the measuring object. think of these targets as CMM Points. so basically its an optical CMM, which only measures the coordinate of those sticker targets. and it does the job accurate to .01mm in very large settings.
one advantage of those targets is that you can then export those targets from the photogrammerty application and use them in the 3d structured light scanner application, that makes it possible to make an accurate scan of a very large object without using a very large calibration panel, because enough number of already measured target points will fit inside one shot of the scanner, and shots are progressively aligned to those targets.

User avatar
avogra
Site Admin
Posts: 95
Joined: 11 Nov 2016, 17:27
Location: Munich, Germany
Contact:

Re: Can someone explain photogrammetry?

Post by avogra » 23 Jun 2017, 04:53

That sounds very sophisticated. If your first contact with 3d scanning is those cheapo methods like kinect or simple laser scanners, david looks like high end, but apparently there is a world beyond.
Thank you for the insight!

User avatar
Micr0
Posts: 178
Joined: 15 Nov 2016, 15:20

Re: Can someone explain photogrammetry?

Post by Micr0 » 23 Jun 2017, 13:56

3dxcan wrote:
17 Jun 2017, 18:06
avogra wrote:
12 Jun 2017, 09:47
After replying to the question of 3dxcan, I realized that I have a very weak understanding of photogrammetry. Maybe someone can help me out?

My understanding of photogrammetry is, that basically you take several photographs of the surface to be reconstructed from different positions. The algorithm then tries to find matching features in those photographs. As each feature appears in several photographs, it's position in space can be reconstructed by triangulation. The position of the camera, which is needed for triangulation, is reconstructed from the photographs themselves, supported by target points in some systems.
This lets me assume, that perfectly uniform surfaces can't be reconstructed, because there are no particular features to be identified. Best objects would be heavily textured ones, like for example wood. At the same time, resolution would be limited by the granularity of the texture.

I think that photogrammetry is much better than that. So where is my mistake?
My experience with photogrammetry comes from working with GOM Tritop System. it uses a set of coded targets (stickers with barcodes in round shapes) to match sequential pictures made around the object to construct its 3d model. then it can precisely measure coordinates of uncoded targets which sticks to the measuring object. think of these targets as CMM Points. so basically its an optical CMM, which only measures the coordinate of those sticker targets. and it does the job accurate to .01mm in very large settings.
one advantage of those targets is that you can then export those targets from the photogrammerty application and use them in the 3d structured light scanner application, that makes it possible to make an accurate scan of a very large object without using a very large calibration panel, because enough number of already measured target points will fit inside one shot of the scanner, and shots are progressively aligned to those targets.
Can you point to some more info on this? I have struggled with how to make very large calibration panels accurately for David for a while now. Getting very large panels printed is easy. Getting them perfectly flat, square and true is extremely difficult. I've gotten almost good enough, but
the more I do this, the more I dislike David's calibration process and the more I want a better way.
µ

3dxcan
Posts: 64
Joined: 02 Jun 2017, 17:21

Re: Can someone explain photogrammetry?

Post by 3dxcan » 24 Jun 2017, 13:22

I have the manual for TRITOP, it helps to better understand photogrammetry procedure, which includes calibration of the photogrammetry camera. you should be interested in that too. it works simply by 2 bars, with target points at each end. I can send you the manual. send me your email address in private message.

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest