Hareket Veri Tabanlarında Fizik Tabanlı Benzerlik Arama

Abstract

Human motion capture data is a digital representation of the complex spatio temporal structure of three dimensional human motion. With the rapid development of motion capturing technologies and the more common use of motion capture data in the field of computer graphics and animation, methods for the reuse of the recorded motions in a database are gaining in importance both for efficiency and cost reasons. As a result of this, the identification and extraction of similar motions within some data set are of central importance for data driven approaches. This work aims to propose a physics based similarity model for retrieval of motion capture data.

Motion expression is an important basis of constructing motion databases for purposes of efficient and effective motion capture data organization, classification, analysis and retrieval. For this, it is essential to create an abstraction over motion capture data in the form of human motion features. In this approach, we describe physics based human motion features consist of joint torques. Physics based features implicitly include the gravity, ground reaction forces and some knowledge of the remaining body parts. These features are discriminative and low dimensional representation of a human action which preserves information of the original high dimensional data. We utilize these low dimensional physics based features extracted from motion data with kd tree data structure to search for similar motion poses. Using spatial and temporal alignment methods of motion data, we expand searching for similar motion poses to searching similarities in motion capture data.