Hi all,
my goal is to create a 3D model of a track or a road in order to integrate it into a race simulation. A file usable by blender or 3DSMAX is the final goal.
I have already tested photogrammetry with drone photos: About 180 photos to scan an area of about 500mx500m. The result has an accuracy of about 3cm, which is excellent.
The problem is that the drone used for this is expensive (around 1500 € - too expensive for us).
I saw this system: Project Video Scanner (https://projectvideoscanner.blogspot.co ... -page.html)
The principle is to reconstruct a point cloud and a 3D model from a shot with 2 or 3 video cameras (webcams) as with a 3D camera.
Do you know if it is possible to do the same with StereoPi? Has anyone ever done this kind of treatment?
The most basic processing would be to save the photos for reprocessing in third-party software.
The most useful processing would be to directly output the point cloud or the 3D model.
Thanks for reading me.
Olivier HOT
building 3D mesh
- Realizator
- Site Admin
- Posts: 912
- Joined: Tue Apr 16, 2019 9:23 am
- Contact:
Re: building 3D mesh
Hi Oliver,
AFAIK, high-resolution photogrammetry requires A LOT of computational power (i.e. a powerful desktop system).
Your basic approach (to save the photos for reprocessing in third-party software) can be done without any issues. You can either do a sequence of the high-resolution photos (like 2x8 or 2x12 MPix each) or record a high-quality video (FullHD) with a high bitrate (like 20-30 Mbit) to take images from.
As for the onboard processing - you can do this with lower resolution. In the upcoming article, we'll describe the real-time and post-processed Pointcloud calculation. But this approach is intended to obtain a single snap of the 3D world, and you need to combine all that 3D clouds to the one global map. A similar task is solved in ROS for the SLAM. It has octomap approach, where the world map is building using "Lego-style" cubes. But I don't think this method precision will be enough for your case.
AFAIK, high-resolution photogrammetry requires A LOT of computational power (i.e. a powerful desktop system).
Your basic approach (to save the photos for reprocessing in third-party software) can be done without any issues. You can either do a sequence of the high-resolution photos (like 2x8 or 2x12 MPix each) or record a high-quality video (FullHD) with a high bitrate (like 20-30 Mbit) to take images from.
As for the onboard processing - you can do this with lower resolution. In the upcoming article, we'll describe the real-time and post-processed Pointcloud calculation. But this approach is intended to obtain a single snap of the 3D world, and you need to combine all that 3D clouds to the one global map. A similar task is solved in ROS for the SLAM. It has octomap approach, where the world map is building using "Lego-style" cubes. But I don't think this method precision will be enough for your case.
Eugene a.k.a. Realizator