From gait, dance to martial art, human movements provide rich, complex yet coherent spatiotemporal patterns reflecting characteristics of a group or an individual. We develop computational methods for motion perception from multimodal data. In particular, we advance our understanding of physics from visual input by constructive models to learn dynamics from kinematics. In this talk, I present a trilogy on understanding human movements:
(1) Gait analysis from video data: A group theoretical analysis of periodic patterns offers both effective viewing angle categorization and human identification from similar viewpoints.
(2) Dance analysis and synthesis (mocap, music, Mechanical Turks): we explore the complex relationship between human perception of dance quality/dancer's gender and dance movements. Using a novel multimedia dance-texture representation, our learning-based method is applied for dance segmentation, analysis and synthesis of new dancers.
(3) Taiji (Tai Chi) movements understanding from kinematics to dynamics (mocap, video, foot pressure): we investigate Taiji sequences (5 min) performed by subjects from beginners to masters to understand the quantified relation between pose and stability.