Items with no label
Announcements
The Intel sign-in experience has changed to support enhanced security controls. If you sign in, click here for more information.
3331 Discussions

I am looking for a method for 3D reconstruction of faces using RGB-D images.

tguan
New Contributor I
1,797 Views

I am using the SR300 camera, I plan the first step of 3D geometry reconstruction, then texture mapping. Is there a code sample for python?

1 Solution
MartyG
Honored Contributor III
917 Views

For the SR300, the best option I can think of is software called CloudCompare, which has been demonstrated to be usable with RealSense for capturing a face as a point cloud.

 

http://www.danielgm.net/cc/

 

1.png

 

 2.png

 

The images above were captured with a D415 model, though I don't believe that CloudCompare is tied to a specific camera model, and there was a case where an SR300 was successfully used with CloudCompare

 

https://www.reddit.com/r/computervision/comments/7ifc71/3d_point_cloud_labelling_tool/

View solution in original post

9 Replies
MartyG
Honored Contributor III
917 Views

Pyntcloud would likely fit your needs. It allows you to create models using a type of depth scan called a point cloud, and then apply further processing such as conversion to a solid mesh-based model. It has been recommended in the past by Dorodnic the RealSense SDK Manager as a point cloud solution for Python.

 

Pyntcloud

https://github.com/daavoo/pyntcloud

 

Documentation

https://pyntcloud.readthedocs.io/en/latest/

 

Example tutorials

https://github.com/daavoo/pyntcloud/tree/master/examples

 

Converting a point cloud to a mesh with Pyntcloud (advice by Pyntcloud's creator)

https://stackoverflow.com/questions/44997054/point-cloud-to-volume

 

tguan
New Contributor I
917 Views

I used pyntcloud before, but I can only run it on a web page. So I tried to use py-pcl to splicing the face cloud. Do you have a good plan or idea to let me learn from it?

MartyG
Honored Contributor III
917 Views

I am not familiar with py-pcl. Another approach you could consider would be a technique called Poisson Surface Reconstruction. Below is an example for Python.

 

https://github.com/mmolero/pypoisson

tguan
New Contributor I
917 Views

thank you very much .I learned about pypoisson.It can meet my needs. Now I have a problem with the installation. I am using python3, the installation steps prompted in the webpage I failed to install using pip.​

MartyG
Honored Contributor III
917 Views

Have you tried the pip installation instruction below, please? Ignore the clickable link, and copy-and-paste the whole line.

 

pip install git+https://github.com/mmolero/pypoisson.git

tguan
New Contributor I
917 Views

Yes, I tried it and showed that the wheel installation failed. Is there a way to download the pypoisson wheel file directly?

MartyG
Honored Contributor III
917 Views

I'm not familiar with the concept of Python wheel files (I'm usually a Windows guy). You can research the subject though by googling for 'python wheel installation'. Here's an example:

 

https://stackoverflow.com/questions/21222114/how-do-i-install-python-libraries-in-wheel-format

 

There are also alternative installation instructions on the front page of pypoisson:

 

$ git clone --recursive git://github.com/mmolero/pypoisson.git

$ python setup.py build

$ python setup.py install

tguan
New Contributor I
917 Views

Thank you very much for inspiring me. Now I am considering the use of ICP algorithm to achieve the fusion of face point cloud data. I want to know what methods I can recommend using the SR300 camera. Here I will collect facial point cloud data from several different angles and then perform point cloud registration.Really thank you for your help.

MartyG
Honored Contributor III
918 Views

For the SR300, the best option I can think of is software called CloudCompare, which has been demonstrated to be usable with RealSense for capturing a face as a point cloud.

 

http://www.danielgm.net/cc/

 

1.png

 

 2.png

 

The images above were captured with a D415 model, though I don't believe that CloudCompare is tied to a specific camera model, and there was a case where an SR300 was successfully used with CloudCompare

 

https://www.reddit.com/r/computervision/comments/7ifc71/3d_point_cloud_labelling_tool/

Reply