Matjaz Gams - Academia.edu (original) (raw)

Papers by Matjaz Gams

Research paper thumbnail of Detecting Falls with Location Sensors and Accelerometers

Proceedings of the AAAI Conference on Artificial Intelligence

Due to the rapid aging of the population, many technical solutions for the care of the elderly ar... more Due to the rapid aging of the population, many technical solutions for the care of the elderly are being developed, often involving fall detection with accelerometers. We present a novel approach to fall detection with location sensors. In our application, a user wears up to four tags on the body whose locations are detected with radio sensors. This makes it possible to recognize the user’s activity, including falling any lying afterwards, and the context in terms of the location in the apartment. We compared fall detection using location sensors, accelerometers and accelerometers combined with the context. A scenario consisting of events difficult to recognize as falls or non- falls was used for the comparison. The accuracy of the methods that utilized the context was almost 40 percentage points higher compared to the methods without the context. The accuracy of pure location-based methods was around 10 percentage points higher than the accuracy of accelerometers combined with the ...

Research paper thumbnail of Context-aware MAS to support elderly people (demonstration)

Adaptive Agents and Multi-Agents Systems, Jun 4, 2012

This paper presents a context-aware, multiagent system for care of the elderly. The system combin... more This paper presents a context-aware, multiagent system for care of the elderly. The system combines state-of-the-art sensor technologies to detect falls and other health problems, and calls for help in the case of an emergency or issues a warning in cases not needing urgent attention. When deployed at the home of an elderly person it provides them with 24-hour monitoring. Consequently, the elderly may live alone at home, even at an advanced age. The health problems are detected with six groups of agents processing the sensor data and augmenting the data with higher-level information, such as the posture of the person, his/her activity and the context of the situation's environment. The system has been tested in several live demonstrations, where it achieved an excellent performance in complex situations. The system is based on the set of agents observing the elderly person from various points of view, and combining the location and inertial sensors to provide context awareness.

Research paper thumbnail of Using accelerometers to improve position-based activity recognition

This paper will present the results of the research conducted on wireless accelerometers in fall ... more This paper will present the results of the research conducted on wireless accelerometers in fall detection and activity recognition. This research is a part of the Confidence project, whose goal is to provide a health monitoring system for the elderly. Normally, position-based body tags are used to detect postures and activities. This paper reports the results of using accelerometers as both a supplement to and substitute of the position tags. It introduces a combined approach based on machine-learning and wave analysis. Preliminary results indicate an important increase in activity recognition accuracy when using both acceleration and position tags. 3 MOVING FILTER An important step in classifying activities is the ability to detect whether the person is moving. It helps to distinguish between static activities (e.g. lying, sitting) and dynamic ones (e.g. standing up, going down). The moving detection used in our approach uses the data sent by the chest

Research paper thumbnail of Confidence: Ubiquitous Care System to Support Independent Living

The Confidence system aims at helping the elderly stay independent longer by detecting falls and ... more The Confidence system aims at helping the elderly stay independent longer by detecting falls and unusual movement which may indicate a health problem. The system uses location sensors and wearable tags to determine the coordinates of the user's body parts, and an accelerometer to detect fall impact and movement. Machine learning is combined with domain knowledge in the form of rules to recognize the user's activity. The fall detection employs a similar combination of machine learning and domain knowledge. It was tested on five atypical falls and events that can be easily mistaken for a fall. We show in the paper and demo that neither sensor type can correctly recognize all of these events on its own, but the combination of both sensor types yields highly accurate fall detection. In addition, the detection of unusual movement can observe both the user's micro-movement and macro-movement. This makes it possible for the Confidence system to detect most types of threats to t...

Research paper thumbnail of Surveillance Systems An Intelligent Indoor Surveillance System

CEPIS UPGRADE is the anchor point for UPENET (UPGRADE European NETwork), the network of CEPIS mem... more CEPIS UPGRADE is the anchor point for UPENET (UPGRADE European NETwork), the network of CEPIS member societies’ publications, that currently includes the following ones: • inforewiew, magazine from the Serbian CEPIS society JISA • Informatica, journal from the Slovenian CEPIS society SDI • Informatik-Spektrum, journal published by Springer Verlag on behalf of the CEPIS societies GI, Germany, and SI, Switzerland • ITNOW, magazine published by Oxford University Press on behalf of the British CEPIS society BCS • Mondo Digitale, digital journal from the Italian CEPIS society AICA • Novática, journal from the Spanish CEPIS society ATI • OCG Journal, journal from the Austrian CEPIS society OCG • Pliroforiki, journal from the Cyprus CEPIS society CCS • Tölvumál, journal from the Icelandic CEPIS society ISIP

Research paper thumbnail of Evolution of Activity Monitoring Through Various Projects

The paper presents the evolution of activity monitoring utilising wearable sensors through severa... more The paper presents the evolution of activity monitoring utilising wearable sensors through several sequential projects running in the last decade. It covers the change and improvement towards sensor technology that is more affordable and comfortable for the users and development of additional functionalities to achieve accurate activity monitoring in terms of activity recognition. All presented systems are developed to perform on-line activity recognition.

Research paper thumbnail of Context-based fall detection and activity recognition using inertial and location sensors

Journal of Ambient Intelligence and Smart Environments, 2014

Accidental falls are some of the most common sources of injury among the elderly. A fall is parti... more Accidental falls are some of the most common sources of injury among the elderly. A fall is particularly critical when the elderly person is injured and cannot call for help. This problem is addressed by many fall-detection systems, but they often focus on isolated falls under restricted conditions, not paying enough attention to complex, real-life situations. To achieve robust performance in real life, a combination of body-worn inertial and location sensors for fall detection is studied in this paper. A novel context-based method that exploits the information from the both types of sensors is designed. It considers body accelerations, location and elementary activities to detect a fall. The recognition of the activities is of great importance and also is the most demanding of the three, thus it is treated as a separate task. The evaluation is performed on a real-life scenario, including fast falls, slow falls and fall-like situations that are difficult to distinguish from falls. All possible combinations of six inertial and four location sensors are tested. The results show that: (i) context-based reasoning significantly improves the performance; (ii) a combination of two types of sensors in a single physical sensor enclosure is the best practical solution.

Research paper thumbnail of Adapting activity recognition to a person with Multi-Classifier Adaptive Training

Journal of Ambient Intelligence and Smart Environments, 2015

Activity-recognition classifiers, which label an activity based on sensor data, have decreased cl... more Activity-recognition classifiers, which label an activity based on sensor data, have decreased classification accuracy when used in the real world with a particular person. To improve the classifier, a Multi-Classifier Adaptive-Training algorithm (MCAT) is proposed. The MCAT adapts activity recognition classifier to a particular person by using four classifiers to utilise unlabelled data. The general classifier is trained on the labelled data available before deployment and retrieved in the controlled environment. The specific classifier is trained on a limited amount of labelled data belonging to the new person in the new environment. A domain-independent meta-classifier decides whether to classify a new instance with the general or specific classifier. The final, second meta-classifier decides whether to include the new instance into the training set of the general classifier. The general classifier is periodically retrained, gradually adapting to the new person in the new environment. The adaptation results were evaluated for statistical significance. Results showed that the MCAT outperforms competing approaches and significantly increases the initial activity-recognition classifier classification accuracy.

Research paper thumbnail of How Accurately Can Your Wrist Device Recognize Daily Activities and Detect Falls?

Sensors, 2016

Although wearable accelerometers can successfully recognize activities and detect falls, their ad... more Although wearable accelerometers can successfully recognize activities and detect falls, their adoption in real life is low because users do not want to wear additional devices. A possible solution is an accelerometer inside a wrist device/smartwatch. However, wrist placement might perform poorly in terms of accuracy due to frequent random movements of the hand. In this paper we perform a thorough, large-scale evaluation of methods for activity recognition and fall detection on four datasets. On the first two we showed that the left wrist performs better compared to the dominant right one, and also better compared to the elbow and the chest, but worse compared to the ankle, knee and belt. On the third (Opportunity) dataset, our method outperformed the related work, indicating that our feature-preprocessing creates better input data. And finally, on a real-life unlabeled dataset the recognized activities captured the subject's daily rhythm and activities. Our fall-detection method detected all of the fast falls and minimized the false positives, achieving 85% accuracy on the first dataset. Because the other datasets did not contain fall events, only false positives were evaluated, resulting in 9 for the second, 1 for the third and 15 for the real-life dataset (57 days data).

Research paper thumbnail of Fall Detection Using Location Sensors and Accelerometers

IEEE Pervasive Computing, 2015

The rapid aging of the population drives the development of pervasive solutions for the care of t... more The rapid aging of the population drives the development of pervasive solutions for the care of the elderly, which often involve fall detection with accelerometers. These solutions are very accurate in laboratory conditions, but can fail in some real-life situations. To overcome this, we present the Confidence system, which detects falls mainly with location sensors. A user wears one to four tags whose locations are detected with sensors. This allows recognizing the user's activity, including falling and lying afterwards, and the context in terms of the location in the apartment. A scenario consisting of events difficult to recognize as falls or non-falls was used to compare the Confidence system with accelerometer-based fall-detection methods, some of them augmented with the context from a location sensor. The accuracy of the methods that utilized the context was around 30 percentage points higher compared to the methods without the context. The Confidence system was also successfully validated in a real-life setting with elderly users.

Research paper thumbnail of Three-layer Activity Recognition Combining Domain Knowledge and Meta- classification Author list

Journal of Medical and Biological Engineering

One of the essential tasks of healthcare and smart-living systems is to recognize the current act... more One of the essential tasks of healthcare and smart-living systems is to recognize the current activity of a particular user. Such activity recognition (AR) is demanding when only limited sensors are used, such as accelerometers. Given a small number of accelerometers, intelligent AR systems often use simple architectures, either general or specific for their AR. In this paper, a system for AR named TriLAR is presented. TriLAR has an AR-specific architecture consisting of three layers: (i) a bottom layer, where an arbitrary number of AR methods can be used to recognize the current activity; (ii) a middle layer, where the predictions from the bottom-layer methods are inputs for a hierarchical structure that combines domain knowledge and meta-classification; and (iii) a top layer, where a hidden Markov model is used to correct spurious transitions between the recognized activities from the middle layer. The middle layer has a hierarchical, three-level structure. First, a meta-classifie...

Research paper thumbnail of Context-based ensemble method for human energy expenditure estimation

Applied Soft Computing, 2015

Monitoring human energy expenditure (EE) is important in many health and sports applications, sin... more Monitoring human energy expenditure (EE) is important in many health and sports applications, since the energy expenditure directly reflects the intensity of physical activity. The actual energy expenditure is unpractical to measure; therefore, it is often estimated from the physical activity measured with accelerometers and other sensors. Previous studies have demonstrated that using a person's activity as the context in which the EE is estimated, and using multiple sensors, improves the estimation. In this study, we go a step further by proposing a context-based reasoning method that uses multiple contexts provided by multiple sensors. The proposed Multiple Contexts Ensemble (MCE) approach first extracts multiple features from the sensor data. Each feature is used as a context for which multiple regression models are built using the remaining features as training data: for each value of the context feature, a regression model is trained on a subset of the dataset with that value. When evaluating a data sample, the models corresponding to the context (feature) values in the evaluated sample are assembled into an ensemble of regression models that estimates the EE of the user. Experiments showed that the MCE method outperforms (in terms of lower root means squared error and lower mean absolute error): (i) five single-regression approaches (linear and non-linear); (ii) two ensemble approaches: Bagging and Random subspace; (iii) an approach that uses artificial neural networks trained on accelerometer-data only; and (iv) BodyMedia (a state-of-the-art commercial EE-estimation device).

Research paper thumbnail of Discovering Health Problems in the Elderly using Data Mining Approach

This paper is presenting a generalized approach to detection of health problems and falls of the ... more This paper is presenting a generalized approach to detection of health problems and falls of the elderly for the purpose of prolonging autonomous living of elderly using a novel data mining approach. The movement of the user is captured with the motion capture system, which consists of the body-worn tags, whose coordinates are acquired by the sensors situated in the apartment. Output time-series of coordinates are modeled with the proposed data mining approach in order to recognize the specific health problem or fall. The approach is general in a sense that it uses k-nearest neighbor algorithm and dynamic time warping with time-series of all measurable joint angles for the attributes instead of the more specific approach with medically defined attributes. It is two-step approach; in the first step it classifies person's activities into five activities including different types of falls. In the second step it classifies classified walking instances from the first step into five different health states; one healthy and four unhealthy. Even though the new approach is more general and can be used to differentiate also from other types of activities or health problems, it achieves very high classification accuracies, similar to the more specific approaches from the literature.

Research paper thumbnail of Recognizing Human Activities and Detecting Falls in Real-time

The paper presents a system that recognizes human activities and detects falls in real-time. It c... more The paper presents a system that recognizes human activities and detects falls in real-time. It consists of two wearable accelerometers placed on the user's torso and thigh. The system is tuned for robustness and real-time performance by combining domain-specific rules and classifiers trained with machine learning. The offline evaluation of the system's performance was conducted on a dataset containing a wide range of activities and different types of falls. The F-measure of the activity recognition and fall detection were 96% and 78%, respectively. Additionally, the system was evaluated at the EvAAL-2013 activity recognition competition and awarded the first place, achieving the score of 83.6%, which was for 14.2 percentage points better than the secondplace system. The competition's evaluation was performed in a living lab using several criteria: recognition performance, user-acceptance, recognition delay, system installation complexity and interoperability with other systems.

Research paper thumbnail of Multi-Classifier Adaptive Training: Specialising an Activity Recognition Classifier Using Semi-supervised Learning

Lecture Notes in Computer Science, 2012

When an activity recognition classifier is deployed to be used with a particular user, its perfor... more When an activity recognition classifier is deployed to be used with a particular user, its performance can often be improved by adapting it to that user. To improve the classifier, we propose a novel semisupervised Multi-Classifier Adaptive Training algorithm (MCAT) that uses four classifiers. First, the General classifier is trained on the labelled data available before deployment. Second, the Specific classifier is trained on a limited amount of labelled data specific to the new user in the current environment. Third, a domain-independent meta-classifier decides whether to classify a new instance with the General or Specific classifier. Fourth, another meta-classifier decides whether to include the new instance in the training set for the General classifier. The General classifier is periodically retrained, gradually adapting to the new user in the new environment where it is deployed. The results show that our new algorithm outperforms competing approaches and increases the accuracy of the initial activity recognition classifier by 12.66 percentage points on average.

Research paper thumbnail of Recognition of Patterns of Health Problems and Falls in the Elderly Using Data Mining

Lecture Notes in Computer Science, 2012

We present a generalized data mining approach to the detection of health problems and falls in th... more We present a generalized data mining approach to the detection of health problems and falls in the elderly for the purpose of prolonging their autonomous living. The input for the data mining algorithm is the output of the motion-capture system. The approach is general since it uses a k-nearestneighbor algorithm and dynamic time warping with the time series of all the measurable joint angles for the attributes instead of a more specific approach with medically defined attributes. Even though the presented approach is more general and can be used to differentiate other types of activities or health problems, it achieves very high classification accuracies, similar to the more specific approaches described in the literature.

Research paper thumbnail of A Network of Sensor and Actuator Agents for Building Automation Systems

Atlantis Ambient and Pervasive Intelligence, 2013

Due to at least two reasons, energy consumption for heating, ventilation, air conditioning (HVAC)... more Due to at least two reasons, energy consumption for heating, ventilation, air conditioning (HVAC) and domestic hot water (DHW) systems should be reduced. First, the total energy consumption is high since nearly 20% of total energy in USA is accounted for HVAC systems. Sencond reason is exploitation of renevable energy sources, which depend on current time and weather situation. Thus, the energy management should improve to effective by managing currently available energy. For this purpose, the sensor actor agent network is proposed with an example on domestic hot water heating, taking into account several aspects including the occupant presence. The simulation model was used, where the water consumption affected system dynamics. The agent-based schema combined with the simulator enables an occupant to choose the most subjectively desirable policy of the DHW control mechanism.

Research paper thumbnail of Behavior Analysis Based on Coordinates of Body Tags

Lecture Notes in Computer Science, 2009

This paper describes fall detection, activity recognition and the detection of anomalous gait in ... more This paper describes fall detection, activity recognition and the detection of anomalous gait in the Confidence project. The project aims to prolong the independence of the elderly by detecting falls and other types of behavior indicating a health problem. The behavior will be analyzed based on the coordinates of tags worn on the body. The coordinates will be detected with radio sensors. We describe two Confidence modules. The first one classifies the user's activity into one of six classes, including falling. The second one detects walking anomalies, such as limping, dizziness and hemiplegia. The walking analysis can automatically adapt to each person by using only the examples of normal walking of that person. Both modules employ machine learning: the paper focuses on the features they use and the effect of tag placement and sensor noise on the classification accuracy. Four tags were enough for activity recognition accuracy of over 93 % at moderate sensor noise, while six were needed to detect walking anomalies with the accuracy of over 90 %.

Research paper thumbnail of Competitive Live Evaluations of Activity-Recognition Systems

IEEE Pervasive Computing, 2015

In order to ensure the validity and usability of activity recognition approaches, an agreement on... more In order to ensure the validity and usability of activity recognition approaches, an agreement on a set of standard evaluation methods is needed. Due to the diversity of the sensors and other hardware employed, designing and accepting standard tests is a difficult task. This article presents an initiative to evaluate activity recognition systems: a living-lab evaluation established through an annual competition − EvAAL-AR (Evaluating Ambient Assisted Living Systems through Competitive Benchmarking − Activity Recognition). In the competition, each team brings their own activity-recognition system, which is evaluated live on the same activity scenario performed by an actor. The evaluation criteria attempt to capture the practical usability: recognition accuracy, user acceptance, recognition delay, installation complexity, and interoperability with ambient assisted living systems. The article also presents the competing systems with emphasis on two best-performing ones: (i) a system that achieved the best recognition accuracy, and (ii) a system that was evaluated as the best overall. Finally, the article presents lessons learned from the competition and ideas for future development of the competition and of the activity recognition field in general.

Research paper thumbnail of Efficient Activity Recognition and Fall Detection Using Accelerometers

Communications in Computer and Information Science, 2013

Ambient assisted living (AAL) systems need to understand the user's situation, which makes activi... more Ambient assisted living (AAL) systems need to understand the user's situation, which makes activity recognition an important component. Falls are one of the most critical problems of the elderly, so AAL systems often incorporate fall detection. We present an activity recognition (AR) and fall detection (FD) system aiming to provide robust real-time performance. It uses two wearable accelerometers, since this is probably the most mature technology for such purpose. For the AR, we developed an architecture that combines rules to recognize postures, which ensures that the behavior of the system is predictable and robust, and classifiers trained with machine learning algorithms, which provide maximum accuracy in the cases that cannot be handled by the rules. For the FD, rules are used that take into account high accelerations associated with falls and the recognized horizontal orientation (e.g., falling is often followed by lying). The system was tested on a dataset containing a wide range of activities, two different types of falls and two events easily mistaken for falls. The Fmeasure of the AR was 99 %, even though it was never tested on the same persons it was trained on. The F-measure of the FD was 78 % due to the difficulty of the events to be recognized and the need for real-time performance, which made it impossible to rely on the recognition of long lying after a fall.

Research paper thumbnail of Detecting Falls with Location Sensors and Accelerometers

Proceedings of the AAAI Conference on Artificial Intelligence

Due to the rapid aging of the population, many technical solutions for the care of the elderly ar... more Due to the rapid aging of the population, many technical solutions for the care of the elderly are being developed, often involving fall detection with accelerometers. We present a novel approach to fall detection with location sensors. In our application, a user wears up to four tags on the body whose locations are detected with radio sensors. This makes it possible to recognize the user’s activity, including falling any lying afterwards, and the context in terms of the location in the apartment. We compared fall detection using location sensors, accelerometers and accelerometers combined with the context. A scenario consisting of events difficult to recognize as falls or non- falls was used for the comparison. The accuracy of the methods that utilized the context was almost 40 percentage points higher compared to the methods without the context. The accuracy of pure location-based methods was around 10 percentage points higher than the accuracy of accelerometers combined with the ...

Research paper thumbnail of Context-aware MAS to support elderly people (demonstration)

Adaptive Agents and Multi-Agents Systems, Jun 4, 2012

This paper presents a context-aware, multiagent system for care of the elderly. The system combin... more This paper presents a context-aware, multiagent system for care of the elderly. The system combines state-of-the-art sensor technologies to detect falls and other health problems, and calls for help in the case of an emergency or issues a warning in cases not needing urgent attention. When deployed at the home of an elderly person it provides them with 24-hour monitoring. Consequently, the elderly may live alone at home, even at an advanced age. The health problems are detected with six groups of agents processing the sensor data and augmenting the data with higher-level information, such as the posture of the person, his/her activity and the context of the situation's environment. The system has been tested in several live demonstrations, where it achieved an excellent performance in complex situations. The system is based on the set of agents observing the elderly person from various points of view, and combining the location and inertial sensors to provide context awareness.

Research paper thumbnail of Using accelerometers to improve position-based activity recognition

This paper will present the results of the research conducted on wireless accelerometers in fall ... more This paper will present the results of the research conducted on wireless accelerometers in fall detection and activity recognition. This research is a part of the Confidence project, whose goal is to provide a health monitoring system for the elderly. Normally, position-based body tags are used to detect postures and activities. This paper reports the results of using accelerometers as both a supplement to and substitute of the position tags. It introduces a combined approach based on machine-learning and wave analysis. Preliminary results indicate an important increase in activity recognition accuracy when using both acceleration and position tags. 3 MOVING FILTER An important step in classifying activities is the ability to detect whether the person is moving. It helps to distinguish between static activities (e.g. lying, sitting) and dynamic ones (e.g. standing up, going down). The moving detection used in our approach uses the data sent by the chest

Research paper thumbnail of Confidence: Ubiquitous Care System to Support Independent Living

The Confidence system aims at helping the elderly stay independent longer by detecting falls and ... more The Confidence system aims at helping the elderly stay independent longer by detecting falls and unusual movement which may indicate a health problem. The system uses location sensors and wearable tags to determine the coordinates of the user's body parts, and an accelerometer to detect fall impact and movement. Machine learning is combined with domain knowledge in the form of rules to recognize the user's activity. The fall detection employs a similar combination of machine learning and domain knowledge. It was tested on five atypical falls and events that can be easily mistaken for a fall. We show in the paper and demo that neither sensor type can correctly recognize all of these events on its own, but the combination of both sensor types yields highly accurate fall detection. In addition, the detection of unusual movement can observe both the user's micro-movement and macro-movement. This makes it possible for the Confidence system to detect most types of threats to t...

Research paper thumbnail of Surveillance Systems An Intelligent Indoor Surveillance System

CEPIS UPGRADE is the anchor point for UPENET (UPGRADE European NETwork), the network of CEPIS mem... more CEPIS UPGRADE is the anchor point for UPENET (UPGRADE European NETwork), the network of CEPIS member societies’ publications, that currently includes the following ones: • inforewiew, magazine from the Serbian CEPIS society JISA • Informatica, journal from the Slovenian CEPIS society SDI • Informatik-Spektrum, journal published by Springer Verlag on behalf of the CEPIS societies GI, Germany, and SI, Switzerland • ITNOW, magazine published by Oxford University Press on behalf of the British CEPIS society BCS • Mondo Digitale, digital journal from the Italian CEPIS society AICA • Novática, journal from the Spanish CEPIS society ATI • OCG Journal, journal from the Austrian CEPIS society OCG • Pliroforiki, journal from the Cyprus CEPIS society CCS • Tölvumál, journal from the Icelandic CEPIS society ISIP

Research paper thumbnail of Evolution of Activity Monitoring Through Various Projects

The paper presents the evolution of activity monitoring utilising wearable sensors through severa... more The paper presents the evolution of activity monitoring utilising wearable sensors through several sequential projects running in the last decade. It covers the change and improvement towards sensor technology that is more affordable and comfortable for the users and development of additional functionalities to achieve accurate activity monitoring in terms of activity recognition. All presented systems are developed to perform on-line activity recognition.

Research paper thumbnail of Context-based fall detection and activity recognition using inertial and location sensors

Journal of Ambient Intelligence and Smart Environments, 2014

Accidental falls are some of the most common sources of injury among the elderly. A fall is parti... more Accidental falls are some of the most common sources of injury among the elderly. A fall is particularly critical when the elderly person is injured and cannot call for help. This problem is addressed by many fall-detection systems, but they often focus on isolated falls under restricted conditions, not paying enough attention to complex, real-life situations. To achieve robust performance in real life, a combination of body-worn inertial and location sensors for fall detection is studied in this paper. A novel context-based method that exploits the information from the both types of sensors is designed. It considers body accelerations, location and elementary activities to detect a fall. The recognition of the activities is of great importance and also is the most demanding of the three, thus it is treated as a separate task. The evaluation is performed on a real-life scenario, including fast falls, slow falls and fall-like situations that are difficult to distinguish from falls. All possible combinations of six inertial and four location sensors are tested. The results show that: (i) context-based reasoning significantly improves the performance; (ii) a combination of two types of sensors in a single physical sensor enclosure is the best practical solution.

Research paper thumbnail of Adapting activity recognition to a person with Multi-Classifier Adaptive Training

Journal of Ambient Intelligence and Smart Environments, 2015

Activity-recognition classifiers, which label an activity based on sensor data, have decreased cl... more Activity-recognition classifiers, which label an activity based on sensor data, have decreased classification accuracy when used in the real world with a particular person. To improve the classifier, a Multi-Classifier Adaptive-Training algorithm (MCAT) is proposed. The MCAT adapts activity recognition classifier to a particular person by using four classifiers to utilise unlabelled data. The general classifier is trained on the labelled data available before deployment and retrieved in the controlled environment. The specific classifier is trained on a limited amount of labelled data belonging to the new person in the new environment. A domain-independent meta-classifier decides whether to classify a new instance with the general or specific classifier. The final, second meta-classifier decides whether to include the new instance into the training set of the general classifier. The general classifier is periodically retrained, gradually adapting to the new person in the new environment. The adaptation results were evaluated for statistical significance. Results showed that the MCAT outperforms competing approaches and significantly increases the initial activity-recognition classifier classification accuracy.

Research paper thumbnail of How Accurately Can Your Wrist Device Recognize Daily Activities and Detect Falls?

Sensors, 2016

Although wearable accelerometers can successfully recognize activities and detect falls, their ad... more Although wearable accelerometers can successfully recognize activities and detect falls, their adoption in real life is low because users do not want to wear additional devices. A possible solution is an accelerometer inside a wrist device/smartwatch. However, wrist placement might perform poorly in terms of accuracy due to frequent random movements of the hand. In this paper we perform a thorough, large-scale evaluation of methods for activity recognition and fall detection on four datasets. On the first two we showed that the left wrist performs better compared to the dominant right one, and also better compared to the elbow and the chest, but worse compared to the ankle, knee and belt. On the third (Opportunity) dataset, our method outperformed the related work, indicating that our feature-preprocessing creates better input data. And finally, on a real-life unlabeled dataset the recognized activities captured the subject's daily rhythm and activities. Our fall-detection method detected all of the fast falls and minimized the false positives, achieving 85% accuracy on the first dataset. Because the other datasets did not contain fall events, only false positives were evaluated, resulting in 9 for the second, 1 for the third and 15 for the real-life dataset (57 days data).

Research paper thumbnail of Fall Detection Using Location Sensors and Accelerometers

IEEE Pervasive Computing, 2015

The rapid aging of the population drives the development of pervasive solutions for the care of t... more The rapid aging of the population drives the development of pervasive solutions for the care of the elderly, which often involve fall detection with accelerometers. These solutions are very accurate in laboratory conditions, but can fail in some real-life situations. To overcome this, we present the Confidence system, which detects falls mainly with location sensors. A user wears one to four tags whose locations are detected with sensors. This allows recognizing the user's activity, including falling and lying afterwards, and the context in terms of the location in the apartment. A scenario consisting of events difficult to recognize as falls or non-falls was used to compare the Confidence system with accelerometer-based fall-detection methods, some of them augmented with the context from a location sensor. The accuracy of the methods that utilized the context was around 30 percentage points higher compared to the methods without the context. The Confidence system was also successfully validated in a real-life setting with elderly users.

Research paper thumbnail of Three-layer Activity Recognition Combining Domain Knowledge and Meta- classification Author list

Journal of Medical and Biological Engineering

One of the essential tasks of healthcare and smart-living systems is to recognize the current act... more One of the essential tasks of healthcare and smart-living systems is to recognize the current activity of a particular user. Such activity recognition (AR) is demanding when only limited sensors are used, such as accelerometers. Given a small number of accelerometers, intelligent AR systems often use simple architectures, either general or specific for their AR. In this paper, a system for AR named TriLAR is presented. TriLAR has an AR-specific architecture consisting of three layers: (i) a bottom layer, where an arbitrary number of AR methods can be used to recognize the current activity; (ii) a middle layer, where the predictions from the bottom-layer methods are inputs for a hierarchical structure that combines domain knowledge and meta-classification; and (iii) a top layer, where a hidden Markov model is used to correct spurious transitions between the recognized activities from the middle layer. The middle layer has a hierarchical, three-level structure. First, a meta-classifie...

Research paper thumbnail of Context-based ensemble method for human energy expenditure estimation

Applied Soft Computing, 2015

Monitoring human energy expenditure (EE) is important in many health and sports applications, sin... more Monitoring human energy expenditure (EE) is important in many health and sports applications, since the energy expenditure directly reflects the intensity of physical activity. The actual energy expenditure is unpractical to measure; therefore, it is often estimated from the physical activity measured with accelerometers and other sensors. Previous studies have demonstrated that using a person's activity as the context in which the EE is estimated, and using multiple sensors, improves the estimation. In this study, we go a step further by proposing a context-based reasoning method that uses multiple contexts provided by multiple sensors. The proposed Multiple Contexts Ensemble (MCE) approach first extracts multiple features from the sensor data. Each feature is used as a context for which multiple regression models are built using the remaining features as training data: for each value of the context feature, a regression model is trained on a subset of the dataset with that value. When evaluating a data sample, the models corresponding to the context (feature) values in the evaluated sample are assembled into an ensemble of regression models that estimates the EE of the user. Experiments showed that the MCE method outperforms (in terms of lower root means squared error and lower mean absolute error): (i) five single-regression approaches (linear and non-linear); (ii) two ensemble approaches: Bagging and Random subspace; (iii) an approach that uses artificial neural networks trained on accelerometer-data only; and (iv) BodyMedia (a state-of-the-art commercial EE-estimation device).

Research paper thumbnail of Discovering Health Problems in the Elderly using Data Mining Approach

This paper is presenting a generalized approach to detection of health problems and falls of the ... more This paper is presenting a generalized approach to detection of health problems and falls of the elderly for the purpose of prolonging autonomous living of elderly using a novel data mining approach. The movement of the user is captured with the motion capture system, which consists of the body-worn tags, whose coordinates are acquired by the sensors situated in the apartment. Output time-series of coordinates are modeled with the proposed data mining approach in order to recognize the specific health problem or fall. The approach is general in a sense that it uses k-nearest neighbor algorithm and dynamic time warping with time-series of all measurable joint angles for the attributes instead of the more specific approach with medically defined attributes. It is two-step approach; in the first step it classifies person's activities into five activities including different types of falls. In the second step it classifies classified walking instances from the first step into five different health states; one healthy and four unhealthy. Even though the new approach is more general and can be used to differentiate also from other types of activities or health problems, it achieves very high classification accuracies, similar to the more specific approaches from the literature.

Research paper thumbnail of Recognizing Human Activities and Detecting Falls in Real-time

The paper presents a system that recognizes human activities and detects falls in real-time. It c... more The paper presents a system that recognizes human activities and detects falls in real-time. It consists of two wearable accelerometers placed on the user's torso and thigh. The system is tuned for robustness and real-time performance by combining domain-specific rules and classifiers trained with machine learning. The offline evaluation of the system's performance was conducted on a dataset containing a wide range of activities and different types of falls. The F-measure of the activity recognition and fall detection were 96% and 78%, respectively. Additionally, the system was evaluated at the EvAAL-2013 activity recognition competition and awarded the first place, achieving the score of 83.6%, which was for 14.2 percentage points better than the secondplace system. The competition's evaluation was performed in a living lab using several criteria: recognition performance, user-acceptance, recognition delay, system installation complexity and interoperability with other systems.

Research paper thumbnail of Multi-Classifier Adaptive Training: Specialising an Activity Recognition Classifier Using Semi-supervised Learning

Lecture Notes in Computer Science, 2012

When an activity recognition classifier is deployed to be used with a particular user, its perfor... more When an activity recognition classifier is deployed to be used with a particular user, its performance can often be improved by adapting it to that user. To improve the classifier, we propose a novel semisupervised Multi-Classifier Adaptive Training algorithm (MCAT) that uses four classifiers. First, the General classifier is trained on the labelled data available before deployment. Second, the Specific classifier is trained on a limited amount of labelled data specific to the new user in the current environment. Third, a domain-independent meta-classifier decides whether to classify a new instance with the General or Specific classifier. Fourth, another meta-classifier decides whether to include the new instance in the training set for the General classifier. The General classifier is periodically retrained, gradually adapting to the new user in the new environment where it is deployed. The results show that our new algorithm outperforms competing approaches and increases the accuracy of the initial activity recognition classifier by 12.66 percentage points on average.

Research paper thumbnail of Recognition of Patterns of Health Problems and Falls in the Elderly Using Data Mining

Lecture Notes in Computer Science, 2012

We present a generalized data mining approach to the detection of health problems and falls in th... more We present a generalized data mining approach to the detection of health problems and falls in the elderly for the purpose of prolonging their autonomous living. The input for the data mining algorithm is the output of the motion-capture system. The approach is general since it uses a k-nearestneighbor algorithm and dynamic time warping with the time series of all the measurable joint angles for the attributes instead of a more specific approach with medically defined attributes. Even though the presented approach is more general and can be used to differentiate other types of activities or health problems, it achieves very high classification accuracies, similar to the more specific approaches described in the literature.

Research paper thumbnail of A Network of Sensor and Actuator Agents for Building Automation Systems

Atlantis Ambient and Pervasive Intelligence, 2013

Due to at least two reasons, energy consumption for heating, ventilation, air conditioning (HVAC)... more Due to at least two reasons, energy consumption for heating, ventilation, air conditioning (HVAC) and domestic hot water (DHW) systems should be reduced. First, the total energy consumption is high since nearly 20% of total energy in USA is accounted for HVAC systems. Sencond reason is exploitation of renevable energy sources, which depend on current time and weather situation. Thus, the energy management should improve to effective by managing currently available energy. For this purpose, the sensor actor agent network is proposed with an example on domestic hot water heating, taking into account several aspects including the occupant presence. The simulation model was used, where the water consumption affected system dynamics. The agent-based schema combined with the simulator enables an occupant to choose the most subjectively desirable policy of the DHW control mechanism.

Research paper thumbnail of Behavior Analysis Based on Coordinates of Body Tags

Lecture Notes in Computer Science, 2009

This paper describes fall detection, activity recognition and the detection of anomalous gait in ... more This paper describes fall detection, activity recognition and the detection of anomalous gait in the Confidence project. The project aims to prolong the independence of the elderly by detecting falls and other types of behavior indicating a health problem. The behavior will be analyzed based on the coordinates of tags worn on the body. The coordinates will be detected with radio sensors. We describe two Confidence modules. The first one classifies the user's activity into one of six classes, including falling. The second one detects walking anomalies, such as limping, dizziness and hemiplegia. The walking analysis can automatically adapt to each person by using only the examples of normal walking of that person. Both modules employ machine learning: the paper focuses on the features they use and the effect of tag placement and sensor noise on the classification accuracy. Four tags were enough for activity recognition accuracy of over 93 % at moderate sensor noise, while six were needed to detect walking anomalies with the accuracy of over 90 %.

Research paper thumbnail of Competitive Live Evaluations of Activity-Recognition Systems

IEEE Pervasive Computing, 2015

In order to ensure the validity and usability of activity recognition approaches, an agreement on... more In order to ensure the validity and usability of activity recognition approaches, an agreement on a set of standard evaluation methods is needed. Due to the diversity of the sensors and other hardware employed, designing and accepting standard tests is a difficult task. This article presents an initiative to evaluate activity recognition systems: a living-lab evaluation established through an annual competition − EvAAL-AR (Evaluating Ambient Assisted Living Systems through Competitive Benchmarking − Activity Recognition). In the competition, each team brings their own activity-recognition system, which is evaluated live on the same activity scenario performed by an actor. The evaluation criteria attempt to capture the practical usability: recognition accuracy, user acceptance, recognition delay, installation complexity, and interoperability with ambient assisted living systems. The article also presents the competing systems with emphasis on two best-performing ones: (i) a system that achieved the best recognition accuracy, and (ii) a system that was evaluated as the best overall. Finally, the article presents lessons learned from the competition and ideas for future development of the competition and of the activity recognition field in general.

Research paper thumbnail of Efficient Activity Recognition and Fall Detection Using Accelerometers

Communications in Computer and Information Science, 2013

Ambient assisted living (AAL) systems need to understand the user's situation, which makes activi... more Ambient assisted living (AAL) systems need to understand the user's situation, which makes activity recognition an important component. Falls are one of the most critical problems of the elderly, so AAL systems often incorporate fall detection. We present an activity recognition (AR) and fall detection (FD) system aiming to provide robust real-time performance. It uses two wearable accelerometers, since this is probably the most mature technology for such purpose. For the AR, we developed an architecture that combines rules to recognize postures, which ensures that the behavior of the system is predictable and robust, and classifiers trained with machine learning algorithms, which provide maximum accuracy in the cases that cannot be handled by the rules. For the FD, rules are used that take into account high accelerations associated with falls and the recognized horizontal orientation (e.g., falling is often followed by lying). The system was tested on a dataset containing a wide range of activities, two different types of falls and two events easily mistaken for falls. The Fmeasure of the AR was 99 %, even though it was never tested on the same persons it was trained on. The F-measure of the FD was 78 % due to the difficulty of the events to be recognized and the need for real-time performance, which made it impossible to rely on the recognition of long lying after a fall.