Localization is a crucial topic in navigation, especially in autonomous vehicles navigation. It is usually done by using a global positioning system (GPS) sensor. Even though there have been many studies of vehicle localization in recent years, most of them combine GPS sensor with other sensors to get a more accurate result [1]. In this thesis, we propose a novel image-based vehicle localization by utilizing vision sensor and computer vision techniques to extract vehicle surrounding text landmarks and to locate the vehicle position. Firstly, we explore the feasibility of image-based vehicle localization by using text landmark of a position to locate vehicle position. A text landmark model, a location matching algorithm and a basic localization model are proposed, which allow a vehicle to find the best matching location in the database by cross-checking the text landmarks from query image and reference location images. Secondly, we propose two more robust localization models by applying vehicle moving distance and heading direction data as part of inputs, which significantly improve the localization accuracy. Finally, we simulate an experiment to evaluate our three different localization models and further prove the robustness of our model through experimental results. / Master of Science / In modern days, global positioning system (GPS) is the major approach to locate positions. However, GPS is not as reliable as we thought. Under some environmental situations, GPS cannot provide continuous navigation information. Besides, GPS signals can be jammed or spoofed by malicious attackers. In this thesis, we aim to explore how to locate the vehicle’s position without using GPS sensor. Here, we propose a novel image-based vehicle localization by utilizing vision sensor and computer vision techniques to extract vehicle surrounding text landmarks and to locate the vehicle position. Various tools and techniques are explored in the process of the research. With the explored result, we propose several localization models and simulate an experiment to prove the robustness of these models.

Vehicle self-localization, the ability of a vehicle to determine its own location, is vital for many aspects of Intelligent Transportation Systems (ITS) and telematics where it is often a building block in a more complex system. Navigation systems are perhaps the most obvious example, requiring knowledge of the vehicle's location on a map to calculate a route to a desired destination. Other pervasive examples are the monitoring of vehicle fleets for tracking shipments or dispatching emergency vehicles, and in public transit systems to inform riders of time-of-arrival thereby assisting trip planning. These system often depend on Global Positioning System (GPS) technology to provide vehicle localization information; however, GPS is challenged in urban environments where satellite visibility and multipath conditions are common. Vehicle localization is made more robust to these issues through augmentation of GPS-based localization with complementary sensors, thereby improving the performance and reliability of systems that depend on localization information. This thesis investigates the augmentation of vehicle localization systems with visual context. Positioning the vehicle with respect to objects in its surrounding environment in addition to using GPS constraints the possible vehicle locations, to provide improved localization accuracy compared to a system relying solely on GPS. A modular system architecture based on Bayesian filtering is proposed in this thesis that enables existing localization systems to be augmented by visual context while maintaining their existing capabilities. It is shown in this thesis that localization errors caused by GPS signal multipath can be reduced by positioning the vehicle with respect to visually-detected intersection road markings. This error reduction is achieved when the identities of the detected road marking and the road being driven are known a priori. It is further shown how to generalize the approach to the situation when the identities of these parameters are unknown. In this situation, it is found that the addition of visual context to the vehicle localization system reduces the ambiguity of identifying the road being driven by the vehicle. The fact that knowledge of the road being driven is required by many applications of vehicle localization makes this a significant finding. A related problem is also explored in this thesis: that of using vehicle position information to augment machine vision. An approach is proposed whereby a machine vision system and a vehicle localization system can share their information with one another for mutual benefit. It is shown that, using this approach, the most uncertain of these systems benefits the most by this sharing of information. Augmenting vehicle localization with visual context is neither farfetched nor impractical given the technology available in today's vehicles. It is not uncommon for a vehicle today to come equipped with a GPS-based navigation system, and cameras for lane departure detection and parking assistance. The research in this thesis brings the capability for these existing systems to work together.

Vehicle self-localization, the ability of a vehicle to determine its own location, is vital for many aspects of Intelligent Transportation Systems (ITS) and telematics where it is often a building block in a more complex system. Navigation systems are perhaps the most obvious example, requiring knowledge of the vehicle's location on a map to calculate a route to a desired destination. Other pervasive examples are the monitoring of vehicle fleets for tracking shipments or dispatching emergency vehicles, and in public transit systems to inform riders of time-of-arrival thereby assisting trip planning. These system often depend on Global Positioning System (GPS) technology to provide vehicle localization information; however, GPS is challenged in urban environments where satellite visibility and multipath conditions are common. Vehicle localization is made more robust to these issues through augmentation of GPS-based localization with complementary sensors, thereby improving the performance and reliability of systems that depend on localization information. This thesis investigates the augmentation of vehicle localization systems with visual context. Positioning the vehicle with respect to objects in its surrounding environment in addition to using GPS constraints the possible vehicle locations, to provide improved localization accuracy compared to a system relying solely on GPS. A modular system architecture based on Bayesian filtering is proposed in this thesis that enables existing localization systems to be augmented by visual context while maintaining their existing capabilities. It is shown in this thesis that localization errors caused by GPS signal multipath can be reduced by positioning the vehicle with respect to visually-detected intersection road markings. This error reduction is achieved when the identities of the detected road marking and the road being driven are known a priori. It is further shown how to generalize the approach to the situation when the identities of these parameters are unknown. In this situation, it is found that the addition of visual context to the vehicle localization system reduces the ambiguity of identifying the road being driven by the vehicle. The fact that knowledge of the road being driven is required by many applications of vehicle localization makes this a significant finding. A related problem is also explored in this thesis: that of using vehicle position information to augment machine vision. An approach is proposed whereby a machine vision system and a vehicle localization system can share their information with one another for mutual benefit. It is shown that, using this approach, the most uncertain of these systems benefits the most by this sharing of information. Augmenting vehicle localization with visual context is neither farfetched nor impractical given the technology available in today's vehicles. It is not uncommon for a vehicle today to come equipped with a GPS-based navigation system, and cameras for lane departure detection and parking assistance. The research in this thesis brings the capability for these existing systems to work together.

Abstract : Embedded intelligence in vehicular applications is becoming of great interest since the last two decades. Position estimation has been one of the most crucial pieces of information for Intelligent Transportation Systems (ITS). Real time, accurate and reliable localization of vehicles has become particularly important for the automotive industry. The significant growth of sensing, communication and computing capabilities over the recent years has opened new fields of applications, such as ADAS (Advanced driver assistance systems) and active safety systems, and has brought the ability of exchanging information between vehicles. Most of these applications can benefit from more accurate and reliable localization. With the recent emergence of multi-vehicular wireless communication capabilities, cooperative architectures have become an attractive alternative to solving the localization problem. The main goal of cooperative localization is to exploit different sources of information coming from different vehicles within a short range area, in order to enhance positioning system efficiency, while keeping the cost to a reasonable level. In this Thesis, we aim to propose new and effective methods to improve vehicle localization performance by using cooperative approaches. In order to reach this goal, three new methods for cooperative vehicle localization have been proposed and the performance of these methods has been analyzed. Our first proposed cooperative method is a Cooperative Map Matching (CMM) method which aims to estimate and compensate the common error component of the GPS positioning by using cooperative approach and exploiting the communication capability of the vehicles. Then we propose the concept of Dynamic base station DGPS (DDGPS) and use it to generate GPS pseudorange corrections and broadcast them for other vehicles. Finally we introduce a cooperative method for improving the GPS positioning by incorporating the GPS measured position of the vehicles and inter-vehicle distances. This method is a decentralized cooperative positioning method based on Bayesian approach. The detailed derivation of the equations and the simulation results of each algorithm are described in the designated chapters. In addition to it, the sensitivity of the methods to different parameters is also studied and discussed. Finally in order to validate the results of the simulations, experimental validation of the CMM method based on the experimental data captured by the test vehicles is performed and studied. The simulation and experimental results show that using cooperative approaches can significantly increase the performance of the positioning methods while keeping the cost to a reasonable amount. / Résumé : L’intelligence embarquée dans les applications véhiculaires devient un grand intérêt depuis les deux dernières décennies. L’estimation de position a été l'une des parties les plus cruciales concernant les systèmes de transport intelligents (STI). La localisation précise et fiable en temps réel des véhicules est devenue particulièrement importante pour l'industrie automobile. Les améliorations technologiques significatives en matière de capteurs, de communication et de calcul embarqué au cours des dernières années ont ouvert de nouveaux champs d'applications, tels que les systèmes de sécurité active ou les ADAS, et a aussi apporté la possibilité d'échanger des informations entre les véhicules. Une localisation plus précise et fiable serait un bénéfice pour ces applications. Avec l'émergence récente des capacités de communication sans fil multi-véhicules, les architectures coopératives sont devenues une alternative intéressante pour résoudre le problème de localisation. L'objectif principal de la localisation coopérative est d'exploiter différentes sources d'information provenant de différents véhicules dans une zone de courte portée, afin d'améliorer l'efficacité du système de positionnement, tout en gardant le coût à un niveau raisonnable. Dans cette thèse, nous nous efforçons de proposer des méthodes nouvelles et efficaces pour améliorer les performances de localisation du véhicule en utilisant des approches coopératives. Afin d'atteindre cet objectif, trois nouvelles méthodes de localisation coopérative du véhicule ont été proposées et la performance de ces méthodes a été analysée. Notre première méthode coopérative est une méthode de correspondance cartographique coopérative (CMM, Cooperative Map Matching) qui vise à estimer et à compenser la composante d'erreur commune du positionnement GPS en utilisant une approche coopérative et en exploitant les capacités de communication des véhicules. Ensuite, nous proposons le concept de station de base Dynamique DGPS (DDGPS) et l'utilisons pour générer des corrections de pseudo-distance GPS et les diffuser aux autres véhicules. Enfin, nous présentons une méthode coopérative pour améliorer le positionnement GPS en utilisant à la fois les positions GPS des véhicules et les distances inter-véhiculaires mesurées. Ceci est une méthode de positionnement coopératif décentralisé basé sur une approche bayésienne. La description détaillée des équations et les résultats de simulation de chaque algorithme sont décrits dans les chapitres désignés. En plus de cela, la sensibilité des méthodes aux différents paramètres est également étudiée et discutée. Enfin, les résultats de simulations concernant la méthode CMM ont pu être validés à l’aide de données expérimentales enregistrées par des véhicules d'essai. La simulation et les résultats expérimentaux montrent que l'utilisation des approches coopératives peut augmenter de manière significative la performance des méthodes de positionnement tout en gardant le coût à un montant raisonnable.

Les applications pour véhicules autonomes et les systèmes d’aide avancée à la conduite (Advanced Driving Assistance Systems - ADAS) mettent en oeuvre des processus permettant à des systèmes haut niveau de réaliser une prise de décision. Pour de tels systèmes, la connaissance du positionnement précis (ou localisation) du véhicule dans son environnement est un pré-requis nécessaire. Cette thèse s’intéresse à la détection de la structure de scène, au processus de localisation ainsi qu’à la modélisation d’erreurs. A partir d’un large spectre fonctionnel de systèmes de vision, de l’accessibilité d’un système de cartographie ouvert (Open Geographical Information Systems - GIS) et de la large diffusion des systèmes de positionnement dans les véhicules (Global Positioning System - GPS), cette thèse étudie la performance et la fiabilité d’une méthode de localisation utilisant ces différentes sources. La détection de marquage sur la route réalisée par caméra monoculaire est le point de départ permettant de connaître la structure de la scène. En utilisant, une détection multi-noyau avec pondération hiérarchique, la méthode paramétrique proposée effectue la détection et le suivi des marquages sur la voie du véhicule en temps réel. La confiance en cette source d’information a été quantifiée par un indicateur de vraisemblance. Nous proposons ensuite un système de localisation qui fusionne des informations de positionnement (GPS), la carte (GIS) et les marquages détectés précédemment dans un cadre probabiliste basé sur un filtre particulaire. Pour ce faire, nous proposons d’utiliser les marquages détectés non seulement dans l’étape de mise en correspondance des cartes mais aussi dans la modélisation de la trajectoire attendue du véhicule. La fiabilité du système de localisation, en présence d’erreurs inhabituelles dans les différentes sources d’information, est améliorée par la prise en compte de différents indicateurs de confiance. Ce mécanisme est par la suite utilisé pour identifier les sources d’erreur. Cette thèse se conclut par une validation expérimentale des méthodes proposées dans des situations réelles de conduite. Leurs performances ont été quantifiées en utilisant un véhicule expérimental et des données en libre accès sur internet. / Autonomous Vehicles (AV) applications and Advanced Driving Assistance Systems (ADAS) relay in scene understanding processes allowing high level systems to carry out decision marking. For such systems, the localization of a vehicle evolving in a structured dynamic environment constitutes a complex problem of crucial importance. Our research addresses scene structure detection, localization and error modeling. Taking into account the large functional spectrum of vision systems, the accessibility of Open Geographical Information Systems (GIS) and the widely presence of Global Positioning Systems (GPS) onboard vehicles, we study the performance and the reliability of a vehicle localization method combining such information sources. Monocular vision–based lane marking detection provides key information about the scene structure. Using an enhanced multi-kernel framework with hierarchical weights, the proposed parametric method performs, in real time, the detection and tracking of the ego-lane marking. A self-assessment indicator quantifies the confidence of this information source. We conduct our investigations in a localization system which tightly couples GPS, GIS and lane makings in the probabilistic framework of Particle Filter (PF). To this end, it is proposed the use of lane markings not only during the map-matching process but also to model the expected ego-vehicle motion. The reliability of the localization system, in presence of unusual errors from the different information sources, is enhanced by taking into account different confidence indicators. Such a mechanism is later employed to identify error sources. This research concludes with an experimental validation in real driving situations of the proposed methods. They were tested and its performance was quantified using an experimental vehicle and publicly available datasets.

A critical requirement for safe autonomous driving is to have an accurate state estimate of thevehicle. One of the most ubiquitous yet reliable ways for this task is through the integrationof the onboard Inertial Navigation System (INS) and the Global Navigation Satellite System(GNSS). This integration can further be assisted through fusion of information from otheronboard sensors. On top of that, a ground vehicle enforces its own set of rules, through non-holonomic constraints, which along with other vehicle dynamics can aid the state estimation.In this project, a sequential probabilistic inference approach has been followed, that fusesthe high frequency, short term accurate INS estimates, with low frequency, drift free GPSobservations. The fusion of GPS and IMU has been sought through a modular asynchronousloosely coupled framework, capable of augmenting additional observation sources to facilitatethe state estimation and tracking process. Besides GPS and IMU, the applied strategy makesuse of wheel speed sensor measurements, nonholonomic constraints and online estimationof IMU sensor biases as well wheel speed scalling factor. Theses augmentations have beenshown to increase the robustness of the localization module, under periods of GPS outage.The Extended Kalman Filter (EKF) has seen extensive usage for such sensor fusion tasks,however, the performance can be limited due to the propagation of the covariance throughlinearization of the underlying non-linear model. The Unscented Kalman Filter (UKF) avoidsthe issue of linearization based on jacobians. Instead, it uses a carefully chosen set ofsample points in order to accurately map the probability distribution. Correspondingly, thesurrounding literature also indicates towards the UKF out performing EKF in such tasks.Therefore, the present thesis also seeks to evaluate these claims.The EKF and SRUKF (Square Root UKF) instances of the developed algorithm have beentested on real sensor logs, recorded from a Scania test vehicle. Under no GPS outage situation,the implemented localization algorithm performs within a position RMSE of 60cm.The robustness of the localization algorithm, to GPS outages, is evaluated by simulating0-90% lengths of GPS unavailability, during the estimation process. Additionally, to unfoldthe impact of parameters, the individual modules within the suggested framework wereisolated and analysed with respect to their contribution towards the algorithm’s localizationperformance.Out of all, the online estimation of IMU sensor biases proved to be critical for increasingthe robustness of the suggested localization algorithm to GPS shortage, especially for the EKF.In terms of the distinction, both the EKF and the SRUKF performed to similar capabilities,however, the UKF showed better results for higher levels of GPS cuts. / Ett kritiskt krav för säker autonom körning är att ha en korrekt tillståndsuppskattning avfordonet. Ett av de mest förekommande men ändå tillförlitliga sätten för denna uppgift ärgenom integrationen av det inbyggda tröghetsnavigationssystemet (INS) och med Satellitnavi-gation (GNSS). Denna integration kan ytterligare underlättas genom sammanslagning avinformation från andra sensorer ombord. Utöver det upprätthåller ett markfordon sin egenuppsättning regler, genom icke-holonomiska begränsningar, som tillsammans med annanfordonsdynamik kan hjälpa till vid tillståndsuppskattningen.I detta projekt har en sekventiell probabilistisk slutledning följts, som sammansmälterde högfrekventa, kortsiktiga exakta INS-uppskattningarna, med lågfrekventa, driftfria GPS-observationer. Sammanslagningen av GPS och IMU har sökts genom ett modulärt asynkrontlöst kopplat ramverk, som kan utökas med ytterligare observationskällor för att underlättatillståndsuppskattningen och spårningsprocessen. Förutom GPS och IMU använder dentillämpade strategin mätningar av hjulhastighetssensorer, icke-holonomiska begränsningaroch onlineuppskattning av IMU-sensorbias samt hjulhastighetsskalningsfaktor. Dessa tillägghar visat sig öka robustheten hos lokaliseringsmodulen under perioder utan GPS-signal.Extended Kalman Filter (EKF) har sett omfattande användning för sådana sensorfusionsup-pgifter, men prestandan kan begränsas på grund av spridningen av kovariansen genomlinearisering av den underliggande icke-linjära modellen. Unscented Kalman Filter (UKF)undviker frågan om linearisering baserad på jacobianer. Istället använder den en noggrantutvald uppsättning provpunkter för att korrekt kartlägga sannolikhetsfördelningen. På motsva-rande sätt indikerar den omgivande litteraturen också mot UKF att utföra EKF i sådanauppgifter. Därför försöker denna avhandling också utvärdera dessa påståenden.EKF- och SRUKF-instanserna (Square Root UKF) av den utvecklade algoritmen hartestats på sensorloggar, inspelade från ett Scania-testfordon. Utan GPS-avbrott presterar denimplementerade lokaliseringsalgoritmen inom en position RMSE på 60 cm.Robustheten hos lokaliseringsalgoritmen, vid GPS-avbrott, utvärderas genom att simulera0-90% längder av GPS-otillgänglighet under uppskattningsprocessen. Utöver det har deenskilda modulerna inom det föreslagna ramverket isolerats och analyserats med avseendepå deras bidrag till algoritmens lokaliseringsprestanda.Av allt visade sig onlineuppskattningen av IMU-sensorbiaser vara avgörande för att ökarobustheten hos den föreslagna lokaliseringsalgoritmen mot GPS-brist, särskilt för EKF. Närdet gäller distinktionen presterade både EKF och SRUKF med liknande förmåga, men UKFvisade bättre resultat vid längre perioder utan GPS-signal.

Curbs and road markings were designed to provide a visual low-level spatial perception of road environments. In this sense, a perception system capable of detecting those road features is of utmost importance for an autonomous vehicle. In vision-based approaches, few works have been developed for curb detection, and most of the advances on road marking detection have aimed lane markings only. Therefore, to detect all these road features, multiple algorithms running simultaneously would be necessary. Alternatively, as the main contribution of this work, it was proposed to employ an architecture of Fully Convolutional Neural Network (FCNN), denominated as 3CSeg-Multinet, to detect curbs and road markings in a single inference. Since there was no labeled dataset available for training and validation, a new one was generated with Brazilian urban scenes, and they were manually labeled. By visually analyzing experimental results, the proposed approach has shown to be effective and robust against most of the clutter present on images, running at around 10 fps in a Graphics Processing Unit (GPU). Moreover, with the intention of granting spatial perception, stereo vision techniques were used to project the detected road features in a point cloud. Finally, as a way to validate the applicability of the proposed perception system on a vehicle, it was also introduced a vision-based metric localization model for the urban scenario. In an experiment, compared to the ground truth, this localization method has revealed consistency on its pose estimations in a map generated by LIDAR. / Guias e sinalizações horizontais foram projetados para fornecer a percepção visual de baixo nível do espaço das vias urbanas. Deste modo, seria de extrema importância para um veículo autônomo ter um sistema de percepção capaz de detectar tais características visuais. Em abordagens baseadas em visão, poucos trabalhos foram desenvolvidos para detecção de guias, e a maioria dos avanços em detecção de sinalizações horizontais foi focada na detecção de faixas apenas. Portanto, para que fosse possível detectar todas essas características visuais, seria necessário executar diversos algoritmos simultaneamente. Alternativamente, como sendo a principal contribuição deste trabalho, foi proposto a adoção de uma Rede Neural Totalmente Convolutiva, denominado 3CSeg-Multinet, para detectar guias e sinalizações horizontais em apenas uma inferência. Como não havia um conjunto de dados rotulados disponível para treinar e validar a rede, foi gerado um novo conjunto com imagens capturadas em ambiente urbano brasileiro, e foi realizado a rotulação manual. Através de uma análise visual dos resultados experimentais obtidos, o método proposto mostrou-se eficaz e robusto contra a maioria dos fatores que causam confusão nas imagens, executando a aproximadamente 10 fps em uma GPU. Ainda, com o intuito de garantir a percepção espacial, foram usados métodos de visão estéreo para projetar as características detectadas em núvem de pontos. Finalmente, foi apresentado também um modelo de localização métrica baseado em visão para validar a aplicabilidade do sistema de percepção proposto em um veículo. Em um experimento, este método de localização revelou-se capaz de manter as estimativas consistentes com a verdadeira pose do veículo em um mapa gerado a partir de um sensor LIDAR.

Afin d’améliorer la précision des systèmes de navigation ainsi que de garantir la sécurité et la continuité du service, il est essentiel de connaitre la position et l’orientation du véhicule en tout temps. La localisation absolue utilisant des systèmes satellitaires tels que le GPS est souvent utilisée `a cette fin. Cependant, en environnement urbain, la localisation `a l’aide d’un récepteur GPS peut s’avérer peu précise voire même indisponible `a cause des phénomènes de réflexion des signaux, de multi-trajet ou de la faible visibilité satellitaire. Afin d’assurer une estimation précise et robuste du positionnement, d’autres capteurs et méthodes doivent compléter la mesure. Dans cette thèse, des méthodes de localisation de véhicules sont proposées afin d’améliorer l’estimation de la pose en prenant en compte la redondance et la complémentarité des informations du système multi-capteurs utilisé. Tout d’abord, les mesures GPS sont fusionnées avec des estimations de la localisation relative du véhicule obtenues `a l’aide d’un capteur proprioceptif (gyromètre), d’un système stéréoscopique(Odométrie visuelle) et d’un télémètre laser (recalage de scans télémétriques). Une étape de sélection des capteurs est intégrée pour valider la cohérence des observations provenant des différents capteurs. Seules les informations validées sont combinées dans un formalisme de couplage lâche avec un filtre informationnel. Si l’information GPS est indisponible pendant une longue période, la trajectoire estimée par uniquement les approches relatives tend `a diverger, en raison de l’accumulation de l’erreur. Pour ces raisons, les informations d’une carte numérique (route + bâtiment) ont été intégrées et couplées aux mesures télémétriques de deux télémètres laser montés sur le toit du véhicule (l’un horizontalement, l’autre verticalement). Les façades des immeubles détectées par les télémètres laser sont associées avec les informations_ bâtiment _ de la carte afin de corriger la position du véhicule.Les approches proposées sont testées et évaluées sur des données réelles. Les résultats expérimentaux obtenus montrent que la fusion du système stéréoscopique et du télémètre laser avec le GPS permet d’assurer le service de localisation lors des courtes absences de mesures GPS et de corriger les erreurs GPS de type saut. Par ailleurs, la prise en compte des informations de la carte numérique routière permet d’obtenir une approximation de la position du véhicule en projetant la position du véhicule sur le tronc¸on de route correspondant et enfin l’intégration de la carte numérique des bâtiments couplée aux données télémétriques permet d’affiner cette estimation, en particulier la position latérale. / In some dense urban environments (e.g., a street with tall buildings around), vehicle localization result provided by Global Positioning System (GPS) receiver might not be accurate or even unavailable due to signal reflection (multi-path) or poor satellite visibility. In order to improve the accuracy and robustness of assisted navigation systems so as to guarantee driving security and service continuity on road, a vehicle localization approach is presented in this thesis by taking use of the redundancy and complementarities of multiple sensors. At first, GPS localization method is complemented by onboard dead-reckoning (DR) method (inertial measurement unit, odometer, gyroscope), stereovision based visual odometry method, horizontal laser range finder (LRF) based scan alignment method, and a 2D GIS road network map based map-matching method to provide a coarse vehicle pose estimation. A sensor selection step is applied to validate the coherence of the observations from multiple sensors, only information provided by the validated sensors are combined under a loosely coupled probabilistic framework with an information filter. Then, if GPS receivers encounter long term outages, the accumulated localization error of DR-only method is proposed to be bounded by adding a GIS building map layer. Two onboard LRF systems (a horizontal LRF and a vertical LRF) are mounted on the roof of the vehicle and used to detect building facades in urban environment. The detected building facades are projected onto the 2D ground plane and associated with the GIS building map layer to correct the vehicle pose error, especially for the lateral error. The extracted facade landmarks from the vertical LRF scan are stored in a new GIS map layer. The proposed approach is tested and evaluated with real data sequences. Experimental results with real data show that fusion of the stereoscopic system and LRF can continue to localize the vehicle during GPS outages in short period and to correct the GPS positioning error such as GPS jumps; the road map can help to obtain an approximate estimation of the vehicle position by projecting the vehicle position on the corresponding road segment; and the integration of the building information can help to refine the initial pose estimation when GPS signals are lost for long time.

L'estimation de la pose (position et l'attitude) en temps réel est une fonction clé pour les véhicules autonomes routiers. Cette thèse vise à étudier des systèmes de localisation pour ces véhicules en utilisant des capteurs automobiles à faible coût. Trois types de capteurs sont considérés : des capteurs à l'estime qui existent déjà dans les automobiles modernes, des récepteurs GNSS mono-fréquence avec antenne patch et une caméra de détection de la voie regardant vers l’avant. Les cartes très précises sont également des composants clés pour la navigation des véhicules autonomes. Dans ce travail, une carte de marquage de voies avec une précision de l’ordre du décimètre est considérée. Le problème de la localisation est étudié dans un repère de travail local Est-Nord-Haut. En effet, les sorties du système de localisation sont utilisées en temps réel comme entrées dans un planificateur de trajectoire et un contrôleur de mouvement pour faire en sorte qu’un véhicule soit capable d'évoluer au volant de façon autonome à faible vitesse avec personne à bord. Ceci permet de développer des applications de voiturier autonome aussi appelées « valet de parking ». L'utilisation d'une caméra de détection de voie rend possible l’exploitation des informations de marquage de voie stockées dans une carte géoréférencée. Un module de détection de marquage détecte la voie hôte du véhicule et fournit la distance latérale entre le marquage de voie détecté et le véhicule. La caméra est également capable d'identifier le type des marquages détectés au sol (par exemple, de type continu ou pointillé). Comme la caméra donne des mesures relatives, une étape importante consiste à relier les mesures à l'état du véhicule. Un modèle d'observation raffiné de la caméra est proposé. Il exprime les mesures métriques de la caméra en fonction du vecteur d'état du véhicule et des paramètres des marquages au sol détectés. Cependant, l'utilisation seule d'une caméra a des limites. Par exemple, les marquages des voies peuvent être absents dans certaines parties de la zone de navigation et la caméra ne parvient pas toujours à détecter les marquages au sol, en particulier, dans les zones d’intersection. Un récepteur GNSS, qui est obligatoire pour le démarrage à froid, peut également être utilisé en continu dans le système de localisation multi-capteur du fait qu’il permet de compenser la dérive de l’estime. Les erreurs de positionnement GNSS ne peuvent pas être modélisées simplement comme des bruits blancs, en particulier avec des récepteurs mono-fréquence à faible coût travaillant de manière autonome, en raison des perturbations atmosphériques sur les signaux des satellites et les erreurs d’orbites. Un récepteur GNSS peut également être affecté par de fortes perturbations locales qui sont principalement dues aux multi-trajets. Cette thèse étudie des modèles formeurs de biais d’erreur GNSS qui sont utilisés dans le solveur de localisation en augmentant le vecteur d'état. Une variation brutale due à multi-trajet est considérée comme une valeur aberrante qui doit être rejetée par le filtre. Selon le flux d'informations entre le récepteur GNSS et les autres composants du système de localisation, les architectures de fusion de données sont communément appelées « couplage lâche » (positions et vitesses GNSS) ou « couplage serré » (pseudo-distance et Doppler sur les satellites en vue). Cette thèse étudie les deux approches. En particulier, une approche invariante selon la route est proposée pour gérer une modélisation raffinée de l'erreur GNSS dans l'approche par couplage lâche puisque la caméra ne peut améliorer la performance de localisation que dans la direction latérale de la route. / Estimating the pose (position and attitude) in real-time is a key function for road autonomous vehicles. This thesis aims at studying vehicle localization performance using low cost automotive sensors. Three kinds of sensors are considered : dead reckoning (DR) sensors that already exist in modern vehicles, mono-frequency GNSS (Global navigation satellite system) receivers with patch antennas and a frontlooking lane detection camera. Highly accurate maps enhanced with road features are also key components for autonomous vehicle navigation. In this work, a lane marking map with decimeter-level accuracy is considered. The localization problem is studied in a local East-North-Up (ENU) working frame. Indeed, the localization outputs are used in real-time as inputs to a path planner and a motion generator to make a valet vehicle able to drive autonomously at low speed with nobody on-board the car. The use of a lane detection camera makes possible to exploit lane marking information stored in the georeferenced map. A lane marking detection module detects the vehicle’s host lane and provides the lateral distance between the detected lane marking and the vehicle. The camera is also able to identify the type of the detected lane markings (e.g., solid or dashed). Since the camera gives relative measurements, the important step is to link the measures with the vehicle’s state. A refined camera observation model is proposed. It expresses the camera metric measurements as a function of the vehicle’s state vector and the parameters of the detected lane markings. However, the use of a camera alone has some limitations. For example, lane markings can be missing in some parts of the navigation area and the camera sometimes fails to detect the lane markings in particular at cross-roads. GNSS, which is mandatory for cold start initialization, can be used also continuously in the multi-sensor localization system as done often when GNSS compensates for the DR drift. GNSS positioning errors can’t be modeled as white noises in particular with low cost mono-frequency receivers working in a standalone way, due to the unknown delays when the satellites signals cross the atmosphere and real-time satellites orbits errors. GNSS can also be affected by strong biases which are mainly due to multipath effect. This thesis studies GNSS biases shaping models that are used in the localization solver by augmenting the state vector. An abrupt bias due to multipath is seen as an outlier that has to be rejected by the filter. Depending on the information flows between the GNSS receiver and the other components of the localization system, data-fusion architectures are commonly referred to as loosely coupled (GNSS fixes and velocities) and tightly coupled (raw pseudoranges and Dopplers for the satellites in view). This thesis investigates both approaches. In particular, a road-invariant approach is proposed to handle a refined modeling of the GNSS error in the loosely coupled approach since the camera can only improve the localization performance in the lateral direction of the road. Finally, this research discusses some map-matching issues for instance when the uncertainty domain of the vehicle state becomes large if the camera is blind. It is challenging in this case to distinguish between different lanes when the camera retrieves lane marking measurements.As many outdoor experiments have been carried out with equipped vehicles, every problem addressed in this thesis is evaluated with real data. The different studied approaches that perform the data fusion of DR, GNSS, camera and lane marking map are compared and several conclusions are drawn on the fusion architecture choice.

In some dense urban environments (e.g., a street with tall buildings around), vehicle localization result provided by Global Positioning System (GPS) receiver might not be accurate or even unavailable due to signal reflection (multi-path) or poor satellite visibility. In order to improve the accuracy and robustness of assisted navigation systems so as to guarantee driving security and service continuity on road, a vehicle localization approach is presented in this thesis by taking use of the redundancy and complementarities of multiple sensors. At first, GPS localization method is complemented by onboard dead-reckoning (DR) method (inertial measurement unit, odometer, gyroscope), stereovision based visual odometry method, horizontal laser range finder (LRF) based scan alignment method, and a 2D GIS road network map based map-matching method to provide a coarse vehicle pose estimation. A sensor selection step is applied to validate the coherence of the observations from multiple sensors, only information provided by the validated sensors are combined under a loosely coupled probabilistic framework with an information filter. Then, if GPS receivers encounter long term outages, the accumulated localization error of DR-only method is proposed to be bounded by adding a GIS building map layer. Two onboard LRF systems (a horizontal LRF and a vertical LRF) are mounted on the roof of the vehicle and used to detect building facades in urban environment. The detected building facades are projected onto the 2D ground plane and associated with the GIS building map layer to correct the vehicle pose error, especially for the lateral error. The extracted facade landmarks from the vertical LRF scan are stored in a new GIS map layer. The proposed approach is tested and evaluated with real data sequences. Experimental results with real data show that fusion of the stereoscopic system and LRF can continue to localize the vehicle during GPS outages in short period and to correct the GPS positioning error such as GPS jumps; the road map can help to obtain an approximate estimation of the vehicle position by projecting the vehicle position on the corresponding road segment; and the integration of the building information can help to refine the initial pose estimation when GPS signals are lost for long time.