I tried to adjust various map parameters but I get a lot of noise:
SBM:
SGBM:
Current Settings:
Could this be related to the 300mm base between cameras or camera calibration?
Disparity map noise
- Realizator
- Site Admin
- Posts: 900
- Joined: Tue Apr 16, 2019 9:23 am
- Contact:
Re: Disparity map noise
Hey Sencis,
1. According to your first image, youer left and right images are swapped, or you are using the inverses colorizing pattern. With the defaul settings red color means nearby objects, and the dark blue means far objects. Please check this.
2. You can find that the building borders, as well as the tree borders on the right, are sharp. To get the detailed depth map of the branches with the snow, you should decrease the blocksize and reduce the pre-filter cap value. To sau briefly, it it better to do smoothing after getting the map, but not brfore. Pre-filter is equivalent go gauss noice.
3. 300mm base is good for the objects which are far from the camera, and is bad for the objects nearby (like a few meters). If yo'll set the disparity map parameters to see both far and near objects with 300mm base (i.e. mindisparity and disparity range), you'll get a noicy image (and a lot of overcalculations).
1. According to your first image, youer left and right images are swapped, or you are using the inverses colorizing pattern. With the defaul settings red color means nearby objects, and the dark blue means far objects. Please check this.
2. You can find that the building borders, as well as the tree borders on the right, are sharp. To get the detailed depth map of the branches with the snow, you should decrease the blocksize and reduce the pre-filter cap value. To sau briefly, it it better to do smoothing after getting the map, but not brfore. Pre-filter is equivalent go gauss noice.
3. 300mm base is good for the objects which are far from the camera, and is bad for the objects nearby (like a few meters). If yo'll set the disparity map parameters to see both far and near objects with 300mm base (i.e. mindisparity and disparity range), you'll get a noicy image (and a lot of overcalculations).
Eugene a.k.a. Realizator