Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
K-Means and Gaussian Mixture Models on Lie Groups: Application to Geometrical Clustering
No full textIn this article, we derive and implement two new clustering algorithms dedicated to Lie groups, adapted from the well known K-Means algorithm and Gaussian mixture models. More precisely, the K-Means algorithm is reformalized by taking into account the fact that observations belong to a Lie group (LG) with an appropriate intrinsic metric and Gaussian mixture model are defined for data living in LGs. The consistency and the performance of the resulting are numerically validated for data lying on the LG SE(2), by comparison with state-of-the-art methods for synthetic data and pseudo-real data generated using an ultra-sound sensor model
A High Availability Inertial-Vision Data Fusion Using an ES-KF for a Civil Aircraft During a Precision Approach in a GNSS-Challenged Environment
No full textInternational audienceThis paper introduces the architecture of a localization system designed for commercial airliners. The system is designed to maintain guidance during an PBN CAT I approach in a GNSS-challenged environment. It leverages a vision sensor to observe the aircraft’s surroundings. Specifically, this new sensor allows for the detection and tracking of the runway during the final segment of the approach, providing line-of-sight information to the system. The vision sensor is integrated into a multi-sensor system that includes a high-precision navigation-grade Inertial Measurement Unit (IMU) (angular drift of the order of 0.01 degree per hour), a barometric altimeter, and a GNSS receiver. Data fusion is performed using an Error-State Kalman Filter (ES-KF) within a semi-closed loop framework. The ES-KF is typically used for IMU hybridization systems, as its allows for quick testing and modification of propagation and observation models to address challenges like false observability. The system’s design is made in order to address the challenges posed by the limited number of landmarks and the high correlations on the measurement errors elaborated from them, where two points on the runway may be indiscernible from the perspective of the camera, and where distance estimation may be challenging. The system’s architecture, the types of sensors used, and the choice of a tightly-coupled single-feature hybridization improve the system’s operational period and, ultimately, its availability. The regulatory framework for integrating a vision sensor into the localization system of a civil aircraft is discussed. The performance of this vision-integrated localization system is evaluated using a comprehensive Monte Carlo simulator
Assessing Spoofer Impact on GNSS Receivers : Tracking Loops
No full textInternational audienceIn the context of global navigation satellite systems (GNSS), synchronization is crucial for successfully decoding the navigation message and accurately estimating pseudoranges. Synchronization of each received GNSS signal typically involves at least two tracking loops: a delay lock loop (DLL) and a phase lock loop (PLL). The reception of a spoofed signal disrupts the synchronization process, potentially leading to erroneous pseudorange estimation or loss of service. This paper investigates the impact of spoofing on code, carrier phase, and frequency tracking estimates and proposes a transformation-based strategy to characterize the joint DLL and PLL under spoofing, focusing on the system's stable equilibria (SE), linearity and interdependence, transient response, and noise impact. The study reveals the nonlinearity and interdependence of the tracking loops (i.e., PLL and DLL cannot be considered separately) and shows the emergence of multiple SE, leading to potential chaotic behavior and bifurcation
The impact of cross-validation choices on pBCI classification metrics: lessons for transparent reporting
No full textInternational audienceNeuroadaptive technologies are a type of passive Brain-computer interface (pBCI) that aim to incorporate implicit user-state information into human-machine interactions by monitoring neurophysiological signals. Evaluating machine learning and signal processing approaches represents a core aspect of research into neuroadaptive technologies. These evaluations are often conducted under controlled laboratory settings and offline, where exhaustive analyses are possible. However, the manner in which classifiers are evaluated offline has been shown to impact reported accuracy levels, possibly biasing conclusions. In the current study, we investigated one of these sources of bias, the choice of cross-validation scheme, which is often not reported in sufficient detail. Across three independent electroencephalography (EEG) n-back datasets and 74 participants, we show how metrics and conclusions based on the same data can diverge with different cross-validation choices. A comparison of cross-validation schemes in which train and test subset boundaries either respect the block-structure of the data collection or not, illustrated how the relative performance of classifiers varies significantly with the evaluation method used. By computing bootstrapped 95% confidence intervals of differences across datasets, we showed that classification accuracies of Riemannian minimum distance (RMDM) classifiers may differ by up to 12.7% while those of a Filter Bank Common Spatial Pattern (FBCSP) based linear discriminant analysis (LDA) may differ by up to 30.4%. These differences across cross-validation implementations may impact the conclusions presented in research papers, which can complicate efforts to foster reproducibility. Our results exemplify why detailed reporting on data splitting procedures should become common practice
Reconnaissance des émotions à l'aide du Deep Learning
No full textUnderstanding human emotions is crucial in healthcare, human-robot interaction, and marketing. Despite the progress in emotion recognition from one modality, such as a facial video and a sequence of physiological signals, it is still challenging to improve by combining multiple modalities.Moreover, it is difficult to recognize emotions in long sequential data, such as long videos, although most real-world videos of people expressing emotions are long.Existing emotion datasets are limited in volume and quality, making it difficult to develop an effective deep learning-based emotion recognition system. An effective real-world emotion understanding system should be able to recognise emotions from long videos synchronized with multiple modalities.In this thesis, we focus on multimodal emotion recognition from long videos synchronized with physiological signals. Specifically, multimodal emotion methods face three main challenges: (a) learning the emotion representation, (b) learning the representation of fine-grained emotions, as well as (c) combining modalities to predict emotions. In this thesis, we first introduce two large behaviour analysis datasets: INEMO and StressID. INEMO is a multimodal dataset designed to facilitate emotion recognition from watching social media videos. StressID is a multimodal dataset designed for stress identification. Secondly, we propose two pre-training techniques for facial expression recognition: (1) supervised pre-training on synthetic data generated by our video generation method and (2) self-supervised pre-training on multi-view videos. We show that the proposed pre-training techniques allow us to get rich facial representations which allow us to improve fine-grained emotion recognition accuracy.Thirdly, we tackle the problem of emotion recognition from multiple modalities. We propose a framework for multimodal fusion of videos and physiological signals to predict emotions.This framework consists of mainly two steps: (1) extracting features from long raw videos and physiological signals; (2) fusing extracted features to predict emotions using a cross-modality approach based on attention mechanism. Our method leverages the additional modalities resulting in better emotion recognition performance.Our methods have been extensively evaluated on various emotion recognition benchmarks. The proposed methods outperform previous methods, significantly pushing emotion recognition to real-world deployments.La compréhension des émotions humaines est cruciale dans des domaines tels que les soins de santé, l'interaction homme-robot et le marketing. Malgré les progrès réalisés dans la reconnaissance des émotions à partir d'une seule modalité, comme une vidéo faciale ou une séquence de signaux physiologiques, il reste difficile d'améliorer les performances en combinant plusieurs modalités. De plus, reconnaître les émotions dans des données séquentielles longues, telles que des vidéos de longue durée, est un défi, bien que la plupart des vidéos réelles de personnes exprimant des émotions soient longues. Les ensembles de données existants sur les émotions sont limités en volume et en qualité, ce qui complique le développement de systèmes de reconnaissance des émotions basés sur l'apprentissage profond. Un système efficace de compréhension des émotions dans des situations réelles doit être capable de reconnaître les émotions à partir de longues vidéos synchronisées avec plusieurs modalités. Dans cette thèse, nous nous concentrons sur la reconnaissance multimodale des émotions à partir de longues vidéos synchronisées avec des signaux physiologiques. Plus précisément, les méthodes de reconnaissance multimodale des émotions rencontrent trois principaux défis : (a) l'apprentissage de représentations des émotions selon, (b) l'apprentissage de représentations d'émotions fines, et(c) la combinaison de modalités pour prédire les émotions. Dans ce travail, nous introduisons tout d'abord deux grands ensembles de données pour l'analyse comportementale : INEMO et StressID. INEMO est un ensemble de données multimodales conçu pour faciliter la reconnaissance des émotions lors du visionnage de vidéos issues des réseaux sociaux. StressID, quant à lui, est un ensemble de données multimodales conçu pour l'identification du stress. Ensuite, nous proposons deux techniques de pré-entraînement pour la reconnaissance des expressions faciales : 1. un pré-entraînement supervisé sur des données synthétiques générées par notre méthode de génération de vidéos, et 2. un pré-entraînement auto-supervisé sur des vidéos multi-vues. Nous démontrons que ces techniques permettent d'obtenir des représentations faciales riches, améliorant ainsi la précision de la reconnaissance des émotions fines. Enfin, nous abordons la problématique de la reconnaissance des émotions à partir de multiples modalités. Nous proposons un cadre pour la fusion multimodale de vidéos et de signaux physiologiques afin de prédire les émotions. Ce cadre repose principalement sur deux étapes : 1. l'extraction de caractéristiques à partir de longues vidéos brutes et de signaux physiologiques ; 2. la fusion des caractéristiques extraites pour prédire les émotions en utilisant une approche intermodale basée sur un mécanisme d'attention. Notre méthode exploite les modalités supplémentaires, ce qui améliore significativement les performances de reconnaissance des émotions. Nos méthodes ont été évaluées de manière approfondie sur plusieurs ensembles de données de référence pour la reconnaissance des émotions. Les résultats montrent que les méthodes proposées surpassent les approches précédentes, ouvrant la voie à des déploiements concrets dans des environnements réels
GAUGE-COMPATIBLE TENSORS IN STATISTICAL MANIFOLDS
No full textInternational audienceWe investigate gauge-compatible (1, 1)-tensor fields on statistical manifolds and their impact on the cohomology of Koszul-Vinberg (pre-Lie) algebras naturally associated with torsion-free affine connections. Given a pair of dual connections (∇, ∇ * ), we study solutions of the gauge equationand analyze how such tensors induce algebraic structures and double complexes whose associated spectral sequences encode geometric information.Our first contribution is the introduction of a new product based the covariant derivative of a gauge equation solution θ. Provided θ is a weak Nijenhuis tensor, a curvature-like tensor can be defined and when it is vanishing, the product turns out to define a Koszul-Vinberg algebra on the sections of the tangent bundle. Our second main result establishes that, under suitable gauge-compatibility conditions adapted to invariant distributions, the Koszul-Vinberg spectral sequence degenerates at the E 2 -page.These general results are applied to statistical coKähler manifolds, where the structure tensor φ provides a canonical, nontrivial solution of the gauge equation. In this setting, we obtain explicit geometric criteria ensuring spectral degeneracy and show that, for CR-submanifolds, the vanishing of two natural obstruction tensors is equivalent to both gauge flatness and degeneration of the associated spectral sequence. Our results unify gauge-theoretic compatibility, pre-Lie algebra cohomology, and geometric structure, providing explicit models where homological simplifications reflect intrinsic geometric properties.</div
Moment-Matching Array Processing Technique for diffuse source estimation
No full textDirection of Arrival (DOA) estimation is a fundamental problem in signal processing. Diffuse sources, whose power density cannot be represented with a single angular coordinate, are usually characterized based on prior assumptions, which associate the source angular density with a specific set of functions. However, these assumptions can lead to significant estimation biases when they are incorrect. This paper introduces the Moment-Matching Estimation Technique (MoMET), a low-complexity method for estimating the mean DOA, spread, and power of a narrow diffuse source without requiring prior knowledge on the source distribution. The unknown source density is characterized by its mean DOA and its first central moments, which are estimated through covariance matching techniques which fit the empirical covariance of the measurements to that modeled from the moments. The MoMET parameterization is robust to incorrect model assumptions, and numerically efficient. The asymptotic bias and covariance of the new estimator are derived and its performance is demonstrated through simulations
On-line Pick-Freeze Mirror algorithm for Sensitity Analysis
No full textThe main objective of this paper is to propose a new approach for estimating the entire collection of Sobol' indices simultaneously. Our approach exploits the fact that Sobol' indices can be rewritten as solutions to an optimization problem over the simplex of R d , to construct an online sequence of estimators using a stochastic mirror descent algorithm. We prove that our estimation procedure is consistent and provide a non-asymptotic upper bound for its rate of convergence. Furthermore, we demonstrate the numerical accuracy of our method and compare it with other classical estimation procedures
Multi-objective four-dimensional trajectory planning for free-route operations considering weather forecast uncertainty
No full textInternational audienceFree-route Operations (FRO) represent a significant shift in the future air traffic management, offering enhanced system capacity and operational efficiency. However, there remains a lack of trajectory management methods tailored to FRO that can effectively address the challenges of unstructured traffic flows, high traffic complexity, and uncertain weather conditions. This paper proposes a novel multi-module framework for 4D free-routing trajectory management, including trajectory prediction, airspace configuration, and strategic and tactical trajectory planning. Trajectory prediction incorporates ensemble-based weather forecasts to address uncertainty. A routing method tailored to FRO is adopted, combining a two-layer network with a K-shortest path algorithm to generate alternative routes. The core of this framework is a strategic trajectory planning method, integrating flow management and conflict management to optimize safety, efficiency, and flexibility. A complexity metric based on the linear dynamical system is adopted for safety evaluation. A distributed multi-objective optimization method is designed based on the decomposition mechanism, solving subproblems in parallel. The proposed framework is validated through a simulation scenario based on historical data from the western China airspace. Results demonstrate that the proposed method reduces operational risk by 88.52 % and increases flexibility by 51.66 %, with only a 6.84 % increase in cost. Additionally, trade-offs among three objectives are identified from non-dominated solutions, and a multi-criteria decision-making method guides the selection of the ideal solution and maneuver type. The proposed method also demonstrates high computational efficiency, making it a practical decision-support tool for future free-route operations
Stokes-Lagrange and Stokes-Dirac representations of N-dimensional port-Hamiltonian systems for modeling and control
No full textInternational audienceIn this paper, we extended the port-Hamiltonian framework by introducing the concept of Stokes-Lagrange structure, which enables the implicit definition of a Hamiltonian over an N-dimensional domain and incorporates energy ports into the system. This new framework parallels the existing Dirac and Stokes-Dirac structures. We proposed the Stokes-Lagrange structure as a specific case where the subspace is explicitly defined via differential operators that satisfy an integration-by-parts formula. By examining various examples through the lens of the Stokes-Lagrange structure, we demonstrated the existence of multiple equivalent system representations. These representations provide significant advantages for both numerical simulation and control design, offering additional tools for the modeling and control of port-Hamiltonian systems