Authors:
Loic Merckel
and
Toyoaki Nishida
Affiliation:
Kyoto University, Japan
Keyword(s):
Change-point detection, Lie group, Rigid body motion segmentation, Special Euclidean group, Exponential map.
Related
Ontology
Subjects/Areas/Topics:
Advanced User Interfaces
;
Animation Algorithms and Techniques
;
Animation and Simulation
;
Animation from Motion Capture
;
Augmented, Mixed and Virtual Environments
;
Computer Vision, Visualization and Computer Graphics
;
Interactive Environments
Abstract:
Common CAD interfaces for editing spatial motion of virtual objects, which includes both position and orientation information, are often hampered by complexity and lack of intuitiveness. As the demand for motion data is increasing, e.g., in computer graphics or mixed reality, the development of new interfaces that offer a natural means of specifying arbitrary motion becomes essential. A solution consists in relying on live motion capture systems to record user’s gestures through space. In this context, we present a novel method for discovering change-points in a time series of elements in the set of rigid-body motion in space SE(3). The goal is to segment gesture-defined motion with in mind the development of a method for enhancing the user’s intent. Although numerous change-points detection techniques are available for dealing with scalar, or vector, time series, the generalization of these techniques to more complex structures may require overcoming difficult challenges. The group
SE(3) does not satisfy closure under linear combination. Consequently, most of the statistical properties, such as the mean, cannot be properly estimated in a straightforward manner. We present a method that takes advantage of the Lie group structure of SE(3) to adapt a difference of means method. Especially, we show that the change-points in SE(3) can be discovered in its Lie algebra se(3) that forms a vector space. The performance of our method is evaluated through both synthetic and real data.
(More)