Blender Camera Pose about visual-gps-slam HOT 3 CLOSED

Thanks a lot man. It worked.
I put a start for your great work.

from visual-gps-slam.

Hello @YJonmo

Thank you for your question. Unfortunately I have no experience with TSDF yet (thanks for the link).

In my view you do not need to transform the Depth-Maps from Blender, as they are expressed relative to the blender camera coordinate system and not with respect to a specific world coordinate frame. However, depending on the specific TSDF implementation, you might need to triangulate or reproject the 3D points, for which you in turn need the correct camera pose. Therefore, your issue might indeed be linked to a wrong Blender camera pose.

You can check how I did the conversion in the python addon. The coordinate transformation from Blender to DSO can be found from line 34 to line 113 https://github.com/GSORF/Visual-GPS-SLAM/blob/master/02_Utilities/BlenderAddon/addon_vslam_groundtruth_Blender280.py#L34

You might also want to double check your calculation of the camera intrinsics with mine which you can find from lines 326 to 329: https://github.com/GSORF/Visual-GPS-SLAM/blob/master/02_Utilities/BlenderAddon/addon_vslam_groundtruth_Blender280.py#L326

Hope this helps. Kind regards,
Adam

from visual-gps-slam.

Thanks a lot man. It worked.
I put a start for your great work.

Awesome @YJonmo, that is great to read! Thanks for the update and great success for your project.

Kind regards,
Adam

from visual-gps-slam.