Show simple item record

dc.contributor.advisorThiel, David
dc.contributor.authorWixted, Andrew James
dc.date.accessioned2018-01-23T02:23:44Z
dc.date.available2018-01-23T02:23:44Z
dc.date.issued2007
dc.identifier.doi10.25904/1912/504
dc.identifier.urihttp://hdl.handle.net/10072/365884
dc.description.abstractAbstract :The original core question of this thesis related to the real-time wireless collection of data from teams of athletes, in essence a technology - engineering issue. However, as time progressed, more fundamental questions evolved: What data is being collected, what does the data mean and how should it be interpreted? This question lies on the intersection of human movement studies, in the form of biomechanical & physiological sports science, and digital signal processing. The key biomechanical and physiological questions also feed back into the original real-time data collection issue. If data collected from athlete-mounted transducers1 is correctly interpreted, the salient features can be extracted and subsequently stored or transferred across a wireless network to a host system for analysis. Consequently this thesis addresses two intertwined issues, the real time in-situ data collection and the data processing necessary to extract some specific information useful to sports scientists. In-situ athlete monitoring - data collection. In the past, much athlete monitoring was performed under invasive conditions, either laboratory conditions or in specially prepared circumstances that were in-effect portable laboratories. Biomechanical analysis occurred on specially prepared rowing sculls or on running tracks with built-in instrumentation or tracks monitored by specialised high-speed cameras. Physiological monitoring required heart-rate monitoring, sampling of body fluids, monitoring of expired gasses and numerous other invasive techniques. With the advent of Micro-Electro Mechanical Systems (MEMS) such as gyroscopes, magnetometers, accelerometers etc., the opportunity existed for the in-situ monitoring of athletes both in training and in competition. While in-situ athlete monitoring can be simple data logging from a single transducer, the modern sports scientist needs more flexibility to gather the necessary information. Whether data is collected from multiple intra-athlete sensors or multiple inter-athlete sensors there must be a mechanism for synchronising that data so that it can be correctly recombined. This thesis addresses the synchronous data collection through the development of a real-time operating system, designed to operate as a node within a synchronised wireless sensor network. This thesis proposes wireless protocols, or 1 In this thesis the predominate transducer is the inertial sensor or 'accelerometer' although data from other transducers is also analysed. topologies, necessary for the operation of low-powered synchronised wireless sensor networks. Finally, the issue of appropriate data sampling, necessary to maximise retained information and minimise bandwidth, is addressed in a discussion on data compression. This discussion analyses the raw data from the viewpoints of the data collection and transport process and the useful extractable content of the data. Each aspect of the system was filtered through the necessity of minimising the effect of the monitoring system on the system under observation. In the case of elite athletes this means making a system as unobtrusive as possible, which ultimately impacts the power available for data collection, synchronisation, processing and transport. Data interpretation and feature extraction. There appears to be enormous potential for data collection and analysis as it relates to elite athletes and elite level sport. In this thesis only one or two thin slices of sports science, as it relates to elite athletes, were investigated. These investigations addressed specific questions relating to the automation of physiological monitoring of rugby football players. Accelerometers have been used as tools for the estimation of human energy expenditure (EE). The EE estimator was known as 'counts' and, while effective for estimating EE of humans walking, was ineffective for estimating EE during running activities. It was hypothesised that the ineffectiveness was due to some form of biomechanical running efficiency factor that upset the EE estimator, and that this biomechanical efficiency factor may be extractable from the collected data. To extract apparently useful information, a range of signal processing techniques were investigated and their usefulness, in terms of correctness, processing efficiency, accuracy and interpretability analysed. The 'running efficiency factor' was investigated through biomechanical analysis of systematically collected multi-athlete multi-speed treadmill data. The biomechanical analysis (documented in Ch.6) while inconclusive on the original running-efficiency question suggested the possibility of other applications, such as running technique coaching, for this form of data collection and analysis. The biomechanical analysis, in conjunction with an understanding of the related kinematics ultimately lead to the identification of features in the athlete data that could be used as EE estimators. Linear regression analysis identified that the athlete's leg-length appeared to have a correlation with several extracted parameters. One of these parameters was the athlete's natural step-frequency or cadence. By combining, in a linear equation, an athlete's step frequency with the athlete's anthropomorphic measures of leg-length and mass, an EE estimator that was more effective than the accelerometer 'counts' based estimators was generated. This estimator was immune to a range of errors that affect 'counts' based estimators. Initially this EE estimator was extracted using the processing power of a desktop computer. It was necessary to move the processing to the lower power in-situ sensor platform. The underlying signal processing functions were replaced with potentially less accurate, low-power, signal processing functions. The effectiveness of these different techniques was evaluated. While each chapter of this thesis is essentially stand-alone, most chapters describe some major component of the in-situ athlete monitoring system. Therefore, the in-situ athlete monitoring - data collection, interpretation and feature extraction is described by the low-powered real-time operating system (Ch.3) which provides the platform. The optimal sampling system is identified from the signal analysis of Ch.7 as is the data compression system used to log on-board data. Ch.4 describes the various signal processing techniques and identifies appropriate and inappropriate techniques for extracting information. Ch.5 analyses the extraction of the EE estimate and details the low-power implementation of the total rugby feature extraction. Ch.2 describes and proposes wireless network implementations necessary to bring together multiple individual in-situ monitoring systems, to generate a cohesive team-sport in-situ monitoring system.
dc.languageEnglish
dc.publisherGriffith University
dc.publisher.placeBrisbane
dc.rights.copyrightThe author owns the copyright in this thesis, unless stated otherwise.
dc.subject.keywordsAthlete monitoring system
dc.subject.keywordsCohesive team-sport in-situ monitoring system
dc.subject.keywordsIndividual in-situ monitoring systems
dc.titleIn-situ Athlete Monitoring: Data Collection, Interpretation & Feature Extraction
dc.typeGriffith thesis
gro.facultyScience, Environment, Engineering and Technology
gro.rights.copyrightThe author owns the copyright in this thesis, unless stated otherwise.
gro.hasfulltextFull Text
dc.contributor.otheradvisorJames, Daniel
dc.rights.accessRightsPublic
gro.identifier.gurtIDgu1322532171109
gro.source.ADTshelfnoADT0932
gro.thesis.degreelevelThesis (PhD Doctorate)
gro.thesis.degreeprogramDoctor of Philosophy (PhD)
gro.departmentGriffith University. School of Engineering.
gro.griffith.authorWixted, Andrew James


Files in this item

This item appears in the following Collection(s)

Show simple item record