Research Rafael Berkvens

Abstract

Extended Reality (XR) has become the killer application for future wireless networks. XR is expected to be a major source of traffic for 6G. Use cases include education, health care, and gaming. XR requires accurate and real-time pose information (gesture recognition) to enable seamless experience. Recently, researchers have investigated the use of sub-6 GHz Wi-Fi signals for pose estimation and the results are very promising. Radio waves at these frequencies offer limited resolution due to low bandwidth. On the other hand, mmWave frequencies (>30 GHz) not only offer high data-rates but also high spatial resolution. The improved spatial resolution can benefit pose estimation in XR applications where accurate motion tracking is important for immersive and realistic experience. In this project, my aim is to leverage mmWave signals for pose estimation. This idea of using communications signals for sensing is known as Integrated Sensing and Communication (ISAC). To do this, I aim to collect an extensive and realistic dataset of gestures across several people and environments. I will then develop novel signal processing and deep learning-based algorithms for environment independent and multi-user sensing. Moreover, I will investigate a low power solution for pose estimation on the edge devices. Finally, I will develop a real-time prototype, which projects real-world movements onto the virtual world avatar of the user.

Researcher(s)

• Promoter: Famaey Jeroen
• Co-promoter: Berkvens Rafael
• Fellow: Bhat Nabeel Nisar

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

The radio frequency part of the spectrum is filled with various energy sources that are transmitted, reflected, refracted, diffracted, absorbed and scattered by objects and persons in the real world. The energy is transmitted for other reasons than interpreting the environment: it is transmitting information, data that people or machines are sharing. However, like a lighthouse that signals the location of the shore and simultaneously intermittently lights up that shore, radio frequency transmissions carry both data and information about the environment. In the most generic sense, our objective is to be able to look at the radio frequency spectrum like our eyes look at the visual light spectrum and 'see' what is happening. The major research hypothesis of this proposal is that devices and persons can be counted, identified and tracked between rooms by studying changes in received yet unknown radio signals. If we can prove this hypothesis, the academic breakthrough will lead to a novel class of research focusing on interpreting the real world using the below visual light ambient electromagnetic spectrum.

Researcher(s)

• Promoter: Berkvens Rafael
• Co-promoter: Weyn Maarten

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

General goal Optimizing the AI training, network installation and forecasting aspects of an RF-based counting system for crowds using both external data and data from the same RF-based counting system.

Researcher(s)

• Promoter: Berkvens Rafael
• Promoter: Hellinckx Peter
• Fellow: Janssens Robin

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

In places where crowds gather, it is especially important for event organisers to be able to make an accurate estimate of the number of people present. In order to invest in a particular method, a fair comparison of the counting methods is necessary.  Our earlier research strongly pointed to the need for calibration of different counting methods and a pooling and exploitation of knowledge and expertise among these organisers in order to help them professionalise and enable (further) growth. The project will result in a better understanding of technologies for visitor counting through a decision tree based on a fair comparison of (1) the number of attendees at a given time and (2) the number of unique visitors during the event which also provides guidelines for extrapolating the counts. The decision tree will result in, among other things, more accurate predictions, impact analyses, deployment of resources and a better choice of visitor counts based on accuracy.  Results Context and objective Mapping visitor numbers at events has become more important than ever since the corona crisis. Having a clear view of how many visitors are present at a venue is the basis of crowd management. However, measuring crowds is challenging. Organisers, security personnel, security forces and other stakeholders often talk about varying visitor numbers at the same event. Technological counting methods also contradict each other. The need for calibrations for different counting methods is high. This project systematically checked the accuracy of different counting methods. More specifically, this project investigated the employability and accuracy of manual click and quadrant counts, as well as that of four technological counting methods commonly used at events: camera counting, Wi-Fi counting, mobile data counting and radio wave counting. Test events Due to the corona crisis, the events sector went on lockdown for a long time and events could not take place at various times during the course of project. When the sector was allowed to restart, it was first in the form of test events that required government approval. In the next phase, events could go ahead subject to compliance with a limited maximum capacity. Since the summer, the deployment of a Covid Safe Ticket (CST) ensured that events could once again proceed in as normal a manner as possible at full capacity. The research team, together with the various counting method providers, chose to pool their knowledge and expertise and deploy them to ensure a safe restart of the events sector. For this reason, we conducted measurements at test events, events with limited capacity as well as events that used a CST. Moreover, different types of events participated as test cases in this project, which also resulted in a lot of variation in terms of content. In this way, the difficult situation the events sector was in gave an extra dimension to this project and (often in consultation with the National Crisis Centre) we were able to support the events sector in difficult and uncertain times. Counting Guide The results of the research were compiled in a handy tool available on the website www.telwijzer.be. You can use the Counting Guide to determine the most appropriate counting method(s) for your event.

Researcher(s)

• Promoter: Berkvens Rafael
• Co-promoter: Weyn Maarten

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

In order to understand biological processes such as migration, dispersal and disease transmission, we need to know where animals are moving and who they are meeting. While this has been achieved for a lot of larger animals, the vast majority of animals are too small to effectively monitor without compromising on data accuracy or acquisition rates. This has implications not only for research into animal movement and behaviour, but also for applied applications such as better welfare for captive animals and livestock, and environmental monitoring. The recent advances in the Internet of Things (IoT) which has revolutionized various aspects of daily life have enormous potential in the field of wildlife tracking, but as yet have been little exploited, particularly when considering miniaturized options. We developed ProxLogs, an integrated, flexible and accessible monitoring system for small animals, based around recent improvements to Bluetooth Low Energy protocols. This project aims to develop the Minimum Viable Product, test it in operational environments, and investigate the appropriate business model of the system. This will be a state-of-the-art system which will allow the monitoring of far smaller wild and domestic animals at a greatly improved spatiotemporal scale than has previously been achieved, all while ensuring the system remains low cost and accessible for end users through our use of the widely available Bluetooth protocols. In this project we will further validate the prototype and investigate different potential business models.

Researcher(s)

• Promoter: Weyn Maarten
• Co-promoter: Berkvens Rafael
• Co-promoter: Kirkpatrick Lucinda
• Co-promoter: Leirs Herwig

Research team(s)

• Internet Data Lab (IDLab)

Project website

Project type(s)

• Research Project

Abstract

Wireless sensor networks (WSNs) are a key enabler for Internet of Things (IoT) applications in many different industries such as logistics, healthcare, agriculture, smart homes and smart vehicles. The location of one or more of the devices (nodes) in the network is interesting if not crucial information for the application. For this reason, algorithms for the self-localization of wireless sensor networks have been developed which are capable of automatically determining the position of the static nodes relative to each other. However, these methods are not equipped to effectively define a distance estimate in complex environments. Moreover, path loss exponents should be adapted to each link between the nodes in particular for better results. We hypothesize that we can model both the location and signal attenuation of WSN nodes so that the individual communication link distances can be ascribed to node mobility or environmental changes by incorporating context information such as environment layout or likely node locations.

Researcher(s)

• Promoter: Berkvens Rafael

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

Large music festivals such as Studio Brussels' 'De Warmste Week' and Tomorrowland have a need for privacy conscious systems and algorithms that measure and analyse the density and flow of crowds for safety and security purposes. Such systems and algorithms for automated crowd status (density, flow, activity) assessments can alleviate the currently difficult task presented to trained staff and police forces. Our research group was the first to attempt to estimate large numbers of people through passive, radio frequency device free localization, at a venue supporting at least seven thousand attendees at the Tomorrowland music festival. Our preliminary data correlates with the rough density estimates currently available to event organisers, mainly coming from expert opinion based on a limited amount of cameras. My research will focus on three crucial aspects, which I will elaborate upon in this proposal: first, accurately model the relationship between crowd density and radio frequency signal features; second, estimate crowd flows in the environment; and finally, investigate if it is possible to relate crowd activity to the radio frequency signal features. By using radio signal features, the proposed system respects the privacy of individuals by design, making it truly non-intrusive.

Researcher(s)

• Promoter: Weyn Maarten
• Co-promoter: Berkvens Rafael
• Fellow: Kaya Abdil

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

Automatic crowd density estimation can be highly useful for a multitude of applications, examples of which are traffic control, gauging interest at a trade fair and crowd control systems during large-scale events. Classic camera-based setups have several shortcomings, the most notorious of which is the potential for privacyrelated issues to occur. The use of a passive crowd estimator which makes use of an RF-based wireless sensor network (WSN) could provide a solution to this problem. A series of experiments which we performed by installing WSNs in large-scale music festival environments containing thousands of individuals indicated that the influence of the crowd on radio frequency communication within these networks can be used to obtain accurate crowd size estimates. In this project, we seek to validate this core principle for different types and sizes of environments. Furthermore, we wish to investigate how the environment type is related to the network size and the amount of training that is required to obtain accurate results. Finally, an in-depth analysis regarding the crowd density within subregions of these environments and the potential for this approach to allow for crowd flows to be determined, will be investigated as well. Furthermore, to commercialize this proof-of-concept, a go-to-market strategy will be further finetuned. This includes the identification of the different application sectors and to address the different benefits for customers.

Researcher(s)

• Promoter: Weyn Maarten
• Co-promoter: Berkvens Rafael

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

Availability of water resources is under stress due to climatic changes, visible in the recent period of prolonged drought in Belgium, but also elsewhere. However, each year more than 60 mio m3drinking water is lost in Flanders. Leak localization today is very time consuming and labor intensive as operators have to manually place equipment that 'listens' to the water flow during the night. SmartWaterGrid will substantiate, facilitate and automate leak localization to respond more quickly to detected leaks by using innovative modelling to significantly reduce the number of costly sensors needed. To do so, hybrid digital twins of real-time flow and pressure measurements will be augmented with GIS data, physical models, and human feedback from customers and experts. This way, leak localization can be brought from +/- 70km, and weeks to months to exactly localize a leak, up to a soft real-time solution of less than 1 km (street level). Additionally, optimal operating parameters of the wireless sensor network will be determined to minimize battery energy consumption while feeding sufficient data into the digital twin. The type of field service order and required field service skills will also be determined to effectively resolve problems for end customers.

Researcher(s)

• Promoter: Berkvens Rafael
• Co-promoter: Weyn Maarten

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

In recent years, Low Power Wide Area Networks (LPWAN) have received much attention, due to the rise of the Internet of Things (IoT) and the need to localize devices in these long-range networks, using minimal power consumption. Asset tracking is one of the classic applications of LPWAN localization. However, the more accurate a localization algorithm, the more application potential (e.g. home automation, health care solutions and smart cities) there is to use this algorithm. Therefore, we need advanced technologies and algorithms to improve the accuracy and reliability of LPWAN localization. Although feature-based localization is widely used in indoor environments, we will extend the use of this methodology to outdoor environments. Features are defined as signal characteristics, such as signal strength. The class of feature-based localization can be subdivided into different subclasses. Fingerprinting and ranging are two of the most important techniques in the featurebased class. In this research, we will investigate new and existing algorithms to increase the accuracy and reliability of feature-based localization techniques in LPWAN. A comparative study between the accuracy and reliability of LPWAN technologies (Sigfox, LoRaWAN and NB-IoT) will be made as well.

Researcher(s)

• Promoter: Weyn Maarten
• Co-promoter: Berkvens Rafael
• Fellow: Janssen Thomas

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

Nowadays, predictive maintenance, remote monitoring of machinery, real-time communication between employees, etc. is considered essential for efficient operation and management of industrial environments. However, metal-containing ship & harbor environments such as ship compartments, remote control rooms of factories and individual stacked containers, still lack internet coverage, often due to the presence of large blocking metal const ruct ions. This challenge is tackled by the Internet of Shipping (IoS) project, which uses multi-hop sub-GHz wireless mesh network extensions together with locally available wired and wireless technology (e.g., satellite, Wi-Fi, PLC, Ethernet), to provide connectivity and positioning services in challenging shipping environments. The IoS connectivity services will be optimized for (i) safety operations (health monitoring, possibility to trigger alarms), (ii) track and tracing of goods, equipment and employees, (iii) communication between employees (voice calls from remote compartments & locations), and (iv) remote monitoring of machinery operations.

Researcher(s)

• Promoter: Famaey Jeroen
• Co-promoter: Berkvens Rafael

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project

Abstract

In this project we develop and validate an IoT pre-commercial product in the area of industry 4.0 for the (petro)chemical sector. The project focusses on algorithm optimization, power consumption optimization, communication energy budget classification, validation and demonstration in an operational industrial environment and potential patent application.

Researcher(s)

• Promoter: Weyn Maarten
• Co-promoter: Berkvens Rafael

Research team(s)

• Internet Data Lab (IDLab)

Project type(s)

• Research Project