Thales (Canada)

The DeepGlobe Building Extraction Challenge poses the problem of localizing all building polygons in the given satellite images. We can create polygons using an existing instance segmentation algorithm based on Mask R-CNN. However, polygons produced from instance segmentation have irregular shapes, which are far different from real building footprint boundaries and therefore cannot be directly applied to many cartographic and engineering applications. Hence, we present a method combining Mask R-CNN with building boundary regularization. Through the experiments, we find that the proposed method and Mask R-CNN achieve almost equivalent performance in terms of accuracy and completeness. However, compared to Mask R-CNN, our method produces better regularized polygons which are beneficial in many applications.

The current hype of Artificial Intelligence (AI) mostly refers to the success of machine learning and its sub-domain of deep learning. However, AI is also about other areas, such as Knowledge Representation and Reasoning, or Distributed AI, i.e., areas that need to be combined to reach the level of intelligence initially envisioned in the 1950s. Explainable AI (XAI) now refers to the core backup for industry to apply AI in products at scale, particularly for industries operating with critical systems. This paper reviews XAI not only from a Machine Learning perspective, but also from the other AI research areas, such as AI Planning or Constraint Satisfaction and Search. We expose the XAI challenges of AI fields, their existing approaches, limitations and opportunities for Knowledge Graphs and their underlying technologies.

Visual Question Answering (VQA) is a challenging task that has received increasing attention from both the computer vision and the natural language processing communities. Current works in VQA focus on questions which are answerable by direct analysis of the question and image alone. We present a concept-aware algorithm, ConceptBert, for questions which require common sense, or basic factual knowledge from external structured content. Given an image and a question in natural language, ConceptBert requires visual elements of the image and a Knowledge Graph (KG) to infer the correct answer. We introduce a multi-modal representation which learns a joint Concept-Vision-Language embedding. We exploit ConceptNet KG for encoding the common sense knowledge and evaluate our methodology on the Outside Knowledge-VQA (OK-VQA) and VQA datasets. Our code is available at

Large cities’ expanding populations are causing traffic congestion. The maintenance of the city’s road network necessitates ongoing monitoring, growth, and modernization. An intelligent vehicle detection solution is necessary to address road traffic concerns with the advancement of automatic cars. The identification and tracking vehicles on roads and highways are part of intelligent traffic monitoring while driving. In this paper, we have presented how You Only Look Once (YOLO) v5 model may be used to identify cars, traffic lights, and pedestrians in various weather situations, allowing for real-time identification in a typical vehicular environment. In an ordinary or autonomous environment, object detection may be affected by bad weather conditions. Bad weather may make driving dangerous in various ways, whether due to freezing roadways or the illusion of low fog. In this study, we used YOLOv5 model to recognize objects from street-level recordings for rainy and regular weather scenarios on 11 distinct classes of vehicles (car, truck, bike), pedestrians, and traffic signals (red, green, yellow). We utilized freely available Roboflow datasets to train the proposed system. Furthermore, we used real video sequences of road traffic to evaluate the proposed system’s performance. The study results revealed that the suggested approach could recognize cars, trucks, and other roadside items in various circumstances with acceptable results.

In the last five years, the state-of-the-art in computer vision has improved greatly thanks to an increased use of deep convolutional neural networks (CNNs), advances in graphical processing unit (GPU) acceleration and the availability of large labelled datasets such as ImageNet. Obtaining datasets as comprehensively labelled as ImageNet for ship classification remains a challenge. As a result, we experiment with pre-trained CNNs based on the Inception and ResNet architectures to perform ship classification. Instead of training a CNN using random parameter initialization, we use transfer learning. We fine-tune pre-trained CNNs to perform maritime vessel image classification on a limited ship image dataset. We achieve a significant improvement in classification accuracy compared to the previous state-of-the-art results for the Maritime Vessel (Marvel) dataset.

Assessment of mental workload is crucial for applications that require sustained attention and where conditions such as mental fatigue and drowsiness must be avoided. Previous work that attempted to devise objective methods to model mental workload were mainly based on neurological or physiological data collected when the participants performed tasks that did not involve physical activity. While such models may be useful for scenarios that involve static operators, they may not apply in real-world situations where operators are performing tasks under varying levels of physical activity, such as those faced by first responders, firefighters, and police officers. Here, we describe WAUC, a multimodal database of mental Workload Assessment Under physical aCtivity. The study involved 48 participants who performed the NASA Revised Multi-Attribute Task Battery II under three different activity level conditions. Physical activity was manipulated by changing the speed of a stationary bike or a treadmill. During data collection, six neural and physiological modalities were recorded, namely: electroencephalography, electrocardiography, breathing rate, skin temperature, galvanic skin response, and blood volume pulse, in addition to 3-axis accelerometry. Moreover, participants were asked to answer the NASA Task Load Index questionnaire after each experimental section, as well as rate their physical fatigue level on the Borg fatigue scale. In order to bring our experimental setup closer to real-world situations, all signals were monitored using wearable, off-the-shelf devices. In this paper, we describe the adopted experimental protocol, as well as validate the subjective, neural, and physiological data collected. The WAUC database, including the raw data and features, subjective ratings, and scripts to reproduce the experiments reported herein will be made available at: http://musaelab.ca/resources/.

In recent years, various techniques have been applied to modeling radio-wave propagation in railway networks, each one presenting its own advantages and limitations. This paper presents a hybrid channel modeling technique, which combines two of these methods, the ray-tracing (RT) and vector parabolic equation (VPE) methods, to enable the modeling of realistic railway scenarios including stations and long guideways within a unified simulation framework. The general-purpose RT method is applied to analyze propagation in complex areas, whereas the VPE method is reserved for long and uniform tunnel as well as open-air sections. By using the advantages of VPE to compensate for the limitations of RT and vice versa, this hybrid model ensures improved accuracy and computational savings. Numerical results are validated with experimental measurements in various railway scenarios, including an actual deployment site of communication-based train control (CBTC) systems.

In order to address the problem of 3-D localization of an underwater target using a 2-D active sonar with unknown oceanographic factors in a multipath environment with heavy clutter, a novel iterative framework based on Maximum Likelihood Probabilistic Data Association (ML-PDA), which considers ocean sound speed profile (SSP) uncertainty and utilizes multiple detections to realize 3-D position estimation with only bearing and time of flight (ToF) measurements, is proposed. ML-PDA is highly effective in low SNR target detection. However, it is limited by its assumption of at most one target-originated detection within a scan. To estimate the 3-D target state with multipath detections under weak observability conditions, we first extend the ML-PDA into a multipath ML-PDA by enumerating the combined association events formed from multiple detection patterns. In contrast to the situation in air target tracking, the water column is nonhomogeneous and the underwater sound speed profile varies, influenced by uncertain ocean factors, e.g., temperature, salinity, and pressure. The resultant acoustic signal travels in a curvilinear path instead of a straight line. In this article, an SSP-dependent ToF measurement model is derived for both the direct path and the surface-reflected path between two remote nodes, so that the SSP uncertainty can be addressed systematically. By adopting an iterative prediction-update methodology, we first propagate the SSP uncertainty into the modified measurement covariance with the help of the unscented sampling technique. Then, we formulate a new joint likelihood ratio (JLLR) function based on the modified measurement covariance within the multidetection ML-PDA framework. A hybrid optimization method with grid search and particle swarm optimization is applied to solve the complex JLLR objective function and to find the optimal target state estimate from a large surveillance region. Finally, a sequential update technique is used to update the SSP state with the estimated target state and sensor measurements. In subsequent iterations, a more accurate JLLR can be rebuilt based on the updated SSP state, which can help find a better parameter estimate eventually. In addition, the Cramér-Rao lower bound, which quantifies the best possible accuracy in the presence of SSP uncertainties, is derived and analyzed. Numerical simulations confirm the underwater target localization performance of the proposed method in the presence of heavy clutter in an unknown ocean environment with a realistic sound propagation model.

A non-intrusive formulation of the polynomial chaos method is applied to quantify the uncertainties in deterministic models of the indoor radio channel. Deterministic models based on the finite-difference time-domain (FDTD) method and ray tracing are examined. Various sources of parameter uncertainty are considered, including randomness in the material properties, building geometry, and the spatial location of transmitting and receiving antennas. The polynomial chaos results are confirmed against Monte Carlo simulations and experimental measurements. The analysis shows the expected variation in the sector-averaged path loss can be considerable for relatively small input parameter uncertainties, leading to the conclusion that a single simulation run using `nominal values' may be insufficient to adequately characterize the indoor radio channel.

In the era of the internet of things (IoT), software-enabled inter-connected devices are of paramount importance. The embedded systems are very frequently used in both security and privacy-sensitive applications. However, the underlying software (a.k.a. firmware) very often suffers from a wide range of security vulnerabilities, mainly due to their outdated systems or reusing existing vulnerable libraries; which is evident by the surprising rise in the number of attacks against embedded systems. Therefore, to protect those embedded systems, detecting the presence of vulnerabilities in the large pool of embedded devices and their firmware plays a vital role. To this end, there exist several approaches to identify and trigger potential vulnerabilities within deployed embedded systems firmware. In this survey, we provide a comprehensive review of the state-of-the-art proposals, which detect vulnerabilities in embedded systems and firmware images by employing various analysis techniques, including static analysis, dynamic analysis, symbolic execution, and hybrid approaches. Furthermore, we perform both quantitative and qualitative comparisons among the surveyed approaches. Moreover, we devise taxonomies based on the applications of those approaches, the features used in the literature, and the type of the analysis. Finally, we identify the unresolved challenges and discuss possible future directions in this field of research.

The aim of the present study was to explore differences in the clinical expression, clinical diagnoses and management of airway diseases in a primary-care setting. Patients aged >or=35 yrs who had ever smoked were enrolled when they presented for any reason to one of eight rural primary-care practices. Respiratory symptom questionnaires and spirometry were administered. In total, 1,034 patients had acceptable and reproducible spirometry, of whom 550 (53%) were males and 484 (47%) were females. Males smoked more than females (41.2 versus 29.2 pack-yrs) respectively, and were more likely to have a pre-bronchodilator forced expiratory volume in one second/forced vital capacity <0.70 at 22.4 versus 11.8%, respectively. However, more females than males reported breathlessness (51.0 versus 42.8%, respectively), a prior diagnosis compatible with airflow obstruction and taking respiratory medications (23.4 versus 14.9%, respectively). In conclusion, the current results suggest that females are more likely than males to report breathlessness and be prescribed respiratory medications independent of differences in the severity of airflow obstruction.

With the growing market of Electric Vehicles (EV), the procurement of their charging infrastructure plays a crucial role in their adoption. Within the revolution of Internet of Things, the EV charging infrastructure is getting on board with the introduction of smart Electric Vehicle Charging Stations (EVCS), a myriad set of communication protocols, and different entities. We provide in this article an overview of this infrastructure detailing the participating entities and the communication protocols. Further, we contextualize the current deployment of EVCSs through the use of available public data. In the light of such a survey, we identify two key concerns, the lack of standardization and multiple points of failures, which renders the current deployment of EV charging infrastructure vulnerable to an array of different attacks. Moreover, we propose a novel attack scenario that exploits the unique characteristics of the EVCSs and their protocol (such as high power wattage and support for reverse power flow) to cause disturbances to the power grid. We investigate three different attack variations; sudden surge in power demand, sudden surge in power supply, and a switching attack. To support our claims, we showcase using a real-world example how an adversary can compromise an EVCS and create a traffic bottleneck by tampering with the charging schedules of EVs. Further, we perform a simulation-based study of the impact of our proposed attack variations on the WSCC 9 bus system. Our simulations show that an adversary can cause devastating effects on the power grid, which might result in blackout and cascading failure by comprising a small number of EVCSs.

With the burgeoning of wearable devices and passive body/brain-computer interfaces (B/BCIs), automated stress monitoring in everyday settings has gained significant attention recently, with applications ranging from serious games to clinical monitoring. With mobile users, however, challenges arise due to other overlapping (and potentially confounding) physiological responses (e.g., due to physical activity) that may mask the effects of stress, as well as movement artifacts that can be introduced in the measured signals. For example, the classical increase in heart rate can no longer be attributed solely to stress and could be caused by the activity itself. This makes the development of mobile passive B/BCIs challenging. In this paper, we introduce PASS, a multimodal database of Physical Activity and StresS collected from 48 participants. Participants performed tasks of varying stress levels at three different activity levels and provided quantitative ratings of their perceived stress and fatigue levels. To manipulate stress, two video games (i.e., a calm exploration game and a survival game) were used. Peripheral physical activity (electrocardiography, electrodermal activity, breathing, skin temperature) as well as cerebral activity (electroencephalography) were measured throughout the experiment. A complete description of the experimental protocol is provided and preliminary analyses are performed to investigate the physiological reactions to stress in the presence of physical activity. The PASS database, including raw data and subjective ratings has been made available to the research community at http://musaelab.ca/pass-database/. It is hoped that this database will help advance mobile passive B/BCIs for use in everyday settings.

The rapid development of self-driving vehicles requires integrating a sophisticated sensing system to address the various obstacles posed by road traffic efficiently. While several datasets are available to support object detection in autonomous vehicles, it is crucial to carefully evaluate the suitability of these datasets for different weather conditions across the globe. In response to this requirement, we present a novel dataset named the Canadian Vehicle Datasets (CVD). Subsequently, we present deep learning models that use this dataset. The CVD comprises street-level videos which were recorded by Thales, Canada. These videos were collected with high-quality cameras mounted on a vehicle in the Canadian province of Quebec. The recordings were made during daytime and nighttime, capturing weather conditions such as hazy, snowy, rainy, gloomy, nighttime and sunny days. A total of 10000 images of vehicles and other road assets are extracted from the collected videos. A total of 8388 images were annotated with corresponding generated labels 27766 with their respective 11 different classes. We analyzed the performance of the YOLOv8 model trained using the existing RoboFlow dataset. Then, we compared it with the model trained on the expanded version of RoboFlow using the proposed weather-specific dataset, CVD. Final values of improved accuracy of 73.26 %, 72.84 %, and 73.47 % (Precision/Recall/mAP) were reported upon adding the proposed dataset. Finally, the model trained on this diverse dataset exhibits heightened robustness and proves highly beneficial for both autonomous and conventional vehicle operations, making it applicable not only in Canada but also in other countries with comparable weather conditions.

In this paper, the multi-vehicle tracking problem is revisited, with greater consideration being given to the interactions between vehicles. Traditionally, algorithms for tracking multiple vehicles in the multi-lane case assume that vehicles move independently of one another and that longitudinal and lateral vehicle dynamics are mutually independent. However, due to traffic volume, limited lane resources, and traffic heterogeneity, vehicles have to interact with neighboring vehicles for the purposes of maintaining a safe distance from the leading vehicle or improving their navigability by passing slower vehicles. To address the limitations in the literature, this paper proposes a novel multi-vehicle tracking algorithm that integrates the microscopic traffic models (MTM) for modeling interaction behaviors among vehicles in a 2-D road coordinate system. Due to the dependence between the longitudinal and later motions, their corresponding estimates are updated sequentially in a recursive manner. An adaptive deferred decision logic is proposed to improve the accuracy of lateral state estimates and thus improve overall performance. Simulation results show that the proposed MTM-based tracking algorithm can achieve better performance than a conventional multi-lane vehicle tracking algorithm with extension to multi-vehicle tracking, which does not consider interactions among vehicles but updates the longitudinal and lateral motion estimates independently.

Recently, due to the emergence of mobile electroencephalography (EEG) devices, assessment of mental workload in highly ecological settings has gained popularity. In such settings, however, motion and other common artifacts have been shown to severely hamper signal quality and to degrade mental workload assessment performance. Here, we show that classical EEG enhancement algorithms, conventionally developed to remove ocular and muscle artifacts, are not optimal in settings where participant movement (e.g., walking or running) is expected. As such, an adaptive filter is proposed that relies on an accelerometer-based referential signal. We show that when combined with classical algorithms, accurate mental workload assessment is achieved. To test the proposed algorithm, data from 48 participants was collected as they performed the Revised Multi-Attribute Task Battery-II (MATB-II) under a low and a high workload setting, either while walking/jogging on a treadmill, or using a stationary exercise bicycle. Accuracy as high as 95% could be achieved with a random forest based mental workload classifier with ambulant users. Moreover, an increase in gamma activity was found in the parietal cortex, suggesting a connection between sensorimotor integration, attention, and workload in ambulant users.

Large scale Electric Vehicles (EV) penetration is coming with a stupendous energy demand that raises much concerns in the power sector. The impact of this demand is mostly notable at the distribution side as an outcome of EV users home charging preferences. Currently, most EVs charge their batteries through level 1 charger at home. However, the shorter charging times and the declining prices of level 2 chargers favor a switch from level 1 into level 2 chargers at residential premises. As a ramification, this will cause a lamentable peak in the residential load profile, and consequently power utilities will face the impact of elevated number of level 2 chargers with uncontrolled EV charging. To foresee these consequences, using the IEEE-33 Bus radial distribution system, we build a discrete event simulator and present real-life assessment of different EV penetration rates with various level 2 charger adoption rates. We expose that 50% EV penetration along with 50% level 2 chargers deployment may create an undesirable situation on the distribution network. Furthermore, we simulate EV users' charging behavior over different pricing techniques. The collected results show that available pricing techniques cannot maintain the voltage level over minimum desired threshold especially during peak times.

We introduce a Parametric Information Maximization (PIM) model for the Generalized Category Discovery (GCD) problem. Specifically, we propose a bi-level optimization formulation, which explores a parameterized family of objective functions, each evaluating a weighted mutual information between the features and the latent labels, subject to supervision constraints from the labeled samples. Our formulation mitigates the class-balance bias encoded in standard information maximization approaches, thereby handling effectively both short-tailed and long-tailed data sets. We report extensive experiments and comparisons demonstrating that our PIM model consistently sets new state-of-the-art performances in GCD across six different datasets, more so when dealing with challenging fine-grained problems. Our code: https://github.com/ThalesGroup/pim-generalized-category-discovery.

The paper summarises the evolution of rail signalling system from track-circuit signalling to advanced Communication-Based Train Control (CBTC) systems, highlighting the differences, benefits and challenges with regard to improving performance while ensuring safety. It describes the evolution of such systems over the last three decades; from wayside signals to cab signals, to profile-based systems, to communications-based train control systems for an effective means of overcoming the fundamental limitations of the conventional track-circuit based system. CBTC allows trains to operate with much closer headways while ensuring safe train separation and better train control flexibility. Not only do CBTC systems determine the position of the train with a higher degree of accuracy independently of the track circuits, but also offer bidirectional train/wayside communication to enhance train operations and supervision. Operational and performance benefits realised with the newer technology systems are discussed, together with the challenges of implementing such systems. Standardisation for Communication Based Signalling System initiatives is also described and industry trends presented.

Industrial Control Systems (ICSs) are cyber-physical systems that offer attractive targets to threat actors due to the scale of damages, both physical and cyber, that successful exploitation can cause. As such, ICSs often find themselves victims to reconnaissance campaigns - coordinated scanning activity that targets a wide subset of the Internet - that aim to discover vulnerable systems. As these campaigns likely scan broad netblocks of the Internet, some traffic is directed to network telescopes, which are routable, allocated, and unused IP space. In this paper, we explore the threat landscape of ICS devices by analyzing and investigating network telescope traffic. Our network traffic analysis tool takes darknet traffic and generates threat intelligence on scanning campaigns targeting ICSs in the form of campaign fragments, which we leverage in new ways to get more in-depth knowledge of the cybersecurity threats. We investigate the payloads of the identified campaigns using a custom Deep Packet Inspection (DPI) technique to dissect and analyze the packets. We found 13 distinct payload templates and deduced their purpose, and by extension the campaign goals. We use machine learning to classify the sources behind the campaigns and identify threat actors such as botnets, malicious attackers, or researchers, and establish a methodology to rank our campaigns to prioritize our analysis. To conduct our analysis of the threats targeting ICSs, we have leveraged 12.85 TB (330 days) of network traffic received by our observed darknet IP space. Combining these investigative threads, we provide a thorough overview of the threat landscape targeting ICS systems.