Work

Posture Classification for Real life, Real time *

Year: 2012
Type: dissertation
Abstract: In recent years, Body Sensor Networks (BSNs) have been used as the basis of many systems aimed at\nmonitoring bodily parameters ranging from skin temperature, to breathing, to motion. These measurements\ncan then be used to generate additional information related to the monitored subject, such as for\nheat stress prediction or fall detection.\nThis thesis is concerned with the design, development and realistic evaluation of a BSN-based end-toend\nposture classification platform using on-body accelerometers. The work is motivated by applications\nthat require stable, sub-second, end-to-end classification of postures, as well as dynamically configurable\noperation to support exploratory data collection. Classification is performed on-node, thus reducing\nthe amount of data/information transmitted from the wearable nodes to a data sink. The work\nis experimentally-led, and uses an application case study—on-body monitoring of Explosive Ordnance\nDisposal (EOD) operatives—to provide context for system requirements and experimentation performed.\nThis thesis provides three main contributions:\nFirst, the design of a platform that allows real-time on-body classification of static and dynamic\npostures—a capability not present in existing work. The specific posture set consists of six static postures\n(sitting, standing, kneeling, and lying on back, front and one side) and two dynamic postures\n(walking and crawling), of which kneeling and crawling are not commonly considered in the literature.\nClassification is performed on a small, light embedded device using a simple easy-to-implement algorithm.\nThe classification algorithm used is a C4.5 decision tree, with a temporal feature (windowed variance) to\naid in distinguishing dynamic and static postures. Offline classification accuracy is shown to be 96.3%\nbased on data gathered from subjects in a laboratory environment, and real-time on-node classification\naccuracy is shown to be comparable to this figure (95.5%).\nSecond, further advance beyond the state of the art is presented through an investigation into posture\ntransitions. Posture transitions cause transient (<1s) posture changes in the classifier output and\nare shown to reduce classifier accuracy by 2% for every transition / minute for classifiers not specifically\ndesigned to handle transitions. Three posture filters that remove such transient posture changes\nare designed, implemented and tested on experimental data. The best performing filter, Exponentially\nWeighted Voting (EWV), is shown to reduce posture change events by 75.2% and increase accuracy by 1%\n(over unfiltered results). Compared to streaming raw data, an event-based posture classification system\nis shown to reduce transmissions by 98.5% (66-fold reduction).\nFinally, a broad investigation is presented into the effect of both system-related and training-process\nfactors on the accuracy of machine learning-based posture classifiers. The factors analysed include i) temporal\nand feature parameters, ii) sensor sampling rate, iii) number of sensors used, iv) posture class\naggregation and v) number of subjects used for training. Optimal parameters are determined for the\nmotivating EOD application, with a range of parameter values shown to guide development of other\nclassifiers.\nThrough the novel contributions presented, this thesis provides a solid groundwork for further research\nin BSN based posture classification systems and simplifies optimisation of machine-learning classifiers for\nspecific posture classification applications.
Author: Rednic, R.
Institution: Coventry University
Language: English
Open Access status: green
OpenAlex ID: W7014575026
Cites: 108
Cited by:
Related to: 10