Modelling Two-person and Human-object Close Interactions
Introduction
Close interactions, not necessarily with any contacts, between different body parts of single or multiple characters or with the environment are common in computer animation and 3D computer games. Yoga, wrestling, dancing and moving through a constrained environment are some examples. In such motions, the spatial relationships between different body parts of characters are important in capturing the semantics of the scene. When an animator synthesizes or edits such movements, special care is needed to preserve these spatial relationships, for example, "arching back to avoid a punch", "hands extending around each other", "two bodies moving synchronously in close proximity" or "getting into a small car by bending down". However, traditionally, such spatial relationships exist only in the animator’s mind and are not digitally embedded into the data. Although humans use spatial relationships to recognize semantics of interactions, their usage has not been considered much in character animation.
We introduce a simple structure called an interaction mesh [Ho et al. SIGGRAPH2010, Ho et al. VRST2014] to represent such spatial relationships for editing and retargeting motions that involve close interactions between body parts of single or multiple articulated characters. Such an approach can also be used for controlling humanoid robots [Ho and Shum ICRA2013], synthesing virtual parner in VR [Ho et al. ACM TOMM2013]. The spatial relations between characters can also be used for facilitating the sampling process to generate a huge amount of pose-pairs [Yin et al. IEEE TVCG2019].
The spatial relations can also be used for analyzing interactions which can be used for motion retrieval [Shen et al. IEEE TVCG2020] and action recognition [Men et al. ICPR2020, Ho et al. VRST2015].
Our recent work [Goel et al. CGF2022, Men et al. C&G2022] further model human-human close interactions using Generative Adversarial Networks (GANs) to create a wide variety of interactions.
Readers are also referred to a closely related project on topology-based motion analysis and synthesis.
We introduce a simple structure called an interaction mesh [Ho et al. SIGGRAPH2010, Ho et al. VRST2014] to represent such spatial relationships for editing and retargeting motions that involve close interactions between body parts of single or multiple articulated characters. Such an approach can also be used for controlling humanoid robots [Ho and Shum ICRA2013], synthesing virtual parner in VR [Ho et al. ACM TOMM2013]. The spatial relations between characters can also be used for facilitating the sampling process to generate a huge amount of pose-pairs [Yin et al. IEEE TVCG2019].
The spatial relations can also be used for analyzing interactions which can be used for motion retrieval [Shen et al. IEEE TVCG2020] and action recognition [Men et al. ICPR2020, Ho et al. VRST2015].
Our recent work [Goel et al. CGF2022, Men et al. C&G2022] further model human-human close interactions using Generative Adversarial Networks (GANs) to create a wide variety of interactions.
Readers are also referred to a closely related project on topology-based motion analysis and synthesis.
Publications
The Team