Service Recommendations with Deep Learning: A Study on Neural Collaborative Engines, Pasquale De Rosa, Michel Deriaz, Marco De Marco, Luigi Laura, Pacific Asia Journal of the Association for Information Systems Vol. 14 No. 2, pp. 59-70, 2022.
Abstract. Background: The present paper aims to investigate the adoption of Neural Networks for recommendation systems and to propose Deep Learning architectures as advanced frameworks for designing Collaborative Filtering engines. Recommendation systems are data-driven infrastructures which are widely adopted to create effective and cutting-edge smart services, allowing to personalize the value proposition and adapt it to changes and variations in customers' preferences. Method: Our research represents an exploratory investigation on the adoption of Neural Networks for Recommendation Systems, inspired by the findings of a recent study on service science that highlighted the suitability of those models for designing cutting-edge recommenders capable of overcoming stable traditional benchmarks like the Singular Value Decomposition and the k-Nearest Neighbors algorithms. Following this study, we designed a more "complex" Feed-Forward Neural Network, trained on the "Movielens 100K" dataset using the Mean-Squared Error function to approximate the model loss generated and the Adaptive Moment Estimation algorithm (Adam) for the parameters optimization. Results: The results of this study demonstrate the primary role of Feed-Forward Neural Networks for designing advanced Collaborative recommenders, consolidating and even improving the outcomes of the work that inspired our research. Conclusion: Given these assumptions, we confirm the suitability of Feed-Forward Neural Networks as effective recommendation algorithms, laying the foundations for further studies in neural-based recommendation science.
Scalable Incremental Similarity-based Learning with Neural Filters for Large Scale Timeseries Classification Systems, Nils Schaetti, Pasquale De Rosa, Montavon Stéphane, David Deillon, Michel Deriaz, in the 6th International Conference on Smart and Sustainable Technologies (SplitTech 2021), Split and Bol, Croatia, 2021.
Abstract. Automatic health monitoring and activity recognition systems provide specific information for caregivers and health professionals to prevent injury or disease. With the improvement of sensor technologies, wireless communication and machine learning, systems can now be aware of changes in the user’s state and its environment in order to provide activityaware automatic health predictions and information. However, these systems face the challenge to recognize specific patterns and activities in a large and quickly evolving database of sensor data. In this paper, an innovative approach to classify timeseries in a large dynamical space of classes is evaluated. The algorithm is based on the Conceptor Framework (CF) and close to the Reservoir Computing paradigm in machine learning. Compared to traditional approaches for classification, Conceptors allow the recognition of a large number of dynamical patterns incrementally without interference with previous learning and can easily adapt to new classes and users on the fly. For this empirical study, we gathered a dataset of timeseries extracted from sensors placed on horses during training sessions and evaluated various methods on a task of predicting which horse is the source of a given series. The dataset was gathered as a part of the HorseTrack project which aims to create a system able to automatically detect injuries in horses. Our results show that incremental similaritybased learning with Conceptors is able to solve effectively this task with short training time while classical ESNs are not able to perform better than the majority baseline.
Artificial intelligence and GPS sensor technology for 3D analyses in the biomechanics of jumping horses, Stéphane Montavon, David Deillon, Johnathan Bertolaccini, Michel Deriaz, in the Swiss Review of Military and Disaster Medicine (SRMDM), Virtual, 2021.
Abstract. Pathologies of the locomotor system in the horse are strongly linked to competition and daily work sessions. In the show jumping horse, repetitive loads and unsuitable surface conditions too often cause serious and potentially career-ending injuries. In this context, the authors wanted to demonstrate the use of a non-invasive sensor integrated in the girth to characterize the important criteria that define good biomechanics in a jumping horse (Alogo Move Pro Sensor). Based on AI (artificial intelligence) and GPS (global positioning system) sensor technology, this device accurately measures parameters that may help in identifying crucial aspects of jumps relevant either for performance or injury prevention. In a preliminary study, it was possible to qualitatively and quantitatively analyse the three critical phases of a jump sequence, namely approach, jumping parabola and move off. Different parameters were measured for each of these phases. Thanks to state-of-the-art technology used in aeronautical guidance, the sensor allows the display of unique data such as the real trajectory in 3D. A set of analysis algorithms was developed and used. Two objectives, which could be verified in practice, have been formulated. These results were similar and in line with previously published data by other authors.
Service Recommendations with Deep Learning: a Study on Neural Collaborative Engines, Pasquale De Rosa, Michel Deriaz, Marco De Marco, Luigi Laura, in the sixth version of ICTO international conference (ICTO2020), Virtual, 2020.
Abstract. The present paper aims to investigate the adoption of Neural Networks for recommendation systems and to propose Deep Learning architectures as advanced frameworks for designing Collaborative Filtering engines. Recommendation systems are data-driven infrastructures which are widely adopted to create effective and cutting-edge smart services, allowing to personalize the value proposition and adapt it to changes and variations in customers' preferences. For this purpose we will introduce a Collaborative Filtering algorithm based on the adoption of a "deep" Feed-Forward Network, inspired by a recent research on neural-based service recommenders; given these assumptions, we will confirm the suitability of Feed-Forward Neural Networks as effective recommendation algorithms, laying the foundations for further studies in neural-based recommendation science.
A Queue Management Approach for Social Distancing and Contact Tracing, Athanasios I. Kyritsis, Michel Deriaz, in the Third IEEE International Conference on Artificial Intelligence for Industries (ai4i 2020), Virtual, 2020.
Abstract. Social distancing is necessary to prevent the rapid spread of a highly contagious disease, such as COVID-19, at least until a vaccine is found and mass-produced. By reducing the probability of an uninfected person coming close or in physical contact with an infected one, the disease transmission in the community can be suppressed. Although social distancing is simple to comprehend, it is not always easy to implement, mainly because not all public spaces are designed with this requirement in mind. In this paper, we present a queue management tool that can be used to allow people that wait for a service practice social distancing. In our approach, people are asked to join a virtual queue, in order to avoid crowds in physical waiting rooms or long waiting queues. Machine learning is used to predict the estimated waiting time of queuers, so they are called just in time to get served. We use past data and machine learning to predict how busy a location will be so that customers can pick the best time to visit the service. Finally, we present the method we use to monitor people taking a service at any time and implement contact tracing in a privacy-preserving manner.
A Machine Learning Approach to Waiting Time Prediction in Queueing Scenarios, Athanasios I. Kyritsis, Michel Deriaz, in the Second IEEE International Conference on Artificial Intelligence for Industries (ai4i 2019), Laguna Hills, California, 2019.
Abstract. Physically queueing is a reality on many industries that provide services or sell goods. Waiting in a queue can be stressful and exhausting for the clients because of the enforced idle time, and may lead to decreased customer satisfaction. Queueing theory has been widely used to assess client waiting times, to optimize staff schedules, and to increase the robustness of a queueing system against a variable demand for service. In this paper, we are exploring how multiple industries that require queues can benefit from machine learning to predict the clients’ waiting times. We begin by predicting waiting times on bank queues, and then we propose how the procedure can be generalized to more industries and automatized. A publicly available dataset containing entries of people queueing in banks is initially utilized, and after training a fully connected neural network, a mean absolute error of 3.35 minutes in predicting client waiting times was achieved. We are then presenting a web application that is managing queues of different scenarios and industries. The queues may have unique parameters, and the system can adapt to each queue as it creates a per queue optimally trained neural network for waiting time prediction. The use and the capabilities of the system are validated with the use of a simulator. Machine learning, therefore, proves to be a viable alternative to queueing theory for predicting waiting time.
Gait Pattern Recognition Using a Smartwatch Assisting Postoperative Physiotherapy, Athanasios I. Kyritsis, Geoffrey Willems, Michel Deriaz, Dimitri Konstantas, International Journal of Semantic Computing, Vol. 13, No. 2, p. 245–257, 2019.
Abstract. Postoperative rehabilitation is led by physiotherapists and is a vital program that reestablishes joint motion and strengthens the muscles around the joint after an orthopedic surgery. Modern smart devices have affected every aspect of human life. Newly developed technologies have disrupted the way various industries operate, including the healthcare one. Extensive research has been carried out on how smartphone inertial sensors can be used for activity recognition. However, there are very few studies on systems that monitor patients and detect different gait patterns in order to assist the work of physiotherapists during the said rehabilitation phase, even outside the time-limited physiotherapy sessions. In this paper, we are presenting a gait recognition system that was developed to detect different gait patterns. The proposed system was trained, tested and validated with data of people who have undergone lower body orthopedic surgery, recorded by Hirslanden Clinique La Colline, an orthopedic clinic in Geneva, Switzerland. Nine different gait classes were labeled by professional physiotherapists. After extracting both time and frequency domain features from the time series data, several machine learning models were tested including a fully connected neural network. Raw time series data were also fed into a convolutional neural network.
Gait Recognition with Smart Devices Assisting Postoperative Rehabilitation in a Clinical Setting, Athanasios I. Kyritsis, Geoffrey Willems, Michel Deriaz, Dimitri Konstantas, in the First IEEE International Conference on Artificial Intelligence for Industries (ai4i 2018), Laguna Hills, California, 2018.
Abstract. Postoperative rehabilitation is a vital program that re-establishes joint motion and strengthens the muscles around the joint after an orthopedic surgery. This kind of rehabilitation is led by physiotherapists who assess each situation and prescribe appropriate exercises. Modern smart devices have affected every aspect of human life. Newly developed technologies have disrupted the way various industries operate, including the healthcare one. Extensive research has been carried out on how smartphone inertial sensors can be used for activity recognition. However, there are very few studies on systems that monitor patients and detect different gait patterns in order to assist the work of physiotherapists during the said rehabilitation phase, even outside the time-limited physiotherapy sessions, and therefore literature on this topic is still in its infancy. In this paper, we are presenting a gait recognition system that was developed to detect different gait patterns including walking with crutches with various levels of weight-bearing, walking with different frames, limping and walking normally. The proposed system was trained, tested and validated with data of people who have undergone lower body orthopedic surgery, recorded by Hirslanden Clinique La Colline, an orthopedic clinic in Geneva, Switzerland. A gait detection accuracy of 94.9% was achieved among nine different gait classes, as these were labeled by professional physiotherapists.
Considerations for the Design of an Activity Recognition System Using Inertial Sensors, Athanasios I. Kyritsis, Michel Deriaz, Dimitri Konstantas, in the 20th IEEE International Conference on E-health Networking, Application & Services (IEEE HealthCom 2018), Ostrava, Czech Republic, 2018.
Abstract. The last decade there has been an increasing research interest in the field of human activity recognition in the frame of designing context-aware applications. There is a plethora of parameters that affect the performance of an activity recognition system. However, designers of such systems often either ignore some factors or even neglect their importance. In this paper, we present and discuss in detail research challenges in human activity recognition using inertial sensors, and we analyse the significance of the existent parameters during the design and the evaluation of such systems. We exemplify the role of the aforementioned parameters with an experiment that was conducted, in which 11 people performed 5 different activities. Data were recorded from the inertial sensors of a wrist-worn smartwatch. We illustrate how various parameters of the system can be configured and demonstrate how they impact the whole system’s performance. This work aims to be used as a concise reference for future endeavours in the field of human activity recognition using inertial sensors of mobile devices in general, and of wrist-worn smartwatches in particular.
Anomaly Detection Techniques in Mobile App Usage Data among Older Adults, Athanasios I. Kyritsis, Michel Deriaz, Dimitri Konstantas, in the 20th IEEE International Conference on E-health Networking, Application & Services (IEEE HealthCom 2018), Ostrava, Czech Republic, 2018.
Abstract. We are living in an era of demographic ageing, and new technologies that support independent living are constantly being created. In this context, more and more mobile applications are developed for this target group. In this paper, we are presenting a multidimensional application that targets older adults. We are monitoring the usage of all different aspects of the app, the amount of daily activity in the form of daily steps and the resting time throughout the day from a connected bracelet the user is wearing. Data amounting to 402 user-days of 6 different users are collected. A set of different datasets are manufactured, and various anomaly detection techniques are employed to identify the abnormalities in the datasets. The results demonstrate that clustering can be of use to detect anomalies in the older adults’ patterns that could be the trigger of appropriate actions, like informing family members or professional caregivers.
User Requirement Analysis for the Design of a Gamified Ambient Assisted Living Application, Athanasios I. Kyritsis, Julia Nuss, Lynnette Holding, Peter Rogers, Michael O'Connor, Panagiotis Kostopoulos, Mervyn Suffield, Michel Deriaz, Dimitri Konstantas, in the 16th International Conference on Computers Helping People with Special Needs (ICCHP 2018), Linz, Austria, 2018.
Abstract. Most countries of the world are heading towards an ageing society. At the same time, newer technologies are constantly created, while the advances in networks and wireless communications allow other technologies like mobile and cloud computing to become ubiquitous. This leads to a problem that we are identifying and confronting, to make the use of modern technology easier for older adults, since it is in principle more easily perceivable by younger people. This paper presents a questionnaire study that took place during the design of a gamified mobile application that targets older people. In total 133 older adults answered the questionnaire consisting of 41 questions, providing an insightful view of their attitude towards modern technology, their health, physical activity tracking, playing games and social interaction using technology. The results provide useful insights to researchers and developers who target this age group for their human-centric applications and services.
Wandering Behaviors Detection for Dementia Patients: a Survey, Abbass Hammoud, Michel Deriaz, Dimitri Konstantas, in The 3rd International Conference on Smart and Sustainable Technologies (SPLITECH 2018), Split, Croatia, 2018.
Abstract. Dementia is an age-associated impairment that could affect about 135 million people worldwide by the year 2050. People with dementia suffer from memory and orientation problems, which cause them to wander and get lost. Advances in technologies and connectivity can be leveraged to reduce the risk of unsafe wandering. In this paper, we present a survey of state-of-the-art technologies and methodologies, which are used for tracking and detection of wandering behaviors. The survey provides a compilation of the most related works in the literature and commercial fields, discusses their aspects and limitations, with the aim to benefit future efforts in this domain. We found that several approaches exist to tackle the problem of wandering, where most of the reviewed works tend to focus on the technical side, rather than adopting a user-centric design. We also observe that the commercial systems are lagging behind the research efforts, which can have a great impact if wisely applied in real world applications. Finally, we review the related sides of security, privacy and ethical concerns around the development of tracking systems, and present general recommendations for developing systems that respect these sides.
Adaptive power switching technique for ultrasonic motion sensors, Abbass Hammoud, Michel Deriaz, Dimitri Konstantas, in Journal of Ambient Intelligence and Humanized Computing, 2018.
Abstract. Smart sensing technologies play a key role in the core of smart systems, which form the rapidly evolving internet of things. In this context, buildings’ occupancy information is an important input that allows smart systems to be seamlessly aware of and responsive to the inhabitants, thus ensuring their comfort. Ultrasonic motion sensors are used to obtain occupancy information of indoor spaces. Although they provide a high accuracy as compared to other sensors, like Passive InfraRed, they require a higher power consumption. In this work, we propose an adaptive power switching technique, which we callpower hopping. This technique allows ultrasound motion sensors to optimize their transmitter power level, in order to best fit their surrounding environment. The objective is to reduce the overall energy consumption of these sensors. We have tested our method using a sensor prototype, and the results show that, depending on the sensor’s environment, a possible saving in the transmitter power can be achieved, which reached up to 78% in our experiments. We also derive an upper bound limit of the method’s convergence time, and we propose an automatic sensing method to detect potential changes in the sensor’s environment.
Enhance daily live and health of elderly people, Panagiotis Kostopoulos, Athanasios I. Kyritsis, Vincent Ricard, Michel Deriaz, Dimitri Konstantas, in The 8th International Symposium on Frontiers in Ambient and Mobile Systems (FAMS 2018), Porto, Portugal, 2018.
Abstract. As people get older, they tend to become more and more vulnerable to physical disabilities and mental illnesses. In order to prevent the deterioration of their quality of life we have created a system that helps elderly to sustain and extend their activities of daily living (ADL). Older people, especially those who may have just left the working environment, can suffer a sense of loss, particularly of value, purpose, confidence. This can lead to mood swings, isolation and possibly depression. The EDLAH2 (Enhance Daily Live And Health) project tries to combat these negative experiences of elderly people and give the opportunity for a fuller lifestyle. These older adults have a determination to live in their homes and enjoy living in their homes for as long as they can. The idea of the system presented in this paper is to bring an increased level of motivation, interest and engagement into areas that may be important but mundane. This results in a greater drive to be involved in this area of action and a positive feeling when rewards are achieved. EDLAH2 enables the continuity of motivation for elderly people. We set goals and achievements in line with realistic expectations for the older adults and importantly, we provide a guide to their well being improvement. Finally we utilize gamification in order to reinforce the elderly people to stay active and improve their well-being.
Enhanced Still Presence Sensing with Supervised Learning over Segmented Ultrasonic Reflections, Abbass Hammoud, Athanasios I. Kyritsis, Michel Deriaz and Dimitri Konstantas, in The Eighth International Conference on Indoor Positioning and Indoor Navigation (IPIN 2017), Sapporo, Japan, 2017.
Abstract. Sensing the presence of people in indoor spaces allows smart systems to be aware of and responsive to the occupants, and paves the way for a wide range of applications. In this paper, we show how the reflection patterns of ultrasonic signals can be leveraged to detect the presence of still persons. We propose the use of supervised learning over segmented reflection patterns, and prove that this method is capable of detecting minute variations in the environment's response. The experimental evaluation of the proposed method in an office and a residential environment shows that it achieves a high presence sensing accuracy in the case of low signal-to-noise ratio (SNR), and a perfect accuracy in the case of high SNR, even in the case of non line-of-sight. Among the different tested classifiers, we found that the linear Support Vector Machine (SVM) achieves the best performance, yielding a presence detection accuracy of 84.3%-98.4% for low SNR, and 100% for high SNR, in the tested environments.
A Multiobjective Optimization Methodology of Tuning Indoor Positioning Systems, Grigorios G. Anagnostopoulos, Michel Deriaz and Dimitri Konstantas, in The Eighth International Conference on Indoor Positioning and Indoor Navigation (IPIN 2017), Sapporo, Japan, 2017.
Abstract. How can the collected data from testing an indoor positioning deployment be transformed into information concerning the optimal tuning of a positioning system in this deployment? How can such kind of accumulated information from several deployments be transformed into more generic knowledge regarding the system’s performance, with respect to several performance goals? In this work, we present a multiobjective optimization methodology of tuning indoor positioning systems, based on real data recorded onsite. Selecting the appropriate tuning for a positioning system is a challenging task, which depends on many factors: the specific deployment, the devices used, the evaluation metrics and their order of significance, the user-case scenarios tested, etc. In order to handle these multiplicities, we introduce the use of multiobjective optimization which allows several objectives to be simultaneously satisfied. We exemplify the methodology performing tests with the GpmStudio platform, a desktop tuning and evaluation platform that supports our Global Positioning Module (GPM). The methodology proves to be a very useful tool in the hands of testers who are designated to optimally tune the positioning system in a variety of scenarios.
Navigational needs and requirements of hospital staff: Geneva University Hospitals case study, Grigorios G. Anagnostopoulos, Michel Deriaz, Jean-Michel Gaspoz, Dimitri Konstantas and Idris Guessous, in The Eighth International Conference on Indoor Positioning and Indoor Navigation (IPIN 2017), Sapporo, Japan, 2017.
Abstract. Navigating in large hospitals is a challenging task. The consequences of difficulties faced by staff, patients and visitors in finding their way in the hospital can be multiple. The HUGApp project goals are to identify the navigational needs and requirements of people within the premises of Geneva University Hospitals (HUG) before proceeding with potential solutions, such as an indoor navigation mobile app. A questionnaire was designed and distributed to staff members with the goal of understanding the current problems in wayfinding inside HUG, investigating the users’ views on the creation of an indoor navigation mobile app, and specifying the user requirements for such an app. A total of 111 members of the primary care division of HUG answered the questionnaire, providing an insightful view of the healthcare professionals.
Power Hopping: an Automatic Power Optimization Method for Ultrasonic Motion Sensors, Abbass Hammoud, Grigorios G. Anagnostopoulos, Athanasios I. Kyritsis, Michel Deriaz, Dimitri Konstantas, in The Fourteenth International Conference on Ubiquitous Intelligence and Computing (UIC-2017), San Francisco, USA, 2017.
Abstract. Ultrasonic motion sensors are used to obtain occupancy information of indoor spaces. Although they provide a high accuracy as compared to other sensors, like Passive InfraRed (PIR), they require a higher power consumption in general. In this paper we propose power hopping, an automatic power optimization method that allows ultrasound motion sensors to optimize their transmitter power level. The objective is to reduce the overall energy consumption of these sensors. We have tested our method using a sensor prototype, and the results show that, depending on the sensor's environment, a possible saving in the transmitter power can be achieved, which can be as high as 78%. We also derive an upper bound limit of the method's convergence time.
UltraSense: a Self-Calibrating Ultrasound-Based Room Occupancy Sensing System, Abbass Hammoud, Michel Deriaz, Dimitri Konstantas, in The Eighth International Conference on Ambient Systems, Networks and Technologies (ANT-2017), Madeira, Portugal, 2017.
Abstract. Smart sensing technologies play a key role in the core of smart systems, which form the rapidly evolving internet of things. In this context, buildings' occupancy information is an important input that allows smart systems to be seamlessly aware of and responsive to the inhabitants, thus ensuring their comfort. In this paper we present UltraSense, an ultrasound-based room occupancy sensing system that relies on unsupervised learning, to automatically calibrate its parameters according to the room's environment. This ability avoids the need for manual calibration of the sensing system for each new environment. While commonly available occupancy detection technologies are limited to line-of-sight (LOS) conditions, UltraSense also operates in non line-of-sight (NLOS) scenarios. The proposed system was implemented and tested in order to characterize its performance. UltraSense was developed for the European research project SmartHeat in the frame of ambient assisted living.
Robust Ultrasound-Based Room-Level Localization System Using COTS Components, Abbass Hammoud, Michel Deriaz and Dimitri Konstantas, in proceedings of The Fourth IEEE International Conference on Ubiquitous Positioning, Indoor Navigation and Location-Based Services (UPINLBS 2016), Shanghai, China, 2016.
Abstract. Location-based services have become very popular in recent years. Although many previous works targeted the problem of indoor localization, several reasons still prevent the widespread adoption of most systems' implementations. Some of these reasons are their insufficient availability, and the need for extensive node deployment and maintenance. In this work, we present a highly accurate room-level indoor localization system that is based on ultrasound technology. Our system is robust to noise, scalable, has a low complexity on the receiver, and does not require synchronization between transmitter and receiver. Moreover, it uses commercial off-the-shelf (COTS) components and does not require special hardware or additional infrastructure to be deployed. The system relies solely on ultrasound, and does not use any RF signals. To deal with the problem of signal interference, we explain how signal collisions can be detected, and we propose a method for collision avoidance. The system was implemented and tested in scenarios with realistic conditions. The results prove that the proposed system is accurate and robust to ambient noise. This work was conducted in the frame of the European project 'SmartHeat', which aims to improve heating conditions of elderly people, and reduce energy consumption. It employs rooms' occupancy information with other inputs, to adapt the heating according to users' habits.
Practical Evaluation and Tuning Methodology for Indoor Positioning Systems, Grigorios G. Anagnostopoulos, Carlos Martínez de la Osa, Tiago Nunes, Abbass Hammoud, Michel Deriaz and Dimitri Konstantas, in proceedings of The Fourth IEEE International Conference on Ubiquitous Positioning, Indoor Navigation and Location-Based Services (UPINLBS 2016), Shanghai, China, 2016.
Abstract. How do we evaluate the performance of an indoor positioning system? In addition, in which way can the system be optimally tuned for a certain environment? These are the questions addressed in this study. We propose a practical, cost efficient methodology for evaluating and tuning indoor positioning systems. The methodology has two main phases. In the first online recording phase, the ground truth information is gathered, and raw signals are recorded. In the second phase, offline positioning algorithms utilise the recorded information to infer position estimations which can then be precisely evaluated. An automatic parameter optimization methodology, which recommends optimal tunings for the positioning algorithm, is presented as a key utility of this work. An overall advantage of the proposed method is the fact that the recorded data guarantee the repeatability of tests, and allow consistent comparisons among different algorithms, creating the perspective of a testbed based on real data. The implementation of the methodology is exemplified with the presentation of the GpmLab Android application and the GpmStudio desktop platform, tools which assist our main positioning framework, the Global Positioning Module (GPM).
Online Self-Calibration of the Propagation Model for Indoor Positioning Ranging Methods, Grigorios G. Anagnostopoulos, Michel Deriaz and Dimitri Konstantas, in proceedings of The Seventh International Conference On Indoor Positioning and Indoor Navigation (IPIN 2016), Madrid, Spain, October 2016.
Abstract. A common problem for indoor positioning methods is the fact that the differences in the reception characteristics among devices may significantly deteriorate the performance of a positioning system. Ranging algorithms for positioning rely on the accuracy of the parameters of the propagation model. This model is used to infer an estimate of the distance of a mobile device from each access point from the Received Signal Strength Indication (RSSI). In this work we present an algorithm which dynamically recalculates and improves the propagation model. The improvement of the model parameters fits the environment’s characteristics and, more importantly, the reception characteristics of the device used. The proposed algorithm is tested with different devices at an indoor deployment covering a large area where Bluetooth Low Energy (BLE) technology is used. The experimental results show that the proposed method offers a significant accuracy improvement to some devices while it slightly improves the performance of those that are more properly tuned.
Positioning Evaluation and Ground Truth Definition for Real Life Use Cases, Carlos Martínez de la Osa, Grigorios G. Anagnostopoulos, Mauricio Togneri, Michel Deriaz and Dimitri Konstantas, in proceedings of The Seventh International Conference On Indoor Positioning and Indoor Navigation (IPIN 2016), Madrid, Spain, October 2016.
Abstract. Evaluating positioning systems has become a matter of heated debate during the last years. There is no clear standard on how these technologies should be evaluated, and no predominant solution for defining the ground truth in order to compare the position estimates. In this paper, we propose a simple and inexpensive solution for tackling both of these problems in real life use cases. With the proposed methodology, it is possible to measure both static and moving targets, by creating a predefined path with checkpoints. Then, a tester, walking over them, while moving or standing still, indicates when the device was over the aforementioned checkpoints. It is also specified how to evaluate the estimates by comparing them with interpolated points of the ground truth trajectory. Two methods are proposed for performing such interpolation. Finally, in order to evaluate the performance of the positioning system as well as the perceived utility of the position estimates from the end user’s point of view, a series of statistical parameters is discussed. Additionally, in the context of perceived utility by the end user, a parameter that measures the occurrence of abrupt changes in the position estimates is proposed.
F2D: A location aware fall detection system tested with real data from daily life of elderly people, Panagiotis Kostopoulos, Athanasios I. Kyritsis, Michel Deriaz and Dimitri Konstantas, in proceedings of The Sixth International Conference on Current and Future Trends of Information and Communication Technologies in Healthcare (ICTH 2016), London, United Kingdom, September 2016.
Abstract. Falls among older people remain a very important public healthcare issue. In the majority of fall events external support is imperative in order to avoid major consequences. Therefore, the ability to automatically detect these fall events could help reducing the response time and significantly improve the prognosis of fall victims. This paper presents a practical real time fall detection system running on a smartwatch (F2D). A decision module takes into account the rebound after the fall and the residual movement of the user, matching a detected fall pattern to an actual fall. The last module of F2D is the location module which makes our system very useful for nursing homes that host elderly people. The fall detection algorithm has been tested by Fondation Suisse pour les Téléthèses (FST), the project partner who is responsible for the commercialization of our system. By testing with real data and achieving an accuracy of 96.01% we have a fall detection system ready to be deployed on the market and by adding the location module we can provide it to nursing homes for elderly people.
A BLE-Based Probabilistic Room-Level Localization Method, Athanasios I. Kyritsis, Panagiotis Kostopoulos, Michel Deriaz and Dimitri Konstantas, in proceedings of The Sixth International Conference On Localization and GNSS (ICL-GNSS 2016), Barcelona, Spain, June 2016.
Abstract. During the last decades, location based services have become very popular and the developed indoor positioning systems have achieved an impressive accuracy. The problem though is that even if the only requirement is room-level localization, those systems are most of the times not cost-efficient and not easy to set-up, since they often require time-consuming calibration procedures. This paper presents a low-cost, threshold-based approach and introduces an algorithm that takes into account both the Received Signal Strength Indication (RSSI) of the Bluetooth Low Energy (BLE) beacons and the geometry of the rooms the beacons are placed in. Performance evaluation was done via measurements in an office environment composed of three rooms and in a house environment composed of six rooms. The experimental results show an improved accuracy in room detection when using the proposed algorithm, compared to when only considering the RSSI readings. This method was developed to provide context awareness to the international research project named SmartHeat. The projects aims to provide a system that efficiently heats a house, room by room, based on the habitants’ habits and preferences.
Stress detection using smart phone data, Panagiotis Kostopoulos, Athanasios Kyritsis, Michel Deriaz and Dimitri Konstantas, in proceedings of The EAI International Conference on Wearables in Healthcare (EAI 2016), Budapest, Hungary, June 2016.
Abstract. In today's society, working environments are becoming more stressful. The problem of occupational stress is generally recognized as one of the major factors leading to a wide spectrum of health problems. However work should, ideally, be a source of health, pride and happiness, in the sense of enhancing motivation and strengthening personal development. In this work, we present StayActive, a system which aims to detect stress and burn-out risks by analyzing the behaviour of the users via their smartphone. The main purpose of StayActive is the use of the mobile sensor technology for detecting stress. Then a mobile service can recommend and present various relaxation activities "just in time" in order to allow users to carry out and solve everyday tasks and problems at work. In particular, we collect data from people's daily phone usage gathering information about the sleeping pattern, the social interaction and the physical activity of the user. We assign a weight factor to each of these three dimensions of wellbeing according to the user's personal perception and build a stress detection system. We evaluate our system in a real world environment with young adults and people working in the transportation company of Geneva. This paper highlights the architecture and model of this innovative stress detection system. The main innovation of this work is addressed in the fact that the way the stress level is computed is as less invasive as possible for the users.
StayActive: An Application for Detecting Stress, Panagiotis Kostopoulos, Tiago Nunes, Kevin Salvi, Mauricio Togneri and Michel Deriaz, in proceedings of The Fourth International Conference on Communications, Computation, Networks and Technologies (INNOV 2015), Barcelona, Spain, November 2015.
Abstract. In today’s society, working environments are becoming more stressful and people working in these environments become prone to various illnesses. But, work should be a source of health, pride and happiness, in the sense of enhancing motivation and strengthening personal development. In this work, we present StayActive, a system which aims to detect stress and burnout risks by analyzing the behaviour of the users via their smartphone. In particular, we collect data from people’s daily phone usage gathering information about the sleeping pattern, the social interaction and the physical activity of the user. We assign a weight factor to each of these three dimensions of wellbeing according to the user’s personal perception and build a stress detection system. We evaluate our system in a real world environment and in a daily-routine scenario. This paper highlights the architecture and model of this innovative stress detection system.
Smart Position Selection in Mobile Localisation, Carlos Martinez, Grigorios G. Anagnostopoulos, Michel Deriaz, in proceedings of The Fourth International Conference on Communications, Computation, Networks and Technologies (INNOV 2015), Barcelona, Spain, November 2015.
Abstract. Which technology should be used in order to be able to locate oneself in any kind of scenario? This has been a recurrent question in the last years. It has become evident that, until now, there is no dominant indoor positioning solution based on a single technology. Outdoors, positioning systems based on satellites have given excellent results. However, a global solution for both kinds of scenarios does not exist. In our study, this problem is dealt with by creating an algorithm able to evaluate positions received from different technologies and choose the most trustworthy one. As a result, we are able to improve the overall accuracy of the user’s position estimation, compared to the ones the different technologies would have given if used independently. In this way, the user is offered a simple solution to have an accurate position in all environments, in a transparent way. The main challenge of using different technologies at the same time is usually the battery consumption. A solution for dealing with this aspect is also proposed in this document. This research has been done in the context of the Ambient Assisted Living (AAL) Enhanced Daily Living and Health (EDLAH) project, where older people can track their lost objects, which requires them to be positioned in a very accurate way.
F2D: A fall detection system tested with real data from daily life of elderly people, Panagiotis Kostopoulos, Tiago Nunes, Kevin Salvi, Michel Deriaz and Julien Torrent, in Proceedings of the seventeenth International Conference on E-health Networking, Application & Services (IEEE HealthCom'15), Boston, USA, October 2015.
Abstract. Falls among older people remain a very important public healthcare issue. Every year over 11 million falls are registered in the U.S. alone. This paper presents a practical real time fall detection system running on a smartwatch (F2D). A decision module takes into account the rebound after the fall and the residual movement of the user, matching a detected fall pattern to an actual fall. The final decision of a fall event is taken based on the location of the user. To the best of our knowledge, this is the first fall detection system which works on an independent smartwatch, being less stigmatizing for the end user. The fall detection algorithm has been tested by Fondation Suisse pour les Téléthèses (FST), the project partner who is responsible for the commercialization of our system. By analyzing real data of activities of daily life of elderly people, we are confident that F2D meets the demands of a reliable and easily extensible system. This paper highlights the innovative algorithm which takes into account the residual movement and the location of the user to increase the fall detection accuracy. By testing with real data we have a fall detection system ready to be deployed on the market.
Automatic Switching Between Indoor and Outdoor Position Providers, Grigorios G. Anagnostopoulos, Michel Deriaz, in proceedings of the Sixth International Conference on Indoor Positioning And Indoor Navigation (IPIN 2015), Banff, Alberta, Canada, October 2015.
Abstract. In which way may an application switch instantly and reliably between an indoor and an outdoor positioning provider as a user enters and exits buildings? In this work we present a robust switching algorithm, utilizing the dynamic accuracy estimation of each position provider as a reliability indication. Our algorithm offers a fast automatic switch between the indoor and the outdoor provider, in a transparent way for the user. We also present experimental results, using GPS outdoors and a Bluetooth provider indoors. This technique was tested in our lab and was afterwards installed at the Hospital of Perugia, Italy, in the context of the Ambient Assisted Living (AAL) Virgilius project, where users can navigate with a smartphone. This study is also a result of the research done in the context of the AAL EDLAH project, for optimizing the selection of the most adequate positioning technology. Accurate position estimations are used as input for the EDLAH object detection module.
Increased Fall Detection Accuracy in an Accelerometer-Based Algorithm Considering Residual Movement, Panagiotis Kostopoulos, Tiago Nunes, Kevin Salvi, Michel Deriaz and Julien Torrent, in Proceedings of the fourth International Conference on Pattern Recognition Applications and Methods (ICPRAM), Lisbon, Portugal, January 2015.
Abstract. Every year over 11 million falls are registered. Falls play a critical role in the deterioration of the health of the elderly and the subsequent need of care. This paper presents a fall detection system running on a smartwatch (F2D). Data from the accelerometer is collected, passing through an adaptive threshold-based algorithm which detects patterns corresponding to a fall. A decision module takes into account the residual movement of the user, matching a detected fall pattern to an actual fall. Unlike traditional systems which require a base station and an alarm central, F2D works completely independently. To the best of our knowledge, this is the first fall detection system which works on a smartwatch, being less stigmatizing for the end user. The fall detection algorithm has been tested by Fondation Suisse pour les Téléthèses (FST), the project partner for the commercialization of our system. Taking advantage of their experience with the end users, we are confident that F2D meets the demands of a reliable and easily extensible system. This paper highlights the innovative algorithm which takes into account residual movement to increase the fall detection accuracy and summarizes the architecture and the implementation of the fall detection system.
Accuracy Enhancements in Indoor Localization with the Weighted Average Technique, Grigorios G. Anagnostopoulos and Michel Deriaz, in Proceedings of the Eighth International Conference on Sensor Technologies and Applications (SENSORCOMM), Lisbon, Portugal, November 2014.
Abstract. Indoor localization is a topic that has drawn great attention over the last decade. One of the main goals of the research in the field is to improve the achieved accuracy. Along with the accuracy, factors like the easiness of deployment and reconfiguration, the cost, the computational complexity, and the ability to tune the desired accuracy in specific areas are also important. In this study, we used Bluetooth Low Energy (BLE) technology, that offers a low cost and is easily deployed and reconfigured. The weighted average method, combined with the selection of the closest beacons and the averaging of the received signal strength indication (RSSI) at the distance domain proposed in this paper, offers an accuracy down to 0.97 meters, depending on the deployment configuration. This method was tested in our lab and was following installed at the hospital in Perugia, Italy, in the context of the Ambient Assisted Living (AAL) Virgilius project, where users can navigate with a smartphone.
Improving Distance Estimation in Object Localisation with Bluetooth Low Energy, Georgia Ionescu, Carlos Martínez de la Osa and Michel Deriaz, in Proceedings of the Eighth International Conference on Sensor Technologies and Applications (SENSORCOMM), Lisbon, Portugal, November 2014.
Abstract. The arrival of Bluetooth Low Energy (BLE) creates opportunities for great innovations. One possible application is object localisation. We present our unique software that can track objects and help finding their location within a house perimeter. With the help of Bluetooth beacons that can be attached to different items, we can estimate the distance between the mobile device and the object with an accuracy of less than one meter. In this paper, we describe our system and the techniques we use, the experiments we conducted along with the results. In addition, we briefly present some work in progress using an indoor positioning system that helps locating the objects.
User Behaviour Recognition for Interacting with an Artistic Mobile Application, Jody Hausmann, Kevin Salvi, Jérôme Van Zaen, Adrian Hindle and Michel Deriaz, in Proceedings of the 4th International Conference on Multimedia Computing and Systems (ICMCS), Marrakesh, Morocco, April 2014.
Abstract. Interacting with smartphones generally requires direct input from the user. We investigated a novel way based on the user's behaviour to interact directly with a phone. In this paper, we present MoveYourStory, a mobile application that generates a movie composed of small video clips selected according to the user's position and his current behaviour when the user is moving. Towards this end, we have implemented an activity recognition module that is able to recognise current activities, like walking, bicycling or travelling in a vehicle using the accelerometer and the GPS embedded in a smartphone. Moreover, we added different walking intensity levels to the recognition algorithm, as well as the possibility of using the application in any position. A user study was done to validate our algorithm. Overall, we achieved 96.7% recognition accuracy for walking activities and 87.5% for the bicycling activity.
Gesture Recognition for Interest Detection in Mobile Applications, Jérôme Van Zaen, Jody Hausmann, Kevin Salvi and Michel Deriaz, in Proceedings of the 4th International Conference on Multimedia Computing and Systems (ICMCS), Marrakesh, Morocco, April 2014.
Abstract. Gestures are a fast and efficient mean to transmit information. They are used in a large number of situations where speaking is not as effective or even not possible, such as to indicate precisely a point of interest or to warn about a danger in a noisy environment. Furthermore, gestures can also be used for intuitive human-computer interfaces where specific tasks would otherwise require navigating through graphical interface menus. Consequently, solutions to provide reliable and accurate gesture recognition have been investigated extensively in the past years. In this paper, we propose a gesture recognition system to detect user interest with a sensor-embedded mobile phone. Specifically, this system uses hidden Markov models to recognize pointing gestures. Once such a gesture has been recognized, it is straightforward to identify the point of interest based on the user location and the phone orientation. In a subject-independent scenario, we obtained a recognition accuracy above 91% with the accelerometer when discriminating between pointing gestures and similar gestures that are common with a mobile phone (e.g. looking at the screen). When using the gyroscope in addition to the accelerometer, the accuracy raised above 98%.
On-board navigation system for smartphones, Mauricio Togneri and Michel Deriaz, in Proceedings of the 4th International Conference on Indoor Positioning and Indoor Navigation (IPIN), Montbéliard, France, October 2013.
Abstract. Several mobile solutions offer the possibility to download maps to use them offline at any moment. However, most of the time a connection to an external server is still needed in order to calculate routes and navigate. This represents an issue when traveling abroad due to roaming costs. In this paper, we propose a solution to this problem through an engine that stores and manages OpenStreetMap's data to consult points of interest, calculate routes and navigate without any connection required. The software manages indoor and outdoor information to provide a full navigation service that works in both environments. Therefore, the same system allows navigating in a highway by car and provides indoor navigation for museums, hospitals and airports among others. The result is an on-board engine for smartphones that provides indoor and outdoor navigation services that does not require Internet connection.
GeoGuild: Location-Based Framework for Mobile Games, Georgia Ionescu, Javier Martín de Valmaseda and Michel Deriaz, in Proceedings of the 3rd International Conference on Social Computing and Its Applications, Karlsruhe, Germany, October 2013.
Abstract. Smartphones' performance is in sustained growth, and their geolocation capabilities bring opportunities for developers to explore new aspects of social gaming. We propose a multiplatform framework that can give a social side to a wide range of mobile games. The framework is independent of the game and provides different location-based tools, such as guilds, territories management and geolocated events. Our set of tools integrates user geo-position on real world maps, allowing the players to team up and conquer territories in their vicinity. The uniqueness of our framework is the fact that it can turn a simple game into a full social experience, whilst combining the real and the virtual world. In this paper we provide an overview of our platform and show how well-established one-on-one games can be rethought as social multiplayer games.
TrustPos Model: Trusting in Mobile Users' Location, Javier Martín de Valmaseda, Georgia Ionescu and Michel Deriaz, in Proceedings of the 10th International Conference on Mobile Web Information Systems, Paphos, Cyprus, August 2013.
Abstract. While social games based on geo-location are gaining popularity, determining the authenticity of the players' geo-position becomes a challenge, since there are ways to counterfeit it, quite accessible to everyone. We propose a solution based on global spatial and temporal observation of the players' interactions. In this paper we present TrustPos, a trust engine model that associates a trustworthiness factor to each player based on the context of the interactions with both the game and other players. The novelty of TrustPos is the fact that our model is based on an internal network of players linked through their interactions, as opposed to previous approaches that are strongly specialized to concrete domains as peer-to-peer networks and social recommenders, not adaptable to location trust concerns.
Harmonization of Position Providers, Anja Bekkelien, Michel Deriaz, in Proceedings of the 3rd International Conference on Indoor Positioning and Indoor Navigation (IPIN), University of New South Wales, Sydney, Australia, November 2012.
Abstract. Hybrid positioning systems have been proposed in order to overcome the limitations of individual location sensing technologies. However, the large differences between the various technologies make integration into a larger system a challenge. This paper proposes a harmonization model which provides different location information sources with a uniform interface. The model creates an abstract representation based on performance criteria and our aim is to provide a basis for the design of location based services. The benefit of our approach is an extensible system that allows for seamless incorporation of new technologies. In addition, it offers a standard format for geographical positions, facilitating higher level treatment of information. To illustrate the usability of the model we implemented a prototype, the Global Positioning Module, which combines several commonly used technologies.
Hybrid Positioning Framework for Mobile Devices, Anja Bekkelien, Michel Deriaz, in Proceedings of the 2nd International Conference on Ubiquitous Positioning, Indoor Navigation, and Location Based Service (UPINLBS), Helsinki, Finland, October 2012.
Abstract. A variety of technologies has emerged in response to an increasing demand for location aware applications. Nevertheless, single-technology systems have several limitations and vulnerabilities and it seems unlikely that such a system will be able to provide a universal solution. In this paper, we present the Global Positioning Module (GPM), a framework that seamlessly combines a multitude of approaches in order to supply mobile devices with indoor and outdoor positioning. The novelty of our work is the way in which position providers are integrated by using an abstraction derived from their performance properties. This allows for a selection of providers based on their suitability to the surrounding environment and to the user's requirements with regards to accuracy, drift, power consumption and so on. Our aim is to provide a foundation for ubiquitous location based services, namely a transparent transition between the plethora of technologies available today.
PhDs done in the team
2019-12-19. Athanasios Kyritsis. Enhancing Wellbeing Using Artificial Intelligence Techniques.
2018-11-23. Abbass Hammoud. Indoor Occupancy Sensing with Ultrasounds.
2017-11-23. Grigorios Anagnostopoulos. Addressing Crucial Issues of Indoor Positioning Systems.
2017-10-19. Panagiotis Kostopoulos. From fall detection to stress pattern using smart devices.