HalfPager

Modeling Human Motion

(MOTION) Modeling Human Motion

Realistic and flexible models of human motion have a truly diverse range of applications. Games, animation, and scenario simulations need rich and interactive depictions of human behaviour. At the same time, they are also key to interpreting human movement as seen by cameras. As such, it is a critical component in developing the next generation of user-aware interfaces for games, assisted living, and other context-aware applications. Human interactions with the world are mediated by motion and contact. Models of human motion are also useful for understanding human biomechanics, human motor control, and extending the abilities of humanoid robots. The goal of this project is to develop and exploit new models of whole-body human motion with application to animation, games, e-commerce, interfaces for new media, modeling and tracking for health care applications, and entertainment robotics (Beaudoin et al, 2008; Wang et al, 2008).

Significant progress has been made in the past decade in developing models based on motion capture data, which can help inform the construction of both kinematic and dynamic models. However, scalability remains a core problem, as it is not feasible to capture the nearly infinite space of all possible human motions. As a result, more abstract models are needed which can generate motion according to the intent and context. Such flexible models have the promise to scale much better than models which simply resequence and blend captured motion data, as is common practice. Many existing models of human motion are focused heavily on flexible walking skills. However, they fail to model the full agility of human motion, and the rich repertoire of movements unrelated to locomotion, such as fine dexterous finger control during manipulation, realistic coordination of attention and gaze, modeling motion style, or interaction with complex passive systems such as clothing. They are not sufficiently rich to support general vision-based, marker-free motion capture.

Modeling human motion is an interdisciplinary endaevor. This project thus brings together investigators with significant expertise in computer animation, computer vision, games, interactive storytelling, physics-based simulation, robotics, machine learning, and perception. The project is structured around four research themes. (1) Authoring and Editing: the exploration and evaluation of new motion representations at various levels of abstraction. (2) Perceiving Human Motion: examing how motions are observed by humans and computers, including motion sonification, i.e., creating real time audio depictions of ongoing motions with numerous potential applications for athletics, and tailoring motion models for computer vision applications such as camera-based interfaces and markerless motion tracking. (3) Large Motion Repertoires: investigating models of climbing, grasping, and manipulation, all of which are involve complex force-based interactions with the environment, with applications to training simulations, robotics, ergonomics, and health care. (4) Motion Models for the Real World: developing techniques to control more biomechanically-faithful simulations and humanoid robots.

Philippe Beaudoin, Stelian Coros, Michiel van de Panne, and Pierre Poulin, Motion-Motif Graphs, Proc. Symposium on Computer Animation 2008, July 2008.

J. M. Wang, D. J. Fleet, A. Hertzmann, Gaussian Process Dynamical Models for Human Motion, IEEE Transactions on Pattern Analysis and Machine Intelligence, Feb 2008. Vol. 30, No. 2. pp. 283-298.

HQP: Graduate students from: UBC, SFU, U of Calgary, U Toronto, McGill, U Montreal. The project will produce highly trained graduates with expertise in human animation, vision-based tracking, physics-based simulation, and robotics. Through internships and as part of their training, they will be exposed to (and encouraged to pursue) a large variety of applications, including camera-based interfaces, next-generation games and simulations, biomechanical simulation, gait rehabilitation, and more.

N+P: The fields of computer animation, human-computer interfaces, computer vision, biomechanics, and humanoid robotics have each evolved their own models of human motion, developed for their specific purposes. As these models grow in sophistication, this project envisions their inevitable convergence. The expertise of the NIs on this project is tailored to pursue this convergence.

K+TE+E: There exist many possibilities to exploit good models of human motion in a commercial setting, ranging from next-generation animation tools; cameras that produce abstract human motion descriptions instead of raw pixels for computer interface, health-care, and assisted living applications; pedestrian-aware traffic management, etc. Many of these are new applications, and so the primary contribution to technology transfer will be through the HQP that has the abilty to envision the new applications and help develop them, in new companies of their own or in existing companies.

Other: