Pre-Processing Effect on the Accuracy of Event-Based Activity Segmentation and Classification through Inertial Sensors

Sensors

Benish Fida¹  Ivan Bernabucci²  Daniele Bibbo²  Silvia Conforto²  Maurizio Schmid² 

[1] Department of Engineering, University of Roma Tre, Via Vito Volterra, 62, Rome 00146, Italy;

关键词: inertial measurement unit;    gait event detection;    dynamic segmentation;    pre-processing;    physical activity;    classification;   

DOI  :  10.3390/s150923095

来源: mdpi

【 摘 要 】

Inertial sensors are increasingly being used to recognize and classify physical activities in a variety of applications. For monitoring and fitness applications, it is crucial to develop methods able to segment each activity cycle, e.g., a gait cycle, so that the successive classification step may be more accurate. To increase detection accuracy, pre-processing is often used, with a concurrent increase in computational cost. In this paper, the effect of pre-processing operations on the detection and classification of locomotion activities was investigated, to check whether the presence of pre-processing significantly contributes to an increase in accuracy. The pre-processing stages evaluated in this study were inclination correction and de-noising. Level walking, step ascending, descending and running were monitored by using a shank-mounted inertial sensor. Raw and filtered segments, obtained from a modified version of a rule-based gait detection algorithm optimized for sequential processing, were processed to extract time and frequency-based features for physical activity classification through a support vector machine classifier. The proposed method accurately detected >99% gait cycles from raw data and produced >98% accuracy on these segmented gait cycles. Pre-processing did not substantially increase classification accuracy, thus highlighting the possibility of reducing the amount of pre-processing for real-time applications.

【 授权许可】

CC BY   
© 2015 by the authors; licensee MDPI, Basel, Switzerland.