Research

Research areas: Machine Learning, Ambient Intelligence, Artificial Intelligence, Wearable Computing, Ambient Assisted Living.

My main research interests include analysis of human behavior and health using primarily machine learning techniques applied on wearable sensors data. I have experience with various sensing technologies: inertial sensors, ultra-wideband location system, heart-rate, body temperature, GSR, and other physiological sensors. I have used these sensors data in order to extract useful information about the user wearing the sensors: recognize activity, detect a fall and estimation of human energy expenditure. Finally, I have some experience with programming sensors platforms and thus, creating a custom firmware adapted to the user’s needs. In particular, I created a custom firmware for the Shimmer sensor platform in TinyOS/Nesc programming language. Later, hundreds of sensors were equipped with the firmware, and were used by patients in a real-life scenario for the Artemis project – CHIRON.

The text below explains some of my research interests by providing more detailed overview and also links to relevant literature.

 

 –The RAReFall system – winner of the EvAAL ’13 Activity Recognition competition

We developed the “Real-time Activity Recognition and Fall Detection” system called “RAReFall” and won the annual competition in Activity Recognition – EvAAL ’13.

About the system: It is tuned for robustness and real-time performance by combining human-understandable rules and classifiers trained with machine learning algorithms. The system consists of two wearable accelerometers sewn into elastic sports-wear, placed on the abdomen and the right thigh. The recognition of the user’s activities and detection of falls is performed on a laptop using the raw sensors’ data acquired through Bluetooth. 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 second-place system. The evaluation was performed in a living lab using several criteria: recognition performance, user-acceptance, recognition delay, system installation complexity and interoperability with other systems.

A presentation about the RAReFall system was given at the Solomon Seminar and is available on the video lectures portal .


RAReFall activity recognition and fall detection system at the EvAAL competition
, Hristijan Gjoreski, Simon Kozina

Relevant papers:
S. Kozina, H. Gjoreski, M. Gams, M. Luštrek. Efficient Activity Recognition and Fall Detection Using Accelerometers. Evaluating AAL Systems Through Competitive Benchmarking Communications in Computer and Information Science, Volume 386, 2013, pp 13-23
[pdf | link ]

H. Gjoreski, M. Gams, M. LuštrekContext-based fall detection and activity recognition using inertial and location sensors. Journal of Ambient Intelligence and Smart Environments (JAISE), Accepted for publication. 2014.

S. Kozina, H. Gjoreski, M. Gams, M. Luštrek. Three-layer activity recognition combining domain knowledge and meta-classification . Journal of medical and biological engineering, 2013.
[pdf | link ]

 

 –Human Energy Expenditure Estimation

Monitoring human energy expenditure is important in many health and sport applications, since the energy expenditure directly reflects the level of physical activity. The actual energy expenditure is unpractical to measure; hence, the field aims at estimating it by measuring the physical activity with accelerometers and other sensors. Current advanced estimators use a context-dependent approach in which a different regression model is invoked for different activities of the user. In our research approach, we go a step further and use multiple contexts corresponding to multiple sensors, resulting in an ensemble of models for energy expenditure estimation. This provides a multi-view perspective, which leads to a better estimation of the energy. The proposed method was experimentally evaluated on a comprehensive set of activities where it outperformed the current state-of-the-art.

Relevant papers:
H. Gjoreski, B. Kaluža, M. Gams, R. Milić, and M. Luštrek. Ensembles of Multiple Sensors for Human Energy Expenditure Estimation. UbiComp 2013.
[pdf | link ]

–Human Fall Detection

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 foster robust performance in real life, a combination of body-worn inertial and location sensors for fall detection is studied. 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.

Relevant papers:
H. Gjoreski, M. Luštrek,and M. Gams. Context-Based Fall Detection using Inertial and Location Sensors. International Joint Conference on Ambient Intelligence (AmI-12), November, 2012.
[pdf | link ]

H. Gjoreski, M. Luštrek, M. Gams. Accelerometer Placement for Posture Recognition and Fall Detection. The 7th International Conference on Intelligent Environments, IE, 2011.
[pdf]

H. Gjoreski,  Adaptive Human Activity Recognition and Fall Detection using Wearable Sensors, Master Thesis.

 –Human Activity Recognition

To be used in a real-world setting, healthcare and smart living systems must understand the user’s situation and context, making activity recognition an essential component of such systems. In this study a novel approach (TriLAR) for activity recognition is presented. The TriLAR has a three-layer structure: (i) low layer, where an arbitrary number of activity recognition methods can be used to recognize the current activity; (ii) medium layer, where the predictions from low layer methods are aggregated; and (iii) high layer, where hidden Markov model is used to correct spurious transitions of the activities from medium layer. This architecture was tested on a dataset recorded by ten volunteers performing a complex 90-minute scenario while wearing accelerometers placed on the chest, thigh and the ankle. The complex scenario included ten atomic activities. The performance of the TriLAR architecture was compared to three other traditional single-layer activity recognition methods: Classification Trees, Naïve Bayes and Support Vector Machine. The TriLAR architecture outperformed all of the single-layer methods, in all sensor body placement combinations. Chest and ankle emerged as the best performing sensor combination, achieving 94.65% accuracy; which was for 3.72 percentage points better than the best performing single-layer method, SVM.  And finally, the results showed that by using advanced layered architecture, such as TriLAR, it is possible to successfully recognize atomic activities using minimum number of sensors, i.e. one or two. The whole sensor system and the TriLAR architecture were adjusted to be used in a real-world setting, coping with the limitations of the sensor battery-life and communication protocols.

Relevant papers:
S. Kozina, H. GjoreskiM. Gams, M. Luštrek. Three-layer activity recognition combining domain knowledge and meta-classification . Journal of medical and biological engineering, 2013.

H. Gjoreski, M. Luštrek, M. Gams. Accelerometer Placement for Posture Recognition and Fall Detection. The 7th International Conference on Intelligent Environments, IE, 2011.
[pdf]

S. Kozina, H. Gjoreski, M. Gams, M. Luštrek. Efficient Activity Recognition and Fall Detection Using Accelerometers. Evaluating AAL Systems Through Competitive Benchmarking Communications in Computer and Information Science, Volume 386, 2013, pp 13-23
[pdf | link ]

H. Gjoreski,  Adaptive Human Activity Recognition and Fall Detection using Wearable Sensors“, Master Thesis