The Secret Tracking Strategies
of the Interrupting Cows
Who we are:
The Interrupting Cows are a tracking system created to capture leg motion.
How we do it:
The tracking is done by finding the brightly marked joints on a character,
calculating the differences between the joints in proportion to the
height, and then calculating the joint angles. A curve is then placed
through the joints, with interpolation to help with smoothing.
What was attempted, but failed:
Between placing the markers and converting the images to the needed
format, information was lost (lossy jpeg compression). This led to
higher-than-expected difficulty
levels in finding the markers.
What worked:
We used an animated figure as our data. We chose this data due to vision
concerns. There is no unexpected noise in an animation, therefore making
the determination of markers easier.
We placed the markers by hand using Adobe Photoshop. The markers were
placed on the ankles, knees, and hip joints, as well as the head (to
determine height). We then wrote
a program to locate these markers and their coordinates, as well as
to calculate the lengths of the limbs and the height of the figure.
We use this information to calculate joint angles, and use the joint
angles to create a motion curve in joint angle space. It deals with
occlusions of markers by interpolating between the known coordinates.
We made several assumptions:
- Limb is maximally extended when joints are straight
- The limbs will be maximally extended at least one time in the data
- perspective only taken into consideration for human as a whole
Results:
We were able to generate curves in joint angle space for the left and
right hip and knee joints. We used 3D data projected into 2D and
attempted to reconstruct the 3D information. We also were able to
calculate limb lengths from this data.
The generated curves:
Movement of Left Hip
Movement of Right Hip
Movement of Left Knee
Movement of Right Knee
Movement of Left Hip and Knee
Movement of Right Hip and Knee
Movement of Both Hips
Movement of Both Knees
How this may be improved:
This could be improved so it can use real motion capture data.
More rigorous geometry calculations used to test for feasibility of joint
angles and limb lengths.
Recognition of behavior to better fill in gaps in data due to occlusion.
The Code:
tracking tar ball
The Creators:
The algorithms for the Tracking Interrupting Cows were developed by Gary
Yngve and Alison Smith, junior computer science majors at Georgia Tech.
Gary may be contacted at gary@cc.gatech.edu
Alison may be contacted at ans@cc.gatech.edu
Back to the Interrupting Cows' Homepage