how to build 3d face reconstruction by using Realsense

[wshao4]

I have a SR300 now, but how to build 3d face reconstruction by using Realsense like the  functions here ?

[MartyG]

If you have Windows, the '2016 R2' version of the RealSense SDK (which is compatible with the SR300) comes with an easy to use free sample program called '3DScan' for creating model scans of objects or the head like the example for Kinect in your video. You can then export the model from 3DScan as an .obj format model file.

You can launch a 1.8 gb download of the 2016 R2 SDK in your browser by left-clicking on the link below.

http://registrationcenter-download.intel.com/akdlm/irc_nas/vcp/9078/intel_rs_sdk_offline_package_10.0.26.0396.exe

[wshao4]

@MartyG​  Is there a python interface? I see this demo using c#

[wshao4]

@MartyG​  Is there C++ interfaces? C++ interfaces will also work😅

[MartyG]

Outside of the 3DScan program, the options for doing 3D reconstruction with the 2016 SDK alone are limited. There are a number of commercial 3D modeling programs that support the SR300.

If you need to use Python, the best option may be to create a point cloud and then convert the point cloud into a solid mesh. A Python program called Pyntcloud has the ability to do such a conversion.

https://forums.intel.com/s/question/0D70P0000068ZfwSAE

[wshao4]

@MartyG​ Is there C++ interfaces? C++ interfaces will also work😅😅 😅

[MartyG]

It is quite straightforward to generate a point cloud in C++ in the current RealSense SDK 2.0, which is available for download from the link below.

https://github.com/IntelRealSense/librealsense/releases/

You can either generate the point cloud with C++ scripting or by using a pre-made program that comes with RealSense SDK 2.0 called the RealSense Viewer.

Once you have exported a point cloud to a file in .ply format though, it would need to be imported into a 3D modeling program to convert it into a solid mesh. A proven way to do this is to use a program called MeshLab.

https://www.andreasjakl.com/capturing-3d-point-cloud-intel-realsense-converting-mesh-meshlab/

You can also import a .ply point cloud file into other 3D modeling software packages such as Blender.

[wshao4]

I can't find the program that only contains model scans of the face roi from SDK 2.0. It may include body parts below the chin. How to eliminate the parts that are not of interest, save only the face area  like the '3DScan' program . @MartyG​ 

[MartyG]

MeshLab has a 'Select Vertices' tool for removing the parts of a point cloud scan that you do not want before you create the solid model.

https://en.wikiversity.org/wiki/MakerBot/Cleaning_Up_Point_Cloud_Meshes_in_Meshlab_For_3D_Printing#Deleting_Unwanted_Background_Points

https://www.youtube.com/watch?v=Q025ayjgD5I

[wshao4]

Hi, I want to do it through C++ code, not visual UI. @MartyG​ 

[MartyG · Accepted solution]

The only C++ I know of for setting auto-exposure ROI is the code snippet highlighted on line 459 of the script in the link below.

https://github.com/IntelRealSense/librealsense/blob/master/include/librealsense2/hpp/rs_sensor.hpp#L459

There is also a Python version of it:

https://github.com/IntelRealSense/librealsense/issues/2681#issuecomment-436391721

[wshao4]

All right.

I'm not very good at coding.

It could take a long time to study.

thank you!