Matjaz Gams

Competitive Live Evaluations of Activity-Recognition Systems

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

Training the Genre Classifier for Automatic Classification of Web Pages

This paper presents experiments on classifying web pages by genre. Firstly, a corpus of 1539 manually labeled web pages was prepared. Secondly, 502 genre features were selected based on the literature and the observation of the corpus. Thirdly, these features were extracted from the corpus to obtain a data set. Finally, two machine learning algorithms, one for induction of decision trees (J48) and one ensemble algorithm (bagging), were trained and tested on the data set. The ensemble algorithm achieved on average 17% better precision and 1.6% better accuracy, but slightly worse recall; F-measure did not vary significantly. The results indicate that classification by genre could be a useful addition to search engines.

Telehealth using ECG sensor and accelerometer

The increasing size of the elderly population is driving the development of ambient assisted living systems and telehealth. The recognition of the users everyday activities and detection of alarming situations are important components of such systems. Moreover, the monitoring of vital signs, like the ECG, has a key role in telecare and telemonitoring systems. Therefore, in this paper we propose a system that monitors the user by combining an ECG sensor and two accelerometers. Our system recognizes the users activities and detects falls using the accelerometer data. The ECG data is analyzed in order to extract relevant physiological signals: heart rate, respiration rate, etc. In order to improve the reliability and robustness of the system, the measured accelerometer signals can be combined with the ECG signal in order to detect anomalies in the users behavior and heart-related problems. The proposed proof-of-concept system could contribute significantly to the quality, unobtrusiveness and robustness of the health care and patient safety.

RAReFall — Real-time activity recognition and fall detection system

This demo paper presents the RAReFall system, which is a real-time activity recognition and fall detection 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 users activities and detection of falls is performed on a laptop using the raw sensors data acquired through Bluetooth. The offline evaluation of the systems 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 99% 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 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.

Fall Detection Using Location Sensors and Accelerometers

The rapid aging of the worlds population is driving the development of pervasive solutions for elder care. These solutions, which often involve fall detection with accelerometers, are accurate in laboratory conditions but can fail in some real-life situations. To overcome this, the authors present the Confidence system, which detects falls mainly with location sensors. A user wears one to four tags. By detecting tag locations with sensors, the system can recognize the users activity, such as falling and then lying down afterward, as well as the context in terms of the location in the home. The authors used a scenario consisting of events difficult to recognize as falls or nonfalls to compare the Confidence system with accelerometer-based fall-detection methods, some augmented with context data from a location sensor. The methods that used context information were approximately 30 percent more accurate than those that did not. The Confidence system was also successfully validated in a real-life setting with elderly users.

Using Smartwatch as Telecare and Fall Detection Device

The constant increase of the elderly population and the need to prolong the independent live, is driving the development of telecare systems. Two of the key factors of such systems are their simplicity of usage and minimal obtrusiveness. Therefore, in this paper we propose a practical, intuitive and simple smartwatch-based telecare system. The system has several functionalities, including: automatic fall detection, activity analysis, SOS red button, location information, and reminders. The proposed system can work indoors as well as outdoors, and is completely independent, i.e., it requires only a SIM card to function and does not depend on base stations, internet connection, Bluetooth and similar. The evaluation of the fall detection algorithm showed that it detects all of the fast falls and minimizes the false positives, achieving 85% accuracy.

Surrogate-agent modeling for improved training

Computer-aided practice can help improve personnel training for demanding scenarios in terms of time and quality. In this paper, we concentrate on asymmetrical conflicts, such as a unit that deals with hostile crowds robbing a store, with the aim of preventing further criminal activity and at the same time minimizing physical and emotional damage. We propose a surrogate-agent modeling approach based on execution of the following loop: (i) observe a human (the unit leader) playing a set of scenarios in a simulated environment and induce strategic patterns of human play; (ii) use patterns to construct a surrogate agent (digital clone); (iii) test the surrogate under all possible circumstances through data farming; and (iv) evaluate the performance and highlight deficiencies in the agents responses, thereby enabling human improvements in new attempts. Experiments on two domains indicate that the proposed approach could significantly improve the training procedure and help trainees to properly perceive the cognitive properties of the crowds.

Intelligent System to Assist the Independent Living of the Elderly

Information technologies can assist the elderly in their daily lives and thus improve their quality of life. Here, we present a system that was developed within the framework of the EU H2020 project IN LIFE to make the elderly feel more secure. The system consists of a smartwatch, a tablet, and a web portal for carers. The smartwatch has several functionalities; the most important are automatic fall detection, SOS button, communication with carers, location detection in case of emergency, and general activity monitoring. The smartwatch communicates with the server via GSM, therefore it does not depend on wireless internet or Bluetooth connection. The tablet acts as a virtual doorman and also as an interface for the carer, while the portal acts as a central information hub for notifications, scheduling of tasks, and user overview. We present the initial feedback from the users in a pilot study and discuss some modifications that we carried out considering this feedback.

Chronic Heart Failure Detection from Heart Sounds Using a Stack of Machine-Learning Classifiers

Chronic heart failure represents a global pandemic, currently affecting over 26 million of patients worldwide. It is a major contributor in the death rate of patients with cardiovascular diseases and results in more than 1 million hospitalizations annually in Europe and North America. Methods for chronic heart failure detection can be utilized to act preventive, improve early diagnosis and avoid hospitalizations or even life-threatening situations, thus highly enhance the quality of patient’s life. In this paper, we present a machine-learning method for chronic heart failure detection from heart sounds. The method consists of: filtering, segmentation, feature extraction and machine learning. The method was tested with a leave-one-subject-out evaluation technique on data from 122 subjects, gathered in the study. The method achieved 96% accuracy, outperforming a majority classifier for 15 percentage points. More specifically, it detects (recalls) 87% of the chronic heart failure subjects with a precision of 87%. The study confirmed that advanced machine learning applied on real-life sounds recorded with an unobtrusive digital stethoscope can be used for chronic heart failure detection.