Thematic axes

Smart cities and the transportation systems of tomorrow rely on the intensive use of sensors and systems with communication capabilities, either ad hoc or via infrastructure. Their functions must meet specific application and environmental constraints. Historically, communication and detection (RADAR) work at LEOST was conducted separately. With the arrival of new standards (beyond 5G...) and new telecommunications strategies, detection and communication functions must work together or even be combined from the design stage. Indeed, since 5G, new communication standards have provided for the introduction of communication between objects and even between intelligences (6G).

Our expertise in this area at LEOST is distinguished by our ability to offer technological solutions from the design phase onwards, taking into account the specificities of the targeted applications, particularly those of intelligent transport systems (rail, road), sensors for cities, and security. This research is carried out as part of European initiatives (V2X projects such as C-Road, InterCor, Secredas, X2Rail, TC-RAIL, EmulRadio4Rail), regional initiatives (DAFI project, CPER 2015-2020 ELSAT2020 and 2021-2027 RITMEA and IMITECH projects) and national initiatives (I-Site FUTURE WESTERN and COMPOTE, DSR-Linked projects).

Translated with DeepL.com (free version)

The major contributions in the Communications and Sensors theme are presented below under two sub-themes: sensors and radio communication.

SENSORS

This sub-theme deals with the design, analysis, and optimization of sensors. The work brings together activities on the design of electromagnetic and acoustic sensors for various applications such as human or system surveillance and road safety.

Over the past five years, our contributions have focused on proposing (i) RadCom detection and communication architectures (LEOST portfolio - road safety); (ii) prototypes based on metamaterials to improve the detection of vulnerable users (LEOST portfolio - road safety); (iii) bio-inspired acoustic and electromagnetic sensors based on the implementation of very low-power neural networks on chips; (iv) vital signs detection radar; and (v) audio flow analysis.

RADIO COMMUNICATION

This sub-theme deals with the design, evaluation, and optimization of radio communications. The constant evolution of consumer communication standards translates into challenges (including coexistence issues) and new application possibilities within our applications (smart transportation and smart city systems). Over the past five years, our work has focused on (i) proposing routing schemes for vehicle ad hoc networks (VANETs); (ii) evaluating the performance of railway communication systems, taking into account radio channel models; (iii) proposing techniques for optimizing the use of spectral resources; (iv) modeling radio channels to improve quality of service in tunnels; (v) methodologies for designing and integrating miniature antennas; (vi) metamaterial reflectors for integrating communication systems in challenging environments (highways, railways). Four major contributions are illustrated below.

The main cyberEM or cybersecurity issues addressed are:

• Analysis of the susceptibility of different communication solutions to interference and attacks;
• Detection and classification of attack situations through monitoring and study of RF activity;
• Analysis of communication protocols at OSI level 2 to determine misuse of these protocols;
• Detection of drones and their interception by intelligent jamming.

The numerous projects that have mobilized researchers on these topics in recent years demonstrate the significant progress made in this area. Since 2012, these include the SECRET (2012-2015), SECOURT (2016-2021), X2Rail (2016-2020), TCRAIL (2018-2019), EmulRadio4Rail (2018-2020), LoRa-R (2020-2023), GLOCAT (2020-2023), SENSORGUARD (2021-2023), and SHADOW (2021-2022). 
As for ongoing projects, there are SLEMBI (2021-2024), DEPOSIA, Resilient Trust, CORTESE, PENTRAIL, and participation in the Railway Safety Chair for the cybersecurity of embedded systems. The Resilient Trust project is a European project involving 26 partners coordinated by members of LEOST's CyberEM theme. 

Finally, EMC activity has been maintained, albeit at a reduced level, particularly as it appears to be essential in a key COSYS topic. In recent years, examples include LUMICAR, in which we studied the susceptibility of light communications to EM interference, and INCIT-EV and CAYD, which deal with EM emissions produced by electric vehicle induction charging systems. 
Similarly, activities related to distributed computing have also focused on cybersecurity at various levels: risk analysis methods (IEC 62443, EBIOS, etc.) and analysis of network protocols to identify vulnerabilities. 

LOCATION TECHNOLOGIES

Advances in intelligent transport systems require the development of a number of technological building blocks, including location technology with improved performance compared to the COTS solutions deployed in our smartphones or for non-safety-critical applications. In particular, solutions must prove their resistance to the local propagation effects encountered by widely used GNSS signals: electromagnetic interference and multipath effects. The international community is therefore working to develop solutions using the GNSS system, hybridized with other sensors (e.g., inertial measurement units), and to achieve the expected performance in terms of accuracy and integrity through fault detection and mitigation solutions. The research conducted at LEOST aims to improve accuracy and integrity performance, particularly by focusing on the detection and mitigation of local propagation effects on GNSS signals (multipath and interference).

The main areas of research developed concern:

• The detection of faults and interference, and the development of GNSS integrity concepts to enable the use of GNSS for critical applications;
• The use of perception sensors, particularly image sensors, to understand the satellite signal reception environment;
• The interpretation of multimedia content for advanced surveillance applications with person segmentation and anomaly detection.

The first two areas are a continuation of our previous work, and our ongoing involvement at European level in projects aimed at introducing GNSS into ERTMS railway signaling systems gives us expertise and European visibility on this subject, often in support of EUSPA. 


The perception module installed on vehicles, whether road or rail, uses a multitude of sensors, including cameras, which can also be used to enhance the positioning module and ensure safer and more accurate location tracking. Our work aims to use vision to perceive, from a zenith camera with a fisheye lens, the environment and therefore the obstacles around the GNSS antenna that interfere with the reception of these signals. This work has been carried out in particular within the eMAPs project (Enhanced Map System - European Union Agency for the Space Program GSA) project and supplemented through international collaborations, notably with the DLR, and internally. Our position at the European level on this subject now leads us to investigate the development of solutions for more complex use cases not yet studied in the literature, such as their operability in poor lighting or weather conditions.

INTERPRETATION OF MULTIMEDIA CONTENT FOR ADVANCED MONITORING

It is essential to provide information on the status of a public transport system, particularly on the deterioration of the safety, security, comfort, and operational conditions of a system component, infrastructure (road, rail), or network before it becomes too advanced and the resilience of the whole is no longer guaranteed. It is therefore necessary to monitor and understand the activity of the system's components and their interactions. For several years, the laboratory has been proposing to use the multimedia flows provided by surveillance systems to detect and identify certain abnormal and dangerous situations. Our research focuses in particular on onboard enclosures (trains/subways), where we propose to develop deep neural models capable of interpreting human activity by jointly analyzing the content of signals from an audio and video surveillance system. We have studied the following two issues: (i) the detection, tracking, and pose estimation of users and (ii) the detection of behavioral anomalies.

To achieve the performance expected by the operator, the algorithms and architectures we offer must prove their effectiveness regardless of the type of cameras used or their position, and must be robust in the face of changes in brightness and lighting, the presence of obstructions, moving backgrounds, or high population density in more or less confined environments.

The end of the previous evaluation period allowed us to switch to the development of neural architectures, which now form the core of our current expertise. We have thus proposed neural architectures that are robust to numerous operating constraints and, in some cases, finely integrate the temporal variation of audio and video signals in order to provide a better representation of passengers and their activities. We have evolved not only within the traditional framework of supervised learning, but have also implemented semi-supervised techniques that are particularly well suited to defining representation spaces in the case of rare instances or events.

Our results have enabled us to address the safety and security challenges of future autonomous transport systems, particularly those outlined in the “Autonomous Train - Passenger Service” program".

CREATION AND PROVISION OF DATABASES

Most of the algorithmic location solutions proposed in the literature offer interesting performance, validated on real data, but these solutions are, on the one hand, rarely compared with each other, and on the other hand, the scenarios proposed are always in ideal conditions for the sensors and without failures. In addition, building databases is a complex task, extremely costly in terms of equipment and time. Consequently, there is an urgent and growing need for shared databases, yet relatively few exist. Those that do exist are very urban-oriented. The work launched at LEOST therefore aims to contribute to this collective effort by the community. With regard to location data, the effort is being carried out as part of the ANR LOCSP project.

This need for real, substantial, labeled databases has become crucial with the advent of artificial intelligence. Unfortunately, the availability of public databases, when it exists, remains marginal in the field of video surveillance. It was therefore with a view to supporting public policy that we created, as part of the EVEREST project, an annotated database of more than ten million images acquired at various resorts in the French ski area, containing potentially dangerous user behavior, for the evaluation of AI-based vision algorithms for the safety of skiers on chairlifts. To facilitate the tedious and time-consuming task of annotating this database, we developed an intuitive and flexible semi-automatic annotation tool that incorporates an automatic face tracking strategy.