On
Windows app, File->Export 3D clouds..., there are a lot of options to export full resolution point cloud and/or filtering noise in it (note that exporting point cloud may be okay with latest release, but for meshing, see version referred by this
post). With the windows app, you can also detect more loop closures in post-processing step before exporting the clouds (this can add more constraints to correct the VIO drift). Other option is to export high density point cloud from the iPhone App and use third party point cloud softwares like CloudCompare and MeshLab (to refer free open source ones).
For "high quality point clouds", in term of point capture, we cannot really do more than the current precision of the LiDAR on the phone. To improve accuracy of the pose, then yeah the gimbal may help, also avoid looking directly at a white wall or textureless surface (step back a little to see more features around it and scan from farther away). Scanning to close loop closures also help to correct VIO drift over time. VIO from ARKit is pretty good, so no need to find loop closures every second, but at least closing a loop before changing from one room to another would help to get high quality for each room. In term of loop closures, if you can go back at the starting position with same orientation, this will help to find a final loop closure before finishing the scan,
cheers,
Mathieu
Locked
