Making 3D models from video clips

VideoTrace is an application that lets you easily create 3D models from things in video clips.

The University of Adelaide's Australian Centre for Visual Technologies is developing VideTrace. I don't think it's available to the public yet.

Picture 1-136 The user interacts with VideoTrace by tracing the shape of the object to be modelled over one or more frames of the video. By interpreting the sketch drawn by the user in light of 3D information obtained from computer vision techniques, a small number of simple 2D interactions can be used to generate a realistic 3D model. Each of the sketching operations in VideoTrace provides an intuitive and powerful means of modelling shape from video, and executes quickly enough to be used interactively. Immediate feedback allows the user to model rapidly those parts of the scene which are of interest and to the level of detail required. The combination of automated and manual reconstruction allows VideoTrace to model parts of the scene not visible, and to succeed in cases where purely automated approaches would fail.
Link (Thanks, Conor!)


  1. If this pans out with any sort of accuracy, it could be a big boon for the field of compositing. Currently to crate a believable camera match with objects in a scene you need to do a lot of careful measurements to get the best effect. The ability to create 3D ‘roughs’ from existing objects would greatly simplify things. Especially basic operations like masking complex moving forms.

  2. I can see this as useful in Architecture for quick sketch models of surrounding buildings in a site plan to get an idea what the proposed structure will look like in context.

    Those models don’t have to be precise.

  3. Three words: open, source, games.

    Imagine walking around the block with a video camera in the morning and having it as a 3D world for your GTA clone by the evening.

    This technology is nothing short of revolutionary. It’s going to completely change the world of game development.

  4. @ #3
    I have to snicker softly at the implication that building/scenery models are somehow holding up Open Source games. That type of stuff is easy. GTA style buildings take a couple hours to whip up, and most of that is texturing.

    What is holding back Open Source games is that 90% of game assets are boring to make. They are not flashy, impressive and are only notable in their absence. Most of the art needs to be done closely adhering to someone else’s vision with a very low percentage of ‘high profile’ assignments.

    Add to that the fact that game assets have very short shelf lives. The entire process needs to be done in no more than a year or two. After that graphics expectations will have moved on.

    So, in short you need a fairly large group of dedicated artists (say a dozen or so) who are willing to donate lots of time doing what other people want them to do. Now, how many Open Source projects out there can you cant that have a dozen people happy to do the gruntwork? And I don’t mean leads who squash bugs in their off time.

    And all the scripting that goes into making those assets interesting is also 90% boring gruntwork
    with little recognition or immediate payoff.

    In short, the problem with Open Source games is that the majority of positions that need to be filled are precisely the hardest sort of people to find for Open Source projects.

  5. Now if you have multiple cameras – stereo view or a 3×3 grid – I wonder if could figure out most of the geometry for you and also the skin mapping.

    1. nice to see this magic sowftware..
      i m researcher in Geomatics Engg., Deptt of Civil IIT-ROORKEE (INDIA)
      it can be very useful for “Generation of Virtual 3-D City Models” ..and the model can be used for Geographical Information System..
      i want this software..
      can u provide this VideoTrace …?????

  6. This looks great — can’t wait for the release.

    Since I haven’t seen it mentioned above, you may want to check out Google Sketchup which does exactly the same thing, though it’s limited to still images rather than video. (But it does support multiple images, IIRC).

  7. Yeah, knew about this 3 months ago. It would be very useful for ArchViz but certainly not for film.

    Sketchup doesn’t do anything close to what this does. This is a little bit more than just camera matching a photo.

  8. Could someone apply this tool to the Zapruder film in order to determine precisely how far Connolly and Kennedy were displaced from the central axis of the car?

  9. …You know what’s going to jump on this like white on rice? The Star Wars & Star Trek mesh creators. They’ll be able to whip out somewhat more accurate meshes in a lot less time. That’s what *I* want this one for.

Comments are closed.