-
Notifications
You must be signed in to change notification settings - Fork 129
Open
Description
Hello, I am trying to perform 3D reconstruction using RGB images, depth images, camera intrinsics and extrinsics. I want to perform the 3D reconstruction using limited number of frames (~ 25) for my use case and started experimenting by extracting the pointcloud (verticels, color) for one frame only.
I observed that the TSDF Layer gives pretty good results for voxel positions even with one frame, but the Color layer is grey at some voxels (particularly around the boundary of the objects) as shown in the picture.
My understanding so far:
TSDFandColorlayer are treated independently in the backend and hence accumulate different weights during the integration process.- When using only one frame,
TSDF Layerhas enough weights in the voxels during the mesh integration (defaulted to grey color (127, 127, 127)) but theColor Layerdoes not have enough weights in the voxels at the highlighted regions to assign rgb value.
Questions:
- Are there any parameters I can tune to handle different minimum weights for different layers during integration, so that rgb values are assigned to all voxels visible from rgb image?
- Does
ViewCalculatorParamsplay a role in the grey voxels by limiting the camera FOV for integration?
Please let me know any leads to handle this issue.
deivse
Metadata
Metadata
Assignees
Labels
No labels