Cycle 1 Reflection: Looking Across, Moving Inside

Cycle 1 was a chance to pivot in terms of how I was approaching this project. At this point I had a working understanding of what the installation was going to do, but performing it in front of my colleagues revealed methodological challenges.

Specifically, the rotoscoping process I was using to remove dancers from the background of prerecorded videos was far too time-consuming. This was how the process worked:

  1. Aquire short films of dancers performing or rehearsing (gathered from friends or shot myself)
  2. Import videos into After Effect
  3. Use the AF rotoscoping tool to begin cutting out each dancer from the video on their own composition layer. The rotoscoping tool attempts to intelligently locate the edge of the object you are trying to isolate in each frame, but its success is based on the contrast of the object from the background, how fast the object is moving, and how many parts of the object move in different direction. A dancer in sweats against a white wall with many other objects around is obviously not a good candidate for predictive rotoscoping. Therefore, each frame must be cleaned up. Additionally, when dancers pass in front or behind each other, or make contact, it is virtually impossible to separate them.
  4. Once each frame is cleaned up, they are exported to a standalone, alpha-channeled video

This is very short test I did with a live and rotoscoped dancers that represents at least eight hours of rotoscoping. Looks good, but incredibly inefficient.

Switching to a green screen capture model should make this process more efficient, allowing me to spend more time on the actual interaction of the video and the live dancer and less on simply compiling the required media. Additionally, the resolution and depth of field of the Kinect was too low and shallow for the live participant to interact with the prerecorded dancer in a meaningful way. Alex suggested upgrade to the Kinect 2, which has increased resolution and depth of field.

Attached is the initial Isadora patch for placing a silhouette of a live person alongside a video:
https://www.dropbox.com/s/54qn8bfmoc2o5hg/Cycle-1.izz?dl=0



Leave a Reply