Dance-Texture Analysis and Synthesis from Motion Capture Data

Open Access
Chiang, I-kao
Area of Honors:
Interdisciplinary in Engineering Science and Computer Science
Bachelor of Science
Document Type:
Thesis Supervisors:
  • Yanxi Liu, Thesis Supervisor
  • Bernhard R Tittmann, Honors Advisor
  • Jesse Louis Barlow, Honors Advisor
  • Judith A Todd, Faculty Reader
  • Texture
  • Dance
  • Synthesis
  • Symmetry
  • Motion Capture
  • Human Motion
A full pipeline from analysis of Motion Capture (MOCAP) data to synthesis of motion texture was introduced. First, a novel methodology was proposed to extract information from MOCAP data by transforming MOCAP data into motion texture. 173 Jamaican dance data were given to 25 raters for evaluation. Motion features were correlated with traits (actual gender, perceived gender, and dance-ability) of the individuals and perceptions of the raters. The result of the correlations between traits and scores of features suggested that females have distinct features around the hip and the hands, whereas males have distinct features around the shoulder and the knees; the result also suggested that good dancers dance more asymmetrically and more dynamically than bad dancers do. Features were computed using all possible combinations of body joints and set of five asymmetries of frieze patterns. Discriminations and reclassifications yielded statistically significant correlations, p-values smaller than 0.001 between scores of features and traits. The developed pipeline was able to synthesize motion in a controllable and efficient fashion that is unachievable by traditional motion synthesis techniques. Finally, the obtained correlation was fitted with linear regression models and was coupled with 2D image synthesis techniques. Furthermore, the presented technology may lead to a breakthrough in medical, social, and media domains.