A Phase Variable Approach for IMU-Based Locomotion Activity Recognition

• Published in: IEEE transactions on biomedical engineering Vol. 65; no. 6; pp. 1330 - 1338
• Main Authors: Bartlett, Harrison L., Goldfarb, Michael
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.06.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Activity recognition; Ascent; Classification; Classification algorithms; Curvature; Discriminant analysis; Feature extraction; Gait; Gait recognition; Identification methods; Inertial platforms; Legged locomotion; Locomotion; Measurement methods; Monitoring; Neural networks; patient monitoring; Pattern recognition; Space vehicles; Thigh; Training
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/TBME.2017.2750139

Objective: This paper describes a gait classification method that utilizes measured motion of the thigh segment provided by an inertial measurement unit. Methods: The classification method employs a phase-variable description of gait, and identifies a given activity based on the expected curvature characteristics of that activity over a gait cycle. The classification method was tested in experiments conducted with seven healthy subjects performing three different locomotor activities: level ground walking, stair descent, and stair ascent. Classification accuracy of the phase variable classification method was assessed for classifying each activity, and transitions between activities, and compared to a linear discriminant analysis (LDA) classifier as a benchmark. Results: For the subjects tested, the phase variable classification method outperformed LDA when using nonsubject-specific training data, while the LDA outperformed the phase variable approach when using subject-specific training. Conclusions: The proposed method may provide improved classification accuracy for gait classification applications trained with nonsubject-specific data. Significance: This paper offers a new method of gait classification based on a phase variable description. The method is shown to provide improved classification accuracy relative to an LDA pattern recognition framework when trained with nonsubject-specific data.