Dans le cadre des simulations interactives, le manque de modèles géométriques reste une des limitations majeurs des simulateurs. Actuellement, les simulateurs commerciaux ne propose pas ou un tout cas, un nombre limité de cas. Un grand nombre des travaux abordent cependant ce sujet tout au long de ces deux dernières décennies. Malgré une vaste littérature, les méthodes ne sont pas adaptées à un contexte interactive, plus particulièrement quand il s'agit des réseaux vasculaires. Dans cette thèse, nous considérons le problème de la segmentation et la reconstruction des vaisseaux sanguins à partir de données patients en 3DRA. Pour ce faire, nous proposons deux nouveaux algorithmes, un pour la segmentation et un autre, pour la reconstruction. Tout d'abord, le réseau vasculaire est construit grâce à un algorithme de suivi de la ligne centrale des vaisseaux. De plus, notre procédure de suivi extrait des point à la surface des vaisseaux de manière robuste. Deuxièmement, ces points sont estimés par une surface implicite (un blobby model) qui est raffinée de façon itérative. Les résultats du suivi et de la reconstruction sont produit à partir de données synthétiques et réelles. Lors de la simulation de la navigation d'outils interventionnels, notre modèle géométrique remplit les exigences des simulations interactives : une prédiction et détection rapide des collisions, l'accès à l'information topologique, une surface lisse et la mise à disposition de quantités différentielles pour la résolution des contacts. / In the context of interactive simulation, the lack of patient specific geometrical models remains one of the major limitations of simulators. Current commercial simulators proposed no or a limited number of cases. However, a vast literature on the subject has been introduced in the past twenty years. Nevertheless, the proposed methods are not adapted to an interactive context, especially when dealing with vascular networks. In this work, we address the problem of blood vessel segmentation and reconstruction from 3DRA patient data. To this end, we propose two novel algorithms for segmentation and reconstruction. First, the vessel tree is built by tracking the vessel centerline. Our dedicated tracking process also extracts points on the vessel surface in a robust way. Second, those points are fitted by an implicit surface (a blobby model) that is iteratively refined. Tracking and reconstruction results are reported on synthetic and patient data. Simulations within an interventional tool navigation context showed that the resulting geometrical model complies with interactive simulation requirements : fast collision detection and prediction, topology information, smoothness and availability of differential quantities for contact response computation.

Mobile Road Mapping Systems (MRMS) are the current solution to the growing demand for high definition road surface maps in wide ranging applications from pavement management to autonomous vehicle testing. The focus of this research work is to improve the accuracy of MRMS by using the principles of Simultaneous Localization and Mapping (SLAM). First a framework for describing the sensor measurement models in MRMS is developed. Next the problem of estimating the road surface from the set of sensor measurements is formulated as a SLAM problem and two approaches are proposed to solve the formulated problem. The first is an incremental solution wherein sensor measurements are processed in sequence using an Extended Kalman Filter (EKF). The second is a post-processing solution wherein the SLAM problem is formulated as an inference problem over a factor graph and existing factor graph SLAM techniques are used to solve the problem. For the mobile road mapping problem, the road surface being measured is one the primary inputs to the dynamics of the MRMS. Hence, concurrent to the main objective this work also investigates the use of the dynamics of the host vehicle of the system to improve the accuracy of the MRMS. Finally a novel method that builds off the concepts of the popular model fitting algorithm, Random Sampling and Consensus (RANSAC), is developed in order to identify outliers in road surface measurements and estimate the road elevations at grid nodes using these measurements. The developed methods are validated in a simulated environment and the results demonstrate a significant improvement in the accuracy of MRMS over current state-of-the art methods. / Doctor of Philosophy / Mobile Road Mapping Systems (MRMS) are the current solution to the growing demand for high definition road surface maps in wide ranging applications from pavement management to autonomous vehicle testing. The objective of this research work is to improve the accuracy of MRMS by investigating methods to improve the sensor data fusion process. The main focus of this work is to apply the principles from the field of Simultaneous Localization and Mapping (SLAM) in order to improve the accuracy of MRMS. The concept of SLAM has been successfully applied to the field of mobile robot navigation and thus the motivation of this work is to investigate its application to the problem of mobile road mapping. For the mobile road mapping problem, the road surface being measured is one the primary inputs to the dynamics of the MRMS. Hence this work also investigates whether knowledge regarding the dynamics of the system can be used to improve the accuracy. Also developed as part of this work is a novel method for identifying outliers in road surface datasets and estimating elevations at road surface grid nodes. The developed methods are validated in a simulated environment and the results demonstrate a significant improvement in the accuracy of MRMS over current state-of-the-art methods.

The usage of Unmanned Aerial Vehicles (UAV) for several applications has in-creased during the past years. One of the possible applications are aerial imagecapturing for detection and surveillance purposes. In order to make the captur-ing process more efficient, multiple, cameraequipped UAV:s could fly in a for-mation and as a result cover a larger area. To be able to receive several imagesequences and stitch those together, resulting in a panoramavideo, a softwareapplication has been developed and tested for this purpose.All functionality are developed in the language C++ by using the software li-brary OpenCV. All implementations of different techniques and methods hasbeen done as generic as possible to be able to add functionality in the future.Common methods in computervision and object recognition such as SIFT, SURF and RANSAC have been tested. / Användningen av UAV:er för olika tillämpningar har ökat under senare år. Ett avmöjliga användningsområden är flygfotografering och övervakning. För att gö-ra bildupptagningen mer effektiv kan flera UAV:er flyga i formation och på så viskunna fotografera ett avsevärt större område. För att kunna ta in flera bildsekven-ser och foga samman dessa till en panoramavideo, har ett program utvecklats ochtestats för denna uppgift.All funktionalitet för inläsning av bilder och video har utvecklats i C++ medprogrambiblioteket OpenCV. Implementeringen av dessa tekniker och metoderhar gjort så generiskt som möjligt för att det ska vara lättare att lägga till and-ra tekniker och utöka programmets funktioner. Olika tekniker som har testatsinkluderar: SIFT, SURF och RANSAC

Dans le cadre des systèmes de positionnement par satellite GNSS (« Global Navigation Satellite Systems »), l’intégritéde la navigation d’un utilisateur est gérée en réception par la détection, l’identification voire l’exclusion de mesures depseudo-distance jugées erronées. Généralement basés sur le concept a posteriori RAIM (« Receiver Autonomous IntegrityMonitoring »), les algorithmes de contrôle autonome d’intégrité fournissent de hautes performances pour l’aviation civile,dont le contexte de navigation est caractérisé par une forte visibilité des satellites et peu de signaux parasites captéspar l’antenne réceptrice. L’algorithme WLSR RAIM est communément utilisé dans ce cadre. Néanmoins, les techniquesRAIM ne sont pas compatibles avec la navigation terrestre en milieu contraint. En effet, le contexte urbain est notammentcaractérisé par un masquage récurrent des signaux satellitaires directs ainsi que la réception de multi-trajets générés parl’environnement proche du récepteur. RAIM ne prend pas en compte l’ensemble des données disponibles en réception,dégradant ainsi fortement ses performances. Il est donc nécessaire de développer des méthodes de contrôle d’intégritécompatibles avec un tel contexte de navigation. Pour cela, la thèse propose d’étudier l’apport d’informations GNSS a priorinon utilisées par les techniques RAIM. Deux paramètres principaux ont été exploités : le signal GNSS brut reçu et lesestimations de directions d’arrivée des signaux satellitaires DOA (« Direction Of Arrival »). La première étape a consisté à implémenter une méthode a priori qui évalue la cohérence du positionnement estimé par rapport au signal brut directement reçu. Cette méthode a été nommée Direct-RAIM (D-RAIM) et a démontré une forte sensibilité de détection, permettant d’anticiper d’éventuels risques sur la navigation et de caractériser plus finement la qualité de l’environnement proche du récepteur. Toutefois, le caractère a priori de l’approche engendre de potentielles non détection d’erreurs en cas de modèle de signal défectueux. Afin de contourner cette limitation, un couplage WLSRRAIM – D-RAIM a été développé, nommé Hybrid-RAIM (H-RAIM). Une telle approche permet de combiner robustesse etsensibilité apportées par ces techniques respectives. Le second axe de recherche a mis en évidence la contribution de l’information des DOA dans un contrôle autonome d’intégrité. L’intégration d’un réseau d’antennes en réception permet d’obtenir l’estimation des DOA pour l’ensemble dela constellation visible. Théoriquement, l’évolution jointe des DOA est directement liée à l’attitude du réseau. Cet aspectpermet donc de détecter toute incohérence sur une ou plusieurs voies en cas d’estimation(s) de DOA biaisée(s), par rapportà l’ensemble de la constellation. L’algorithme RANSAC (« RANdom SAmple Consensus») a été utilisé afin de détecter toutcomportement aberrant dans l’estimation des DOA, et ainsi mesurer la confiance que l’utilisateur peut placer dans chaquevoie. L’algorithme WLSR RAIM RANSAC a ainsi été implémenté. L’intégration de la composante DOA permet d’ajouterun degré de liberté dans le contrôle autonome d’intégrité côté récepteur et ainsi d’affiner la détection voire l’exclusiond’erreurs. Au cours de cette thèse, un récepteur logiciel a été implémenté, permettant de traiter des signaux Galileo, de lagénération du signal jusqu’au positionnement puis au contrôle d’intégrité. Ce récepteur a pu être évalué à partir de donnéessimulées en environnement urbain. / In Global Navigation Satellite Systems (GNSS) applications, integrity is managed at the reception side by detection,identification and exclusion of faulty pseudorange measurements. Usually based on the a posteriori Receiver AutonomousIntegrity Monitoring (RAIM) concept, integrity techniques provide high performances for civil aviation, with a navigationcontext defined by a clear-sky environment. WLSR RAIM is commonly used. Nevertheless, RAIM techniques are notcompatible with a terrestrial navigation in harsh environments. For instance, urban areas are characterized by a poorvisibility and the reception of many multipaths derived from the receiver closed-environment. RAIM does not consider allthe available data in the reception chain, which dramatically deteriorates the detection performances. Hence, it is necessaryto develop integrity process compatible with such a navigation context. This PhD work studies the contribution of GNSSa priori information, disused by conventional RAIM techniques. Two main parameters have been exploited : the receivedraw GNSS signal and the Directions Of Arrival (DOA) estimations.This first step was devoted to the development of an a priori method which evaluates the consistence of the estimatedPosition Velocity Time (PVT) vector of the receiver with respect to the raw GNSS signal. This method has been calledDirect-RAIM (D-RAIM) and has shown high detection sensitivity, allowing the user to anticipate navigation risks and todefine precisely the quality of the receiver closed-environment. However, the a priori aspect of this approach may lead tonavigation error missed detections if the signal model is getting flawed. In order to circumvent this limitation, a WLSRRAIM – D-RAIM coupling has been developed, called Hybrid-RAIM (H-RAIM). Such an approach merges the robustnessand the sensitivity brought by both techniques.The second research step has brought to light the contribution of the DOA information in an autonomous integritymonitoring. Using an antenna array, the user can get the DOA estimations for all satellites in view. Theoretically, the DOAjoint evolution is directly correlated with the array rotation angles. Hence, any mismatch on the DOA estimations withrespect to the global constellation can be detected. RANdom Sample Consensus (RANSAC) algorithm has been used inorder to detect any faulty DOA evolution, derived from inconsistencies in reception linked to potential navigation risks :RANSAC measures the trust that the user can place in each channel. Therefore, a WLSR RAIM RANSAC algorithmhas been developed. The integration of the DOA component adds a degree of freedom in receiver autonomous integritymonitoring, refining the error detection and exclusion.Last but not least, a software receiver has been implemented processing Galileo data, from the signal generation to positioningand integrity monitoring. This software has been evaluated by simulated data characterizing urban environments.

As the market for unmanned air vehicles (UAVs) rapidly expands, the need for algorithmsthat improve the capabilities of those vehicles is also growing. One valuable capability for UAVsis that of persistent tracking—the ability to find and track another moving object, usually on theground, from an aerial platform. This thesis presents a method for tracking multiple ground targetsfrom an airborne camera. Moving objects on the ground are detected by using frame-to-frameregistration. The detected objects are then tracked using the newly developed recursive RANSACalgorithm. Much video tracking work has focused on using appearance-based processing for tracking,with some approaches using dynamic trackers such as Kalman filters. This work demonstratesa fusion of computer vision and dynamic tracking to increase the ability of an unmanned air platformto identify and robustly track moving targets. With a C++ implementation of the algorithmsrunning on the open source Robot Operating System (ROS) framework, the system developed iscapable of processing 1920x1080 resolution video at over seven frames per second on a desktopcomputer.

Recent miniaturization of computer hardware, MEMs sensors, and high energy density batteries have enabled highly capable mobile robots to become available at low cost. This has driven the rapid expansion of interest in multi-rotor unmanned aerial vehicles. Another area which has expanded simultaneously is small powerful computers, in the form of smartphones, which nearly always have a camera attached, many of which now contain a OpenCL compatible graphics processing units. By combining the results of those two developments a low-cost multi-rotor UAV can be produced with a low-power onboard computer capable of real-time computer vision. The system should also use general purpose computer vision software to facilitate a variety of experiments. To demonstrate this I have built a quadrotor UAV based on control hardware from the Pixhawk project, and paired it with an ARM based single board computer, similar those in high-end smartphones. The quadrotor weights 980 g and has a flight time of 10 minutes. The onboard computer capable of running a pose estimation algorithm above the 10 Hz requirement for stable visual control of a quadrotor. A feature tracking algorithm was developed for efficient pose estimation, which relaxed the requirement for outlier rejection during matching. Compared with a RANSAC- only algorithm the pose estimates were less variable with a Z-axis standard deviation 0.2 cm compared with 2.4 cm for RANSAC. Processing time per frame was also faster with tracking, with 95 % confidence that tracking would process the frame within 50 ms, while for RANSAC the 95 % confidence time was 73 ms. The onboard computer ran the algorithm with a total system load of less than 25 %. All computer vision software uses the OpenCV library for common computer vision algorithms, fulfilling the requirement for running general purpose software. The tracking algorithm was used to demonstrate the capability of the system by per- forming visual servoing of the quadrotor (after manual takeoff). Response to external perturbations was poor however, requiring manual intervention to avoid crashing. This was due to poor visual controller tuning, and to variations in image acquisition and attitude estimate timing due to using free running image acquisition. The system, and the tracking algorithm, serve as proof of concept that visual control of a quadrotor is possible using small low-power computers and general purpose computer vision software.

Reliable track continuity is an important characteristic of multiple target tracking (MTT) algorithms. In the specific case of visually tracking multiple ground targets from an aerial platform, challenges arise due to realistic operating environments such as video compression artifacts, unmodeled camera vibration, and general imperfections in the target detection algorithm. Some popular visual detection techniques include Kanade-Lucas-Tomasi (KLT)-based motion detection, difference imaging, and object feature matching. Each of these algorithmic detectors has fundamental limitations in regard to providing consistent measurements. In this thesis we present a scalable detection framework that simultaneously leverages multiple measurement sources. We present the recursive random sample consensus (R-RANSAC) algorithm in a data fusion architecture that accommodates multiple measurement sources. Robust track continuity and real-time performance are demonstrated with post-processed flight data and a hardware demonstration in which the aircraft performs automated target following. Applications involving autonomous tracking of ground targets occasionally encounter situations where semantic information about targets would improve performance. This thesis also presents an autonomous target labeling framework that leverages cloud-based image classification services to classify targets that are tracked by the R-RANSAC MTT algorithm. The communication is managed by a Python robot operating system (ROS) node that accounts for latency and filters the results over time. This thesis articulates the feasibility of this approach and suggests hardware improvements that would yield reliable results. Finally, this thesis presents a framework for image-based target recognition to address the problem of tracking targets that become occluded for extended periods of time. This is done by collecting descriptors of targets tracked by R-RANSAC. Before new tracks are assigned an ID, an attempt to match visual information with historical tracks is triggered. The concept is demonstrated in a simulation environment with a single target, using template-based target descriptors. This contribution provides a framework for improving track reliability when faced with target occlusions.

The use of unmanned systems is becoming widespread in commercial and military sectors. The ability of these systems to take on dull, dirty, and dangerous tasks which were formerly done by humans is encouraging their rapid adoption. In particular, a subset of these undesirable tasks are uniquely suited for small unmanned aerial vehicles such as quadrotor helicopters. Examples of such tasks include surveillance, mapping, and search and rescue. Many of these potential tasks require quadrotors to be deployed in environments where a degree of position estimation is required and traditional GPS-based positioning technologies are not applicable. Likewise, since unmanned systems in these environments are often intended to serve the purpose of scouts or first--responders, no maps or reference beacons will be available. Additionally, there is no guarantee of clear features within the environment which an onboard sensor suite (typically made up of a monocular camera and inertial sensors) will be able to track to maintain an estimate of vehicle position. Up to 90% of the features detected in the environment may produce motion estimates which are inconsistent with the true vehicle motion. Thus, new methods are needed to compensate for these environmental deficiencies and measurement inconsistencies. In this work, a RANSAC-based outlier rejection technique is combined with an Extended Kalman Filter (EKF) to generate estimates of vehicle position in a 2--D plane. A low complexity feature selection technique is used in place of more modern techniques in order to further reduce processor load. The overall algorithm was faster than the traditional approach by a factor of 4. Outlier rejection allows the abundance of low quality, poorly tracked image features to be filtered appropriately, while the EKF allows a motion model of the quadrotor to be incorporated into the position estimate. The algorithm is tested in real-time on a quadrotor vehicle in an indoor environment with no clear features and found to be able to successfully estimate position of the vehicle to within 40 cm, superior to those produced when no outlier rejection technique was used. It is also found that the choice of simple feature selection approaches is valid, as complex feature selection approaches which may take over 10 times as long to run still result in outliers being present. When the algorithm is used for vehicle control, periodic synchronization to ground truth data was required due to nearly 1 second of latency present in the closed--loop system. However, the system as a whole is a valid proof of concept for the use of low quality images for quadrotor position control. The overall results from the work suggest that it is possible for unmanned systems to use visual data to estimate state even in operational environments which are poorly suited for visual estimation techniques. The filter algorithm described in this work can be seen as a useful tool for expanding the operational capabilities of small aerial vehicles.

In this paper, an object recognition system has been developed that uses local image features. In the system, multiple classes of objects can be recognized in an image. This system is basically divided into two parts: object detection and object identification. Object detection is based on SIFT features, which are invariant to image illumination, scaling and rotation. SIFT features extracted from a test image are used to perform a reliable matching between a database of SIFT features from known object images. Method of DBSCAN clustering is used for multiple object detection. RANSAC method is used for decreasing the amount of false detection. Object identification is based on 'Bag-of-Words' model. The 'BoW' model is a method based on vector quantization of SIFT descriptors of image patches. In this model, K-means clustering and Support Vector Machine (SVM) classification method are applied.

The use of unmanned systems is becoming widespread in commercial and military sectors. The ability of these systems to take on dull, dirty, and dangerous tasks which were formerly done by humans is encouraging their rapid adoption. In particular, a subset of these undesirable tasks are uniquely suited for small unmanned aerial vehicles such as quadrotor helicopters. Examples of such tasks include surveillance, mapping, and search and rescue. Many of these potential tasks require quadrotors to be deployed in environments where a degree of position estimation is required and traditional GPS-based positioning technologies are not applicable. Likewise, since unmanned systems in these environments are often intended to serve the purpose of scouts or first--responders, no maps or reference beacons will be available. Additionally, there is no guarantee of clear features within the environment which an onboard sensor suite (typically made up of a monocular camera and inertial sensors) will be able to track to maintain an estimate of vehicle position. Up to 90% of the features detected in the environment may produce motion estimates which are inconsistent with the true vehicle motion. Thus, new methods are needed to compensate for these environmental deficiencies and measurement inconsistencies. In this work, a RANSAC-based outlier rejection technique is combined with an Extended Kalman Filter (EKF) to generate estimates of vehicle position in a 2--D plane. A low complexity feature selection technique is used in place of more modern techniques in order to further reduce processor load. The overall algorithm was faster than the traditional approach by a factor of 4. Outlier rejection allows the abundance of low quality, poorly tracked image features to be filtered appropriately, while the EKF allows a motion model of the quadrotor to be incorporated into the position estimate. The algorithm is tested in real-time on a quadrotor vehicle in an indoor environment with no clear features and found to be able to successfully estimate position of the vehicle to within 40 cm, superior to those produced when no outlier rejection technique was used. It is also found that the choice of simple feature selection approaches is valid, as complex feature selection approaches which may take over 10 times as long to run still result in outliers being present. When the algorithm is used for vehicle control, periodic synchronization to ground truth data was required due to nearly 1 second of latency present in the closed--loop system. However, the system as a whole is a valid proof of concept for the use of low quality images for quadrotor position control. The overall results from the work suggest that it is possible for unmanned systems to use visual data to estimate state even in operational environments which are poorly suited for visual estimation techniques. The filter algorithm described in this work can be seen as a useful tool for expanding the operational capabilities of small aerial vehicles.