3d reconstruction using realsense

[TheNemo05]

i need to create 3D reconstruction using the Depth data captured by D455, i guess it will require co-ordinate data, pixel information etc. Do realsense have a standard procedure to create such 3D reconstruction using realsense.

[MartyG-RealSense]

Hi @TheNemo05 If you are able to make use of commercial software then there are D455-compatible products that can perform the capture. These include RecFusion (Windows) and DotProduct Dot3D (Windows or Android).

https://www.recfusion.net/
https://www.dotproduct3d.com/

If you are able to use C++ then the RealSense SDK's rs-kinfu C++ open-source example program may be helpful. It enables a single camera to be moved around a scene, progressively building up a point cloud scan with frame fusion. Once you are satisfied with the level of detail on the scan then you can export the pointcloud to a .ply format pointcloud data file.

https://github.com/IntelRealSense/librealsense/tree/master/wrappers/opencv/kinfu

If you are using Python then I would recommend performing the reconstruction with Open3D and its RealSense support.

http://www.open3d.org/docs/release/tutorial/sensor/realsense.html

[TheNemo05]

do you have any examples for how to use 3D reconstruction with Open 3d realsense

[MartyG-RealSense]

There are not many references about reconstruction with RealSense in Open3D other than the tutorial that I linked to above, and also isl-org/Open3D#2467 and isl-org/Open3D#5082 and the link below.

https://blog.georgmill.de/2020/09/01/3d-reconstruction-with-open3d-on-a-jetson-nano/

In addition, Open3D's official documentation has a non-RealSense section about its reconstruction system.

http://www.open3d.org/docs/release/tutorial/reconstruction_system/index.html

[TheNemo05]

I am still not able to find sufficient document of open3D with realsense for depth 3D reconstruction.

[MartyG-RealSense]

You could also look at isl-org/Open3D#473

[TheNemo05]

according to this comment isl-org/Open3D#473 (comment) will the camera capture the entire 3D plot wherever the camera is being placed ? Or will it capture the entire scene

[MartyG-RealSense]

The camera will capture whatever is within its field of view (imagine looking through the viewfinder of a photographic camera). Like a photographic camera, a RealSense camera has a field of view size. Detail in the observed scene that is outside of the field of view box will not be captured unless the camera is turned to bring that outer detail into the field of view.

[TheNemo05]

ok this is not working, do we have any procedure to export pointclouds files from .bag file. So later i can stitch multiple pointcloud files into one for my surfacearea measurement

[MartyG-RealSense]

The RealSense SDK's rs-convert tool can convert bag files to pointcloud .ply files.

https://github.com/IntelRealSense/librealsense/tree/master/tools/convert

If you are using Windows and have installed the full RealSense SDK, instructions for launching the rs-convert tool can be found here:

https://support.intelrealsense.com/hc/en-us/community/posts/360051132573/comments/360014080493

If you are using Linux and have installed librealsense with the examples and tools included, rs-convert should be in the /usr/local/bin location and launchable with the command rs-convert