Ad Hoc Networks. 10th EAI International Conference, ADHOCNETS 2018, Cairns, Australia, September 20-23, 2018, Proceedings

Research Article

An Action Recognition Method Based on Wearable Sensors

Download
141 downloads
  • @INPROCEEDINGS{10.1007/978-3-030-05888-3_19,
        author={Fuliang Ma and Jing Tan and Xiubing Liu and Huiqiang Wang and Guangsheng Feng and Bingyang Li and Hongwu Lv and Junyu Lin and Mao Tang},
        title={An Action Recognition Method Based on Wearable Sensors},
        proceedings={Ad Hoc Networks. 10th EAI International Conference, ADHOCNETS 2018, Cairns, Australia, September 20-23, 2018, Proceedings},
        proceedings_a={ADHOCNETS},
        year={2018},
        month={12},
        keywords={Action recognition Wearable sensors Wearing scheme},
        doi={10.1007/978-3-030-05888-3_19}
    }
    
  • Fuliang Ma
    Jing Tan
    Xiubing Liu
    Huiqiang Wang
    Guangsheng Feng
    Bingyang Li
    Hongwu Lv
    Junyu Lin
    Mao Tang
    Year: 2018
    An Action Recognition Method Based on Wearable Sensors
    ADHOCNETS
    Springer
    DOI: 10.1007/978-3-030-05888-3_19
Fuliang Ma1, Jing Tan1, Xiubing Liu1, Huiqiang Wang1, Guangsheng Feng1,*, Bingyang Li1, Hongwu Lv1, Junyu Lin2, Mao Tang3
  • 1: Harbin Engineering University
  • 2: Institute of Information Engineering, Chinese Academy of Sciences
  • 3: Science and Technology Resource Sharing Service Center of Heilongjiang
*Contact email: fengguangsheng@hrbeu.edu.cn

Abstract

In the field of human action recognition, some existing works are mainly focused on macro actions, e.g., the requirements for action recognition is walking or jumping, while others are concentrated on micro actions, e.g., hand waving or leg raising. However, existing works rarely consider the recognition effect of different sensor wearing schemes with various requirements. In this work, the influences of the wearing scheme on action recognition effect are taken into account, a universal action recognition method to adapt different recognition requirements is developed. First, we present an action layered verification model which includes static action layer, dynamic action layer and joint presentation layer, which is used to provide an optional wearing scheme for each layer and to prevent wrong classification problems. Second, we verify the recognition effect of various wearing schemes under different layers. Finally, an action recognition method based on decision tree is introduced to adapt different requirements. The experiments show that the proposed method achieves a desirable recognition effect in comparison to existing ones.