My Research

|| Home || My Research || About Me || My Mentor || Journal || Final Report ||

I am working with Professor Jessica Hodgins on a couple of projects concerning the perception of animations of human motion, with the goal of uncovering guidelines for where we should be spending our time to produce more compelling animations at minimum cost. Read on for project descriptions. If you find that this interests you, you may also be interested in my presentation and final report.

  • Our first project involves the recovering of connected forms from dot pattern motions. It is well known that the human form can be easily recovered from dot-pattern images of a running person, such as that shown in the image below.

    Early studies have shown that subjects can detect walking and running motions, identify the gender of the runner and even recognise the gait of friends simply by observing dot-pattern renditions of these motions. What we are interested in is when this ability breaks down. What determines whether a given dot-pattern is recognisable as human? We believe that it has to do with the physics of the motion depicted - non-intuitive motions which appear to violate the laws of physics cause confusion and are less likely to result in successful detection of the human form.

    We are preparing dot-motion movies using footage from the Carnegie Mellon Motion Capture Lab some of which is pretty fresh and still needs to be cleaned. For example, we are using some footage of people doing Contact Improv, which is a form of contemporary body-contact dance that involves the use of gravity and the body's mechanics. Partners heavily support each other's body weight at all times. These motions provide for some exciting illusions when we separate the two partners. One of our sample movies shows the woman being rolled over the man's shoulders, except we took the man out, so it looks like the woman is flying and spinning through the air. We hypothesise that it is pretty hard to tell exactly what is being depicted when the motion is unfamiliar and unrealistic as in this case. We are also looking at the identification of two-person dot-motions such as that shown in the image below, which depicts two actors forming a wheel-barrow.

    We are also trying to study the effects of occlusion on subjects' abilities to identify the motions as human. Previous studies used movies of actual people with digitised point-lights, which showed that people are more likely to correctly identify a motion as human when all the correct occlusions are presented. We are trying to generate similar movies using Maya, so as to further explore this effect.

    We are working with collaborators at the University of Virginia on this project.

  • Another project we're working on involves scaling laws in animating motion. It is believed that motion, if scaled, "breaks", i.e. ceases to be believable. We want to show that motion, if scaled correctly, can be preserved.

    Intuitively, the size of a creature affects how we expect it to move. A smaller creature is expected to scamper and a larger one to lumber. Our experiments make use of small, normal and large sized blobs, identical to the one shown below.

    In the "natural" condition, the smallest blob is the fastest moving and the largest one the slowest. We have also prepared movies in which the motions are interchanged, i.e. the large blob moves fast, the small blob moves slowly and so on. We predict that, if we compare the normal and incorrectly scaled motions by devising a questionnaire to gauge subjects' response to both motions, we will find that the incorrectly scaled motions break down but the correctly scaled ones don't. We also plan to extend the experiment to scaled human motion.

Marcella Tanzil is my partner in crime in these matters. Harriet provides the comic relief, with her belly-flopping texture-synthesis man.

|| Home || My Research || About Me || My Mentor || Journal || Final Paper ||