An Efficient Approach for Human Motion Data Mining Based on Curves Matching

Abstract

In this paper, we present a novel and efficient approach to retrieve human motion capture data as used in data-driven computer games, animated movies and special effects in the aim of finding a specific motion. From the kinematic chain model, the human motion capture data is transformed to a spatial-temporal invariance representation called the motion feature representation, in which each segment of kinematic chain model is represented by an angle between itself and the root segment. We treat the human motion as a cluster of curves of angle. In the aim of finding a human motion capture data in a very large database, we propose a novel lower bounding distance called LB-Keogh_Lowe to speed up similarity search. In order to reduce the computational cost, we employ techniques to simplify the curves length of both the envelopes curves and the query data. The similarity between two human motions is measured by applying the constrained Dynamic Time Warping. We carry out an experimental analysis with various real motion capture dataset. The results demonstrate the efficiency of our approach in the context of the human motion capture data and the potentiality to apply it in others contexts of the time-series data retrieval.