特異スペクトル変換による動作変化点列を用いた動作解析法

Abstract

動作分析において，分節化は重要な役割を果たす．特異スペクトル変換により，筋力変化だけでなく，周辺環境など外力変化によっても発生した動作変化点を抽出できる．このため，動作分析において特異スペクトル変換は有効な手段であるが，抽出した動作変化点列を直接分析することは困難であった．本稿では，動作変化点列の非類似度を定義する．その非類似度に基づいて動作変化点列を低次元空間へ布置することにより，動作変化点列の関係性を可視化できる．さらに，低次元空間における投影点に対して感度解析を用いることにより，動作変化点列の変化を抽出する方法を提案する．歩行動作の分析に適用した結果により，提案手法の有効性を示す.

Segmentation plays an important role in behavior analysis. Singular Spectrum Transformation (SST) can extract change-points of human's motion caused not only by changes of muscle forces, but also by those of external forces, and it is effective to transform human's motion data into motion change-points sequences. However, it is difficult to analyze change-points sequences to extract motion pattern changes directly from extracted sequences. In this paper, dissimilarity between two arrays of change-points is defined and a method to visualize the similarity of multiple arrays of change-points is described. Moreover, a method to extract the differences of change-points between two classes of change-points sequences based on sensitivity analysis is proposed. The proposed method was applied to analyze walking motion of different walking speed or different identities. The results prove the effectiveness of the proposed method.