Volume XLII-2/W12
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W12, 189-195, 2019
https://doi.org/10.5194/isprs-archives-XLII-2-W12-189-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W12, 189-195, 2019
https://doi.org/10.5194/isprs-archives-XLII-2-W12-189-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

  09 May 2019

09 May 2019

A SKELETON FEATURES-BASED FALL DETECTION USING MICROSOFT KINECT V2 WITH ONE CLASS-CLASSIFIER OUTLIER REMOVAL

O. S. Seredin1, A. V. Kopylov1, S.-C. Huang2, and D. S. Rodionov1 O. S. Seredin et al.
  • 1Tula State University, Institute of Applied Mathematics and Computer Science, 300012 Tula, Russia
  • 2National Taipei University of Technology, Department of Electronic Engineering, Taipei 106, Taiwan

Keywords: Fall Detection, Movement Analysis, Skeleton Description, RGB-D Camera, Privacy Preserving Elderly People Care

Abstract. The real-time and robust fall detection is one of the key components of elderly people care and monitoring systems. Depth sensors, as they became more available, occupy an increasing place in event recognition systems. Some of them can directly produce a skeletal description of the human figure for compact representation of a person’s posture. Skeleton description makes the output of source video or detailed information about the depth outside the system unnecessary and raises the privacy of the entire system. Based on a comparative study of different RGB-D cameras, the most promising model for further development was chosen - Microsoft Kinect v2. The TST Fall Detection Dataset v2 is used here as a base for experiments. The proposed algorithm is based on the skeleton features encoding on the sequence of neighboring frames and support vector machine classifier. A version of a cumulative sum method is applied for combining the individual decisions on the consecutive frames. It is offered to use the one-class classifier for detection of low-quality skeletons. The 0.958 accuracy of our fall detection procedure was obtained in the cross-validation procedure based on the removal of records of a particular person from the database (Leave-one-Person-out).