Advertisement

You're blocking ads, which pay for BlenderNation. Read about other ways to support us.

Turn your webcam into a 3D scanner

49

web_cam_3d_scannerSeveral people reported this: Cambridge University Qi Pan is developing software that will reconstruct textured 3D models from webcam video. A Linux and later a Windows downloadable demo should be available in a couple of months. The results are already quite amazing - check out the video inside.

Links

About the Author

Avatar image for Bart Veldhuizen
Bart Veldhuizen

I have a LONG history with Blender - I wrote some of the earliest Blender tutorials, worked for Not a Number and helped run the crowdfunding campaign that open sourced Blender (the first one on the internet!). I founded BlenderNation in 2006 and have been editing it every single day since then ;-) I also run the Blender Artists forum and I'm Head of Community at Sketchfab.

49 Comments

  1. THIS IS AMAZING!!!! ,just imagine how easy it will be to digitalize your face/head or the entire body...i think it will be better that DAVID 3d scanner,wich require to have a laser .
    hope it will be free as well ;)

  2. VERY Impressive! The applications for this are endless, especially for markerless augmented reality systems! I take my hat off to the person who made it.

  3. amazing !!

    the fun is that google ads are for 'affordable 3d scanners' on this page :)

    can't wait to see the followings of this story

  4. Woow. I did ask in a Blender forum if it was possible to convert a point cloud to a 3D mesh, and here is the answer. With better calculations algorithms it will soon be possible to film a real house (or landscape) and get a rough 3D mesh of it.

    Cool with the reel time functionality, this make the 3D future interesting.

  5. actually look at papers from Polyfeys from the 90's. Similar algorithms are around for years, just that nobody took the effort to make them realtime & opensource.
    great news.

  6. Thats just amazing, i actually have been working with blender for a short time, not very good at modelindo, but im great with textures and riggings/animation. This would solve my modling problems, and i could just make real life models and then get them into the pc with this software.

    Hopefully it would be free... because if it is comercial, im sure it would cost a lot of money.

  7. How does this affect the jobs of modelers in the near future?

    Of course is amazing, it just worries me how automatized is everything. The mesh is not perfect but a Re-mesh algorithm can solve that.

    I hope is not something to worry about.

  8. Segmentation Blendiac. It takes the point cloud and tries to find points that create a face. The face has to be plane. Then it tries to find faces that have the same edge and with these edges it is possible to add the next face. I assume the algorithm uses some coplanarity issues.
    We develop at our company the same stuff. We take LIDAR data from airial planes flying over cities and trying to reconstruct complete cities.
    the problem is that houses are not as simple as shown in this demo. Many cities have very complex roof shapes and of course you have gables, chimneys, antennas and all that stuff.
    Therefore we only reconstruct LOD2 models (Level of Detail). These are generalized buildings.
    As pildanovak already mentioned - the algorithms are old.
    We work together with the University of Vienna
    http://www.ipf.tuwien.ac.at/cd-labor/31.0.html

  9. I'm wondering how complicated the mesh can get using this method (and how it didn't pick out his hand, I suspect that there's some form of straight edge tolerance on this/mixed into the motion tracking)

    Where abouts were the webcam and laser scripts I saw a while back? I've gotta google that again since I've got a single beam laser (for aligning my astronomy telescopes) but need to make some form of relflective mirror on a small motor to throw the laser up and down onto the turntable and the subject matter.

  10. While it is pretty neat, it seems it still has a way to go developement wise.
    The model used is very simple, yet the mesh seems quite messy, and a
    bit inaccurate when it comes to concave surfaces.
    I'm sure this is something that will see some improvement in the future,
    but concerns me that they are showing it off either this early in
    development, or that it's still this poor at handling it at this late in
    development.

  11. now imagine having a computer controlled 3-axis turntable in a dark room with laser to light the model wherever the computer detects inconsistencies and to further map the object with greater precision ... also multiple cameras oriented towards the model from multiple angles ,
    and all available for order-in a small kit
    that would result in a market for this product and with a market there's sure to be some competition - with this competition the product will become better and smaller (more implementable in any location )
    with a broader market we could all soon have this at our disposal ,in our arsenal of open source tools ,
    new comers to the 3d industry would be able to create stunning work with little effort
    and the world will become a better open source world , with less arguments that lead to global disputes
    resulting in a more peaceful world - creating an environment where human beings can grow to accept each others differences

    for brighter future,
    keep up the great work !!!

  12. Quite impressive in that small scale.
    I'm not too much in the textures, but a robust reconstruction of a bigger model (a milling machine for example)
    in a not-so-well-lighted environment (a factory building) with changing lighting. Where you can't move the model, but have to move the camera...... and you get out a (relative) good mesh, that would kick all 3D-Laserscanners out off business. Be careful, they will offer you lot's of money ;)

  13. WOW! This looks really interesting!

    But I also wonder, that it doesn't track the hand even in initialisation stage, when point cloud is created by tracking moving portions of the image. No single point placed on the hand. Ok, the magic is probably the "...motion estimation followed by bundle adjustment" part, which I dont fully understand. :)

    I think it would be very funny to take the camera and move it through the room. I'd love to see the result of this! :-D :-D

    Can't wait to check this software out!

  14. Sorry guys, but u could use this only for scanning boxes and cans!!
    This because feature tracking und probabilistic algorithms are used!!
    For all the other objects take the David Scanner or more professional (expensive!!) systems!

  15. Wonder if in the near future video files can be turned into 3d environments by using tools based on this software. This would enable you to get more accurate models of buildings and people long gone or even give you the opertunity to have a look around inside movies.

  16. Tim, LA: its pretty obvious that at this point the software is limited to very basic shapes, but all things can be created from basic shapes/faces. Wait for USB 3.0 to bring some serious resolution into the mix and then the fun can begin

  17. That's pretty cool. I came across a program a number of years ago that could do this but you had to take seperate images and manually load them into it. When I saw it I thought, 'wouldn't it be nice to have a program that could do this from a video instead', and here it is... :)

    I hope they decide to open source it! That would be really nice.

  18. This is a very nice progress on 3D model recognition.
    it reminds me of the presentation by Thomas Vetter and Volker Blanz in Siggraph 1999:
    A Morphable Model for the Synthesis of 3D Faces
    Can't we work towards a similar 3D recognition from a single 2D image combined with a reference model database? It's probably an extremely enormous task !!!
    Some links:
    Video on ChaosComputerClub (PLEASE LOOK AT THIS VIDEO):
    chaosradio.ccc.de/ctv088.html
    Some Volker Blanz pages
    http://www.mpi-inf.mpg.de/~blanz/
    http://www.informatik.uni-trier.de/~ley/db/indices/a-tree/b/Blanz:Volker.html
    Some Thomas Vetter pages:
    http://www.informatik.uni-trier.de/~ley/db/indices/a-tree/v/Vetter:Thomas.html
    en.scientificcommons.org/thomas_vetter

    Best regards, Anders Otte

Leave A Reply

To add a profile picture to your message, register your email address with Gravatar.com. To protect your email address, create an account on BlenderNation and log in when posting a message.

Advertisement

×