Wavelet/Geometry Processing ELEC484 Project Proposal

So, I just sent this off:

Hi Michael,

Here's what I've got so far for my term project. I'd like to write a program which constructs 3d models of objects placed in front of my Kinect RGBD(Depth) camera. 

1. read an RGBD (RGB and depth) point cloud from my Kinect (OpenNI or libfreenect)
2. convert that point cloud into a mesh object (PCL or VCG)
3. implement a few subdivision schemes (Butterfly or Kobbelt √3?) on the resulting mesh object (straight C++)
4. import the resulting subdivided mesh objects into a viewer program (Meshlab or Blender) 
5. Optional goals: 
1. implement a point cloud to mesh object converter in C++ using isosurface extraction or front meshing algorithms to replace PCL/VCG from #2
2. paint the final mesh object using a Blender python script and the RGB data from the initial point cloud in #1
3. implement an un-subdividing scheme to generate lower polygon count models
4. put those models in a dorky little game using Blender (I've used the Blender game engine before, a little, and there are many Free sample games available to work from)

I plan to spend the next couple of weeks learning as much as I can about all of the tools mentioned below, in hopes of using them in my elec 484 and elec 486 projects, and just for fun. 

My proposed schedule is:

1. Spend the next two weeks learning as much as I can about all of the tools mentioned below, in hopes of using them in my elec 484 and elec 486 projects, and just for fun. Have a working prototype of this project running which does not involve me implementing any course related algorithms (Blender already supports Linear and Catmull-Clark subdivision. If my implementations are even remotely efficient I'll be submitting them to the Blender project :) )

2. If I've succeeded, spend the next month implementing Butterfly and Kobbelt √3 subdivision, and after that as many secondary goals as time allows

3. If I've failed to get a mostly working prototype in the first two weeks, I'll abandon the point cloud to mesh object conversion until the subdivision schemes are working properly.

Terms:

• OpenNI Natural Interaction API (includes Kinect plugin & library) http://www.openni.org/
• freenect (Kinect plugin & library) http://openkinect.org/wiki/Main_Page
• KinectRgbDemo: free Kinect freenect.lib demo which saves Kinect data as point clouds or as mesh surfaces (sample mesh surface screenshot attached)  http://nicolas.burrus.name/index.php/Research/KinectRgbDemoV5
• PCL (Point Cloud Library) – includes acquisition, filtering, feature estimation, surface reconstruction, registration, model fitting, and segmentation http://www.ros.org/wiki/pcl
• VCGLib: supports isosurface extraction and advancing front meshing algorithms http://vcg.sourceforge.net/
• Blender (3D modelling / game engine / content creation / etc toolkit which I used a fair bit in high school) http://blender.org
• Blender import scripts (.ply supported) http://wiki.blender.org/index.php/Extensions:2.5/Py/Scripts

Attachment: (image screenscapped from youtube user MontyVaca's video)

p.s. I submitted my request for patient information form today! I'm gonna have a CD of xray images etc. that I have full legal ownership of for use in projects. I'm excited :)