Activity Recognition System Optimisation Using Triaxial Accelerometers

Li, Z., Li, B. and Le Kernec, J. (2020) Activity Recognition System Optimisation Using Triaxial Accelerometers. In: International Conference on 3D Imaging Technology (IC3DIT2019), Kunming, Yunnan, China, 15-18 Aug 2019, ISBN 9789811538667 (doi: 10.1007/978-981-15-3867-4_15)

[img] Text
193779.pdf - Accepted Version



Activity recognition is required in various applications such as motion analysis and health care. The accelerometer is a small, economical, easily deployed and high-performance sensor, which can continuously provide acceleration data from the body part it is worn. Previously, the human activity recognition system researches utilising triaxial accelerometer have mainly focused on one placement of the sensor, and rarely states the reason for the choice of sensor placement. This paper presents an optimisation method utilising a triaxial accelerometer when the sensor is placed on different body parts. The statistical characteristics-based algorithms use data from a motion-captured database to classify six classes of daily living activities. Feature selection is performed using the principal component analysis (PCA) from a range of features. Robust and sensitive features that highly contribute to the classification performance are selected. Activity classification is performed using the support vector machine (SVM) and K-nearest neighbour (K-NN) and the results are compared. Based on the HDM05 Mocap database with six activity types (overall 89 motions) collected from five subjects, the best place for wearable accelerometers is the waist, followed by chest, head, left wrist, right wrist, humerus and femur. Based on the preliminary results, multi-accelerometers and data fusion methods are utilised for further increasing the accuracy of classification, where the accuracy increases by 6.69% for SVM and 7.99% for KNN. For two sensors, the best placements for sensors are the waist with the left wrist, followed by the waist with the right wrist, waist with chest, waist with the humerus, waist with head and waist with femur. The result provides a guideline for sensor placement when developing an activity recognition system.

Item Type:Conference Proceedings
Additional Information:This work is supported by the Natural Science Foundation Project of Chongqing Science and Technology Commission (Grant No. cstc2015jcyjBX0113).
Glasgow Author(s) Enlighten ID:Le Kernec, Dr Julien and Li, Zhenghui
Authors: Li, Z., Li, B., and Le Kernec, J.
College/School:College of Science and Engineering > School of Engineering > Systems Power and Energy
Copyright Holders:Copyright © Springer Nature Singapore Pte Ltd. 2020
First Published:First published in Kountchev R., Patnaik S., Shi J., Favorskaya M. (eds) Advances in 3D Image and Graphics Representation, Analysis, Computing and Information Technology. Smart Innovation, Systems and Technologies, vol 180.
Publisher Policy:Reproduced in accordance with the publisher copyright policy

University Staff: Request a correction | Enlighten Editors: Update this record