A few accelerometers positioned on different parts of the body can be used to accurately classify... more A few accelerometers positioned on different parts of the body can be used to accurately classify steady state behaviour, such as walking, running, or sitting. Such systems are usually built using supervised learning approaches. Transitions between postures are, however, difficult to deal with using posture classification systems proposed to date, since there is no label set for intermediary postures and also the exact point at which the transition occurs can sometimes be hard to pinpoint. The usual bypass when using supervised learning to train such systems is to discard a section of the dataset around each transition. This leads to poorer classification performance when the systems are deployed out of the laboratory and used on-line, particularly if the regimes monitored involve fast paced activity changes. Time-based filtering that takes advantage of sequential patterns is a potential mechanism to improve posture classification accuracy in such real-life applications. Also, such filtering should reduce the number of event messages needed to be sent across a wireless network to track posture remotely, hence extending the system’s life. To support time-based filtering, understanding transitions, which are the major event generators in a classification system, is key. This work examines three approaches to post-process the output of a posture classifier using time-based filtering: a naïve voting scheme, an exponentially weighted voting scheme, and a Bayes filter. Best performance is obtained from the exponentially weighted voting scheme although it is suspected that a more sophisticated treatment of the Bayes filter might yield better results.
This paper examines the benefits of edge mining -data mining that takes place on the wireless, b... more This paper examines the benefits of edge mining -data mining that takes place on the wireless, battery-powered, and smart sensing devices that sit at the edge points of the Internet of Things. Through local data reduction and transformation, edge mining can quantifiably reduce the number of packets that must be sent, reducing energy usage, and remote storage requirements. In addition, edge mining has the potential to reduce the risk in personal privacy through embedding of information requirements at the sensing point, limiting inappropriate use. The benefits of edge mining are examined with respect to three specific algorithms: linear Spanish inquisition protocol (L-SIP), ClassAct, and bare necessities (BN), which are all instantiations of general SIP. In general, the benefits provided by edge mining are related to the predictability of data streams and availability of precise information requirements; results show that L-SIP typically reduces packet transmission by around 95% (20-fold), BN reduces packet transmission by 99.98% (5000-fold), and ClassAct reduces packet transmission by 99.6% (250-fold). Although energy reduction is not as radical because of other overheads, minimization of these overheads can lead up to a 10-fold battery life extension for L-SIP, for example. These results demonstrate the importance of edge mining to the feasibility of many IoT applications.
Energy, both in terms of its production and its usage has occupied a prime place in research as w... more Energy, both in terms of its production and its usage has occupied a prime place in research as well as politics and world economy for the past few years. The majority of nations are aiming to deliver severe carbon cuts in the next few years. However, achieving a carbon-free future needs more than infrastructure investment and novel efficient technologies for buildings, transportation and other large consumer domains. It needs a better understanding of people as consumers, as well as a better understanding of energy ...
... 28 4.3.2 Tree regeneration . . . . . ... The last method, robot assisted tracking, is used in... more ... 28 4.3.2 Tree regeneration . . . . . ... The last method, robot assisted tracking, is used in rehabilitation. Human movement is reflected using electromechanical and electromagnetic sensors attached to the body [1, 3, 4, 5]. ...
This paper presents the application scenario, conceptual overview and implementation of a monitor... more This paper presents the application scenario, conceptual overview and implementation of a monitoring system targeted at monitoring EOD suit wearers during missions. The system's aim is to deliver prediction of heat stress risk in the operative and provide actuation of a cooling system integrated within the suit. Prior work established that such prediction requires real-time autonomous processing of skin temperature and body acceleration data, and thus a system implementation is presented based on two interacting subsystems that perform the required sensing and data processing. Posture classification is performed with an accuracy of 96.1%, and a heat stress prediction algorithm is demonstrated with an overall accuracy of 88.5% when predicting the occurrence of heat stress within the next 2 minutes.
This paper presents an investigation into the design space for real-time, wearable posture classi... more This paper presents an investigation into the design space for real-time, wearable posture classification systems; specifically, it analyses the impact of various factors/design choices on classification accuracy when using C4.5 decision trees. The factors can be broadly divided into: 1) system factors (such as sensor sampling rate and number of sensors used) and 2) algorithm and training factors (such as quantity of training data and temporal data features used). These factors are analysed in the context of a case study involving postural activity monitoring of Explosive Ordinance Disposal (EOD) operatives. The case study involves classifying a set of eight postures commonly encountered in EOD missions: sitting, walking, crawling, laying (on all sides) and kneeling. Design guidelines and generic lessons for a wider class of applications can be drawn from the work.
We present MercuryLive, a web-enhanced extension to a body sensor network platform for continuous... more We present MercuryLive, a web-enhanced extension to a body sensor network platform for continuous home-based body motion sensing, interactive supervised data collection sessions, and long-term activity data analysis. The major goal of MercuryLive is to enable practical long-term health monitoring in a home setting and henceforth reduce the effort and cost for collecting clinically relevant quantitative measures on patients' health conditions during daily activities. MercuryLive contains three tiers: a central web server for streaming and storage of sensor data, a sensor data collection engine, and a user-friendly web-based GUI client. The platform is currently used in clinical studies on Parkinson's disease.
This letter introduces MercuryLive, a platform to enable home monitoring of patients with Parkins... more This letter introduces MercuryLive, a platform to enable home monitoring of patients with Parkinson's disease (PD) using wearable sensors. MercuryLive contains three tiers: a resource-aware data collection engine that relies upon wearable sensors, web services for live streaming and storage of sensor data, and a web-based graphical user interface client with video conferencing capability. Besides, the platform has the capability of analyzing sensor (i.e., accelerometer) data to reliably estimate clinical scores capturing the severity of tremor, bradykinesia, and dyskinesia. Testing results showed an average data latency of less than 400 ms and video latency of about 200 ms with video frame rate of about 13 frames/s when 800 kb/s of bandwidth were available and we used a 40% video compression, and data feature upload requiring 1 min of extra time following a 10 min interactive session. These results indicate that the proposed platform is suitable to monitor patients with PD to facilitate the titration of medications in the late stages of the disease.
A few accelerometers positioned on different parts of the body can be used to accurately classify... more A few accelerometers positioned on different parts of the body can be used to accurately classify steady state behaviour, such as walking, running, or sitting. Such systems are usually built using supervised learning approaches. Transitions between postures are, however, difficult to deal with using posture classification systems proposed to date, since there is no label set for intermediary postures and also the exact point at which the transition occurs can sometimes be hard to pinpoint. The usual bypass when using supervised learning to train such systems is to discard a section of the dataset around each transition. This leads to poorer classification performance when the systems are deployed out of the laboratory and used on-line, particularly if the regimes monitored involve fast paced activity changes. Time-based filtering that takes advantage of sequential patterns is a potential mechanism to improve posture classification accuracy in such real-life applications. Also, such filtering should reduce the number of event messages needed to be sent across a wireless network to track posture remotely, hence extending the system’s life. To support time-based filtering, understanding transitions, which are the major event generators in a classification system, is key. This work examines three approaches to post-process the output of a posture classifier using time-based filtering: a naïve voting scheme, an exponentially weighted voting scheme, and a Bayes filter. Best performance is obtained from the exponentially weighted voting scheme although it is suspected that a more sophisticated treatment of the Bayes filter might yield better results.
This paper examines the benefits of edge mining -data mining that takes place on the wireless, b... more This paper examines the benefits of edge mining -data mining that takes place on the wireless, battery-powered, and smart sensing devices that sit at the edge points of the Internet of Things. Through local data reduction and transformation, edge mining can quantifiably reduce the number of packets that must be sent, reducing energy usage, and remote storage requirements. In addition, edge mining has the potential to reduce the risk in personal privacy through embedding of information requirements at the sensing point, limiting inappropriate use. The benefits of edge mining are examined with respect to three specific algorithms: linear Spanish inquisition protocol (L-SIP), ClassAct, and bare necessities (BN), which are all instantiations of general SIP. In general, the benefits provided by edge mining are related to the predictability of data streams and availability of precise information requirements; results show that L-SIP typically reduces packet transmission by around 95% (20-fold), BN reduces packet transmission by 99.98% (5000-fold), and ClassAct reduces packet transmission by 99.6% (250-fold). Although energy reduction is not as radical because of other overheads, minimization of these overheads can lead up to a 10-fold battery life extension for L-SIP, for example. These results demonstrate the importance of edge mining to the feasibility of many IoT applications.
Energy, both in terms of its production and its usage has occupied a prime place in research as w... more Energy, both in terms of its production and its usage has occupied a prime place in research as well as politics and world economy for the past few years. The majority of nations are aiming to deliver severe carbon cuts in the next few years. However, achieving a carbon-free future needs more than infrastructure investment and novel efficient technologies for buildings, transportation and other large consumer domains. It needs a better understanding of people as consumers, as well as a better understanding of energy ...
... 28 4.3.2 Tree regeneration . . . . . ... The last method, robot assisted tracking, is used in... more ... 28 4.3.2 Tree regeneration . . . . . ... The last method, robot assisted tracking, is used in rehabilitation. Human movement is reflected using electromechanical and electromagnetic sensors attached to the body [1, 3, 4, 5]. ...
This paper presents the application scenario, conceptual overview and implementation of a monitor... more This paper presents the application scenario, conceptual overview and implementation of a monitoring system targeted at monitoring EOD suit wearers during missions. The system's aim is to deliver prediction of heat stress risk in the operative and provide actuation of a cooling system integrated within the suit. Prior work established that such prediction requires real-time autonomous processing of skin temperature and body acceleration data, and thus a system implementation is presented based on two interacting subsystems that perform the required sensing and data processing. Posture classification is performed with an accuracy of 96.1%, and a heat stress prediction algorithm is demonstrated with an overall accuracy of 88.5% when predicting the occurrence of heat stress within the next 2 minutes.
This paper presents an investigation into the design space for real-time, wearable posture classi... more This paper presents an investigation into the design space for real-time, wearable posture classification systems; specifically, it analyses the impact of various factors/design choices on classification accuracy when using C4.5 decision trees. The factors can be broadly divided into: 1) system factors (such as sensor sampling rate and number of sensors used) and 2) algorithm and training factors (such as quantity of training data and temporal data features used). These factors are analysed in the context of a case study involving postural activity monitoring of Explosive Ordinance Disposal (EOD) operatives. The case study involves classifying a set of eight postures commonly encountered in EOD missions: sitting, walking, crawling, laying (on all sides) and kneeling. Design guidelines and generic lessons for a wider class of applications can be drawn from the work.
We present MercuryLive, a web-enhanced extension to a body sensor network platform for continuous... more We present MercuryLive, a web-enhanced extension to a body sensor network platform for continuous home-based body motion sensing, interactive supervised data collection sessions, and long-term activity data analysis. The major goal of MercuryLive is to enable practical long-term health monitoring in a home setting and henceforth reduce the effort and cost for collecting clinically relevant quantitative measures on patients' health conditions during daily activities. MercuryLive contains three tiers: a central web server for streaming and storage of sensor data, a sensor data collection engine, and a user-friendly web-based GUI client. The platform is currently used in clinical studies on Parkinson's disease.
This letter introduces MercuryLive, a platform to enable home monitoring of patients with Parkins... more This letter introduces MercuryLive, a platform to enable home monitoring of patients with Parkinson's disease (PD) using wearable sensors. MercuryLive contains three tiers: a resource-aware data collection engine that relies upon wearable sensors, web services for live streaming and storage of sensor data, and a web-based graphical user interface client with video conferencing capability. Besides, the platform has the capability of analyzing sensor (i.e., accelerometer) data to reliably estimate clinical scores capturing the severity of tremor, bradykinesia, and dyskinesia. Testing results showed an average data latency of less than 400 ms and video latency of about 200 ms with video frame rate of about 13 frames/s when 800 kb/s of bandwidth were available and we used a 40% video compression, and data feature upload requiring 1 min of extra time following a 10 min interactive session. These results indicate that the proposed platform is suitable to monitor patients with PD to facilitate the titration of medications in the late stages of the disease.
Uploads
Papers by Ramona Rednic
using supervised learning approaches. Transitions between postures are, however, difficult to deal with using posture classification systems proposed to date, since there is no label set for intermediary postures and also the exact point at which the transition occurs can sometimes be hard to pinpoint. The
usual bypass when using supervised learning to train such systems is to discard a section of the dataset around each transition. This leads to poorer classification performance when the systems are deployed out of the laboratory and used on-line, particularly if the regimes monitored involve fast paced activity changes. Time-based filtering that takes advantage of sequential patterns is a potential mechanism to improve posture classification accuracy in such real-life applications. Also, such filtering should reduce the number of event messages needed to be sent across a wireless network to track posture remotely, hence extending the system’s life. To support time-based filtering, understanding transitions, which are the major event generators in a classification system, is key. This work examines three approaches to post-process the output of a posture classifier using time-based filtering: a naïve voting
scheme, an exponentially weighted voting scheme, and a Bayes filter. Best performance is obtained from the exponentially weighted voting scheme although it is suspected that a more sophisticated treatment of the Bayes filter might yield better results.
using supervised learning approaches. Transitions between postures are, however, difficult to deal with using posture classification systems proposed to date, since there is no label set for intermediary postures and also the exact point at which the transition occurs can sometimes be hard to pinpoint. The
usual bypass when using supervised learning to train such systems is to discard a section of the dataset around each transition. This leads to poorer classification performance when the systems are deployed out of the laboratory and used on-line, particularly if the regimes monitored involve fast paced activity changes. Time-based filtering that takes advantage of sequential patterns is a potential mechanism to improve posture classification accuracy in such real-life applications. Also, such filtering should reduce the number of event messages needed to be sent across a wireless network to track posture remotely, hence extending the system’s life. To support time-based filtering, understanding transitions, which are the major event generators in a classification system, is key. This work examines three approaches to post-process the output of a posture classifier using time-based filtering: a naïve voting
scheme, an exponentially weighted voting scheme, and a Bayes filter. Best performance is obtained from the exponentially weighted voting scheme although it is suspected that a more sophisticated treatment of the Bayes filter might yield better results.