Tweet
Hello everyone, hello Roberto,
I'm working in a company that reconstruct the road accidents. We use Zephyr for reconstruction the roadways.
Earlier I contacted with Roberto for questions per email, but may be my questions will be usefull somebody else too, so I decided to continue to communicate through the forum
So, just to cover the entire width of the road and necessary lenght (from 20 up to 150 meters), we use big datasets for Zephyr. It can be from 50 till 600 photos for each project. 20 megapixels photos, fileformat - tiff. Camera - Canon EOS 6d, constant focal length 24 mm. Camera is placed on a holder with height of 3 meter and tilted down for 45 degrees.
We had never problems with photos orientation, even if photos were made in different directions of road.
The main issue, what we have almost in every project, is too less points in dence point cloud. The roadway sometimes is not closed (it has holes), and after mesh generation this holes are still on the road way. We tried to use the dencification filter, but it works not so fine, because we were getting worse quality of points texture and then the worse quality of mesh (we loosing the geometry of curbstone line and the quality of roadbed).
Sorry for comparing, but, for example, the same project of 50 photos we calculated in Zephyr and Agisoft Photoscan. The settings were almsot maximum (more than 90% resolution, very high level of discretization). In Agi we got a 42 million points, in Zephyr - only 10. And the meshes were really different by quality. In the project was the road with 35 meters length and 8 meters width.
The worst result we got by project with curve road with road length for 150 meters (we made 535 photos in both directions). We got only 16 million points and the quality of points positions was so low, that we could not see the markers on the road, which we made ourselves. Agisoft in this project could use only 335 photos but gave us 55 million points.
I would like to ask, how can we improve the number of points in project? Which settings should we use and what we do possible wrong?
What we expect as result: huge dense point cloud or mesh with highest resolution. In the end we use top view of project to make a sketch. On this top view we expect position error of objects (for example road marking line or gully) not more than 2.5-3 cm of real world.
Is it possible in our case of such big projects?
If it nesessary, I can send some examples of projects.
Thank you in advance.
Regards,
Vladyslav Buriakovskyi
I'm working in a company that reconstruct the road accidents. We use Zephyr for reconstruction the roadways.
Earlier I contacted with Roberto for questions per email, but may be my questions will be usefull somebody else too, so I decided to continue to communicate through the forum
So, just to cover the entire width of the road and necessary lenght (from 20 up to 150 meters), we use big datasets for Zephyr. It can be from 50 till 600 photos for each project. 20 megapixels photos, fileformat - tiff. Camera - Canon EOS 6d, constant focal length 24 mm. Camera is placed on a holder with height of 3 meter and tilted down for 45 degrees.
We had never problems with photos orientation, even if photos were made in different directions of road.
The main issue, what we have almost in every project, is too less points in dence point cloud. The roadway sometimes is not closed (it has holes), and after mesh generation this holes are still on the road way. We tried to use the dencification filter, but it works not so fine, because we were getting worse quality of points texture and then the worse quality of mesh (we loosing the geometry of curbstone line and the quality of roadbed).
Sorry for comparing, but, for example, the same project of 50 photos we calculated in Zephyr and Agisoft Photoscan. The settings were almsot maximum (more than 90% resolution, very high level of discretization). In Agi we got a 42 million points, in Zephyr - only 10. And the meshes were really different by quality. In the project was the road with 35 meters length and 8 meters width.
The worst result we got by project with curve road with road length for 150 meters (we made 535 photos in both directions). We got only 16 million points and the quality of points positions was so low, that we could not see the markers on the road, which we made ourselves. Agisoft in this project could use only 335 photos but gave us 55 million points.
I would like to ask, how can we improve the number of points in project? Which settings should we use and what we do possible wrong?
What we expect as result: huge dense point cloud or mesh with highest resolution. In the end we use top view of project to make a sketch. On this top view we expect position error of objects (for example road marking line or gully) not more than 2.5-3 cm of real world.
Is it possible in our case of such big projects?
If it nesessary, I can send some examples of projects.
Thank you in advance.
Regards,
Vladyslav Buriakovskyi
Comment