TR2019-150

Time Series Segmentation with Leg Analysis for Human Motion Analysis


    •  Imamura, M., Inoue, M., Terada, M., Nikovski, D.N., "Time Series Segmentation with Leg Analysis for Human Motion Analysis", International Workshop on Informatics, Mizuno, T., Eds., October 2019, pp. 163-168.
      BibTeX TR2019-150 PDF
      • @inproceedings{Imamura2019oct,
      • author = {Imamura, Makoto and Inoue, Mao and Terada, Masahiro and Nikovski, Daniel N.},
      • title = {Time Series Segmentation with Leg Analysis for Human Motion Analysis},
      • booktitle = {International Workshop on Informatics},
      • year = 2019,
      • editor = {Mizuno, T.},
      • pages = {163--168},
      • month = oct,
      • publisher = {Informatics Society},
      • isbn = {978-4-902523-46-1},
      • url = {https://www.merl.com/publications/TR2019-150}
      • }
  • MERL Contact:
  • Research Area:

    Data Analytics

Abstract:

To enable automated analysis of human motion data collected by acceleration sensors, gyro senors, or motion capture devices, an approach for accurately segmenting primitive actions is required. Whereas most existing approaches use templates of basic actions such as “stand up”, “walk” and “sit down”, we introduce a novel problem called “structural motif discovery” that aims to find segments without templates from repetitive routine motion that consists of regularly ordered (actions?). We also propose a novel segmentation method that approximates the time series with a sequence of convex-shaped patterns by means of leg analysis, which is parameter-free and its complexity is O(N), where N is the length of a given time series. The experimental results show that our proposed method is effective for both simulation data and real data and real data from repetitive assembly operations,