Multidimensional Trajectory Similarity Estimation via Spatial-Temporal Keyframe Selection and Signal Correlation Analysis

Abstract

In this paper, we present a framework for trajectory matching in asynchronous time sequences. The proposed methodology is applied on sequences of traditional Greek dances to quantify the choreographic similarity between them, using body joints' position and orientation, as well as the temporal dimension as inputs. The adopted methodology uses a two-step approach over the provided body joints' trajectories: a) representative frame selection and b) keyframe comparison. The dance act is captured using a single, markerless, low-cost sensor. The recorded sequence is summarized by selecting the most descriptive frames, which are then compared to other dances' keyframes to calculate similarity scores.