CPSC 815

With Dr. Davis, Fall 04

Project 3: Tracking and Match Moving

 

The Goal of this project was to explore the possibilities that match moving and tracking make available to a digital film compositor.

Phase 1 : Two Sizes Too Small

Here is the Shake tree....

Using myself as the brunt of my own joke, I recreated a sight gag from the animated Christmas classic, The Grinch who Stole Christmas. The first step was to create footage of an empty wall, and footage of me standing in front of the wall, without moving the camera.this setup was designed to allow parts or all of me to become transparent, so that the "heart x-ray viewer" I recreated in photoshop could show that my heart was indeed two sizes too small. Now it doesn't look realistic to have my cartoon heart standing deadly still while my body moves about, even if I move ever so slightly. This is where the Stabilize Node in Shake came in. I used the node in Match mode to track my shirt button. I tracked it once foreward and once backward, and averaged the two tracks to smooth it out. I then attached the extracted data from this node to the other composited elements, with this Result:

Phase 2: The Chrinch who stole SIGGRAPH

Here is the Shake tree....

I thought it would be funny if the "Grinch" in this story snuck into the dpa lab and stole the nice juicy monitors which are supposed to arrive any day now. Since we have no such monitors as of yet (ahem) I thought I would composite one for the grinch to steal. I also thought it would be neat to do a "fly around" helicopter shot to make my life really difficult.This would require some two point tracking and some tricky maya animation. I gathered the motion tracking data front the film footage first. I used a jury rigged marker to track the scene with, and I used a stabilize node as before. The difference this time being that I used two tracking points instead of the original one. The reason for this is that the camera also moves in this shot making neccessary not only x and y motion tracking but scale tracking as well. Would that there were a way to track rotation from video as well, but alas, I had to rotoscope that part in Maya. Speaking of this, I created the monitor element and its matte in Maya. I imported the original footage into Maya and matched the motion of the monitor at the number of each frame where a "spike" or sudden change occured in the tracking data, with the hopes that maya would intrpolate faithfully between these major changes. while I did move the monitor with the original footage, I only keyframed its rotation, as I wanted the stabilize node to do the rest of the work. I then imported the video that maya rendered and composited it into the scene, attaching the motion data to it. Here is Result:

Problems Encountered:

Basically this: if you can't track everything, you can't get away with anything. It turns out that mayas interpolation was still too smooth, even though I was only rotoscoping the rotation. The result is a computer monitor that looks like it is weathering an earthquake. It doesn't look too bad if you scrub through more quickly. another thing to note is that a moving animated character might be a better candidate for this than an inanimate object such as the monitor i used, as a characters native motions may serve to disguise the asynchronous motion of the camera.

Final Project is next!