Abstract
[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]


From time-specified scripts to human-walking CG animations


As 3D-CG animations have been widely applied onto various fields like movies, even non-CG-literate people are getting interested in 3D-CG. These people's yielding demand for 3D-objects with motion is taken for granted. But when you use conservative 3D animation tools, you have to specify detailed movements. They generally require the knowledge and skill, and they are time-consuming.

On the contrary, some tools provide environments for automatical generation of 3D-CG from a script, which is easy to use. Working with these tools, the users only have to edit scripts. The scripts are sequence of the events for their animations.

Action of characters and camera-works usually proceed simultaneously. Time-length of cuts or whole length of animations are specified in many cases. To deal with these issues, the scripts should support the expression of parallel progress, synchronization of events and control on length of animations.

Accordingly, we introduce a form of scripts in which time-length of every event in animations is stated explicitly. With the scripts, we can handle the two issues. To synchronize events, you can set the same start-times. The sum of the time-length of all the events tells you the total time-length of your animation without adjustment.

To realize the script-base tool, we have to prepare a method to automatically generate 3D-CG animations that rigidly follow the time-length in scripts.

We choose human-walking event because it is a very common and is consider to be one of the most complicated events, and propose a method for generation of human-walking CG animations with time-specified scripts.

It is difficult to modelize computation of full walking-movements represented by joint angles with only the time-specified scripts. Therefore, we make use of footprint-expressions as intermediate expressions, i.e. footprint-expressions containing coordinates, times, and duration of every step are generated from scripts that read time-length and coordinates of the start and the goal, prior to the computation of joint angles.

We devide the process into three steps. The first one is the creation of footprint-expressions from scripts. The second one, the estimation of body-trajectories using footprint-expressions, and the last one, the determination of joint angles with footprint-expressions and body-trajectories. In the first step, time-length and coodinates for every step are determined with the formula that estimates step length or other parameters of walking. In the second step, footprint-expressions are optimized to minimize deviation from both the dynamical constraints that are based on one of the three dynamical formulas that is selected by the conditions of the step and the constraints on knee angles. In the third step, every joint angle of the lower half of the body is calculated inverse-kinematically, after the plane each leg is on is set according to some adequate assumptions.

We conducted an experiment to make sure that our mothod creates human-walking animations that rigidly follow the time-length that is specified by the scripts. Three different scripts are given to our system, and we get walking animations as outputs of the system for the scripts. These animations shows us they walk in accordance with the given time-length.


Go back to Thesis Page