Final Report

           Over the course of summer 2005, I worked with Prof. Jessica Hodgins at Carnegie Mellon University as part of the CRA-W program for undergraduate women's research. During this time, I had the opportunity to work on several research projects with Professor Hodgins, and to learn more about the fields of computer graphics and human motion. I was able to get a taste of doing academic research in a graduate and professional environment, an experience which was immensely valuable in shaping my future academic plans. I would therefore like to thank CRA-W for this wonderful opportunity.

           My research for DMP Program in Summer 2005 mainly consists of 2 things :

1. To understand the perception of Human Motion in terms of dot patterns and how human would perceive these collections of human animation that we have generated earlier. In addition to that, we also want to find algorithms that would make the human animation more compelling and easier to perceive. So far, what have been covered from previous papers are that human dot patterns can identify genders, feelings, etc.

The main thing of this research is to find out when people would fail to identify the dot patterns. This means that we want to know how important is the law of physics in order to be able to identify the patterns correctly and non-intuitive patterns would lead to confusion.

2. How scaling of human motions matters (scaling law). Earlier papers discussed the idea that scaling of motion matters and that once the motion is scaled, it is said that it is not possible to have the same motions again.

          I worked on several projects with Dr Hodgins over the course of the summer. The first involved motion capture data that was obtained from CMU Motion Capture Database. Our thesis for the first project was "to understand the perception of human motion in terms of dot patterns and how human would perceive these collections of human animation that we have generated earlier." For the first few weeks, we used this data as our basic for our first research. What we did was to locate spheres in the essential joints of the skeletons from the data capture data so that we would be able to identify the movements. In addition of locating fixed spheres on the joints, we also scrambled the spheres to different locations with the trajectory remain preserved so that it would be even harder to perceive. We tried to add occlusion effect on the spheres in order to give depth perception of the spheres when one is behind the others. After a set of experiments have been made, we concluded that we do not actually need it since we used the three dimensional spheres, so the shading of the spheres would be sufficient to be used as a cue to determine the depth.

          The next research that we were doing involved scaling laws that says "bigger creatures have the tendency to move slower and the smaller ones tend to move faster". In order to prove this, we have set up a survey that consists of 9 different movies of 3 different sizes of blobs ( the model that we use for this experiment) and each size has 3 different velocity. Hence, for each size, there is one correct scaling and the other two are wrong scaling.

The calculation is as follows:

 . If we scale the blob by x (in m) and the time by t (in s), gravity (in m/s2) would be scaled by x/t2. Therefore, x / t² = 1 in order to conserve gravity.

 . Solving the equation in terms of t, we get » t = vx . The sizes of the blobs are: 0.5, 1, 2.

                   SAMPLE CALCULATION FOR THE TIME

The correct scaling example for size 0.5: t = v0.5 = 0.707 from the original time

When we plug in this time into the original equation, we get: Gfinal = 9.8 * (x/t2) = 9.8 * (0.5/ 0.7072) = 1. -> This is correct because the gravity remains 9.8

The incorrect scaling example for the size of 0.5 using the velocity the blob size 2.0:

 gfinal = 9.8 * 0.5 / (sqrt(2)2) = 39.2 -> This is incorrect because the gravity is not preserved at 9.8.

In order to know whether the thesis holds true that motion would not look right if the scales are not right, we conducted a survey.

The survey has the following settings:

. A still image of the three blobs to be shown at the beginning of each set for comparison purposes At least 36 -72 subjects needed for the experiment

. 9 different movies with random order of the sizes with the order of the velocities kept consistent

. Questions that are asked in the survey:

       - Is this glob smaller or bigger than the average person?

       - Does this glob move slower or faster than a person normally walks?"

       - How unnaturally or naturally does this glob move?

       - How humanlike does this glob seem?

       - How smart does this glob seem?

            Another project that is done to investigate the law of scaling is to use a human model. We will be using the back-flip motion from the motion capture database and the setup that is going to be used for the survey would be similar to the blobs survey.

                                           REFERENCES

- detect walking and running motions (Johansson, 1976)

- determine the gender of the runner (Cutting, Proffitt & Kozlowski)

- recognize individuals (Cutting & Kozlowski, 1977)