DualMotion: Global-to-Local Casual Motion Design for Character Animations

Yichen Peng, Chunqi Zhao, Haoran Xie, Tsukasa Fukusato, Kazunori Miyata, Takeo Igarashi

Introduction

Creating natural 3D character animation traditionally requires professional keyframing skills or expensive motion capture (Mocap) setups. For novices, navigating massive Mocap databases is often unintuitive, and sketching complex human movement—which involves over 20 joints moving simultaneously—frequently leads to "blank page syndrome." DualMotion addresses these challenges by introducing a hierarchical, sketch-based interface that decomposes the animation process into manageable stages.

Overview of the DualMotion global-to-local editing pipeline.

Methodology: The Global-to-Local Workflow

The core innovation of DualMotion is its two-stage design strategy, which reduces the cognitive load on the user by separating root movement from limb-specific details.

1. Global Stage

In the first stage, the user draws a 2D trajectory on a virtual ground. The system projects 3D motion data from the CMU Graphics Lab Motion Capture Database into the user's view. Using the Fréchet distance metric, it identifies the best matches for the "root" movement, such as walking or jumping along the sketched path.

2. Local Stage

Once the global motion is established, the user focuses on specific limbs (e.g., head or hands). By sketching relative trajectories against the character's center of mass, users can define nuances like punching or waving.

Detailed breakdown of the Global and Local stages using query strokes.

3. Shadow Guidance and Synthesis

To assist users with limited drawing ability, the system employs "Shadow Guidance" (inspired by ShadowDraw). It displays the top five search results as semi-transparent "shadows," allowing users to select a motion rather than generating one from scratch. The final animation is synthesized by blending the rotational data from the local search onto the skeletal structure of the global result, producing a standard BVH file.

Evaluation and Key Results

The system was evaluated using a subset of 55 motion clips from the CMU database, normalized to 100 frames. A user study with 14 participants compared DualMotion to a single-stage baseline.

  • Accuracy and Efficiency: DualMotion achieved significantly lower Mean Square Error (MSE) compared to reference motions ($p < 0.01$). Users completed designs in minutes with a significantly lower operation count than the baseline.
  • Usability: The system received a System Usability Scale (SUS) score of 75.6 ("good"). 86% of participants indicated they would use the tool frequently for rapid prototyping.

Discussion and Future Work

While DualMotion simplifies 3D animation, several areas for improvement remain. Scaling the database will require GPU acceleration to maintain real-time performance. Additionally, "stitching" limb motions can occasionally result in physically implausible poses. Future iterations may incorporate a "double inverted pendulum model" to ensure center-of-gravity balance and utilize the velocity of user strokes for more expressive timing control.

Summary updated

BibTeX
@article{peng2023dualmotion,
  author = {Yichen Peng and Chunqi Zhao and Haoran Xie and Tsukasa Fukusato and Kazunori Miyata and Takeo Igarashi},
  title = {{DualMotion: Global-to-Local Casual Motion Design for Character Animations}},
  journal = {IEICE Trans. Information and Systems},
  year = {2023},
  doi = {10.1587/transinf.2022iip0011},
}
Figure