In addition to analyzing the dance through observation, it is also good to look at dance numerically so as to obtain a more computationally sound model. For this, video tracking was employed. I created a program that will follow a given dancer and track her movements around the stage. From this, numerical data can be extracted and analyzed.

Tracking a Dancer through Digital Image Processing

First the individual image is extracted from the video.

Then the image is converted to black and white.

Next the each pixel in the image is compared to a threshold. If it is below that value it is colored black, if it is above, it is colored white. By trying different threshold values, a number can be found that will extract the dark clothes of the dancers from the background.

The program then looks and identifies the black “blobs” that are left after the thresholding and identifies them as separate objects.

The largest blob (which in this setup should always be Elyssa since she is wearing a black top and a black bottom) is located and its center of mass is found. The position of the center of mass is then said to be the location of the dancer on the stage.

This process is repeated for every frame of the movie so that the dancer is tracked throughout the entire piece. It is necessary to put some bounds on the distance the “dancer blob” can move each frame so that the tracking does not jump to some other large dark object by accident.

Watch the finished video in [streaming QuickTime] or [avi] form

It also helps in the analysis to be able to see the motion of the dancer without any extraneous information visible.

Watch video with just the positional information [streaming QuickTime] or [avi] form

The video tracking program also outputted a list of the positions that Elyssa was in that were used to visualize the path that she took around the stage.

This information was then smoothed out by averaging and spline interpolation and enhanced to get a nice view of the actual path taken over the course of the dance.

From this positional information, it is possible to get information about the velocity of the dancer over both time and space. A list of velocities is first created. Then a graph can be made of the dancer’s speed for each frame of the movie.

From this information along with the positional information, the average velocity for each part of the space can be computed and a velocity matrix can be created. This is used to create a velocity field.

It is also possible to create a three dimensional representation of the velocity field.

click the picture to view a larger version of this graph

It’s also possible to average the velocities over larger regions to get a sense of what regions of the space were covered the most.

From all of this path and velocity information, a list of important statistics can be created