Workshop: Joseph Henry and Xi Zhao

Joseph Henry

Motion simulation in constrained user environments
 
Environments which constrain human motion are common.  Everyday examples of these include entering, exiting and movement within a car or negotiating seating on a packed train or plane.  Motion within these "user environments" is restricted not only by the overall size of the environment but also by the objects within it.  A lot of work goes into designing these environments to promote comfort and ease of motion for the user.  Such properties are conferred both by the geometric parameters of the environment as well as the size, shape and physical capabilities of the end-user.  Comprehensive testing of these designs can be costly and time-consuming.  Generally it involves the building of life-size rigs of the environment along with user studies, where participants are selected to be as representative as possible of the wider population.  We propose that motion synthesis techniques based on an abstraction of the human-environment interaction, namely the interaction mesh, enable us to create novel environment-appropriate motion from example interaction sequences.  In this way we can produce motion for empirically testing the validity of a particular environmental design.  This talk outlines current work and results relevant to this goal, with a focus on its application to car design.  Future directions for this work will also be discussed.  These include how retargeting of synthesised motions to different environment and character combinations might be achieved through exploitation of the properties of the interaction.

 Xi Zhao

We propose a new approach to compute the distance between two 3D scenes based on spatial relationship of objects in the scene. Previous context-based classification methods for scenes rely on discrete contact information, which do not provide enough information for describing scenes where the objects are interacting with one another in a complex fashion. We use medial sheets constructed between different parts of objects to describe this kind of complex interaction.  Given such a representation, we propose  an efficient method to compute the similarity between two different scenes by using a multi-resolution decomposition of medial sheet data. Our method can be applied to scenes composed of the same objects as well as those composed of different objects.