[4K3-GS-3] Data mining: User aid

[4K3-GS-3-05] Motif and Grammatical Inference-based Repetitive Activity Segmentation on Time Series

〇Masahiro Terada¹, Mao Inoue¹, Takeru Shimada¹, Naoki Minami¹, Makoto Imamura¹ (1. Tokai University Takanawa Campus)

Keywords:Time series, Segmentation, Human motion analysis, Sensor data mining, Grammatical inference

In order to analyze human motion data automatically, segmentation processing that extract basic actions that compose the motion important. In many conventional methods, We was extracting basic actions such as “stand up” and “walk” using a template. However, in this paper, we propose a segmentation method that extract basic action automatically and find a series of actions (called a cycle) that composed as a series of basic actions for actions that repeat the same action like factory work. The proposed method consists of a motif discovery process for finding repetitive similar subsequences, and a grammar inference process for symbolizing the motifs and extracting the arrangement pattern as a cycle. Then, we evaluated extraction rate of the basic action and cycle of the proposed method about packing work and the screw tightening work and we confirmed proposed method can extract mostly correct.