With the aid of modern sensors it is possible to create models of buildings. These sensorstypically generate 3D point clouds and in order to increase interpretability and usability,these point clouds are often translated into 3D models.In this thesis a way of translating a 3D point cloud into a 3D model is presented. The basicfunctionality is implemented using Matlab. The geometric model consists of floors, wallsand ceilings. In addition, doors and windows are automatically identified and integrated intothe model. The resulting model also has an explicit representation of the topology betweenentities of the model. The topology is represented as a graph, and to do this GraphML isused. The graph is opened in a graph editing program called yEd.The result is a 3D model that can be plotted in Matlab and a graph describing the connectivitybetween entities. The GraphML file is automatically generated in Matlab. An interfacebetween Matlab and yEd allows the user to choose which rooms should be plotted.

The evolution of technology has led the research into new emerging wearable devices such as the Smart Glasses. This technology provides with new visualization techniques. Augmented Reality is an advanced technology that could significantly ease the execution of much complex operations. Augmented Reality is a combination of both Virtual and Actual Reality, making accessible to the user new tools to safeguard in the transfer of knowledge in several environments and for several processes. This thesis explores the development of an android based image recognition application. The feature point detectors and descriptors are used as they can deal great with the correspondence problems. The selection of best image recognition technique on the smart glasses is chosen based on the time taken to retrieve the results and the amount of power consumed in the process. As the smart glasses are equipped with the limited resources, the selected approach should use low computation on it by making the device operations uninterruptable. The effective and efficient method for detection and recognition of the safety signs from images is selected. The ubiquitous SIFT and SURF feature detectors consume more time and are computationally complex and require very high-level hardware components for processing. The binary descriptors are taken into account as they are light weight and can support low power devices in a much effective style. A comparative analysis is being done on the working of binary descriptors like BRIEF, ORB, AKAZE, FREAK, etc., on the smart glasses based on their performance and the requirements. ORB is the most efficient among the binary descriptors and has been more effective for the smart glasses in terms of time measurements and low power consumption.

Modern day cars are often equipped with a variety of sensors that collect information about the car and its surroundings. The stereo camera is an example of a sensor that in addition to regular images also provides distances to points in its environment. This information can, for example, be used for detecting approaching obstacles and warn the driver if a collision is imminent or even automatically brake the vehicle. Objects that constitute a potential danger are usually located on the road in front of the vehicle which makes the road surface a suitable reference level from which to measure the object's heights. This Master's thesis describes how an estimate of the road surface can be found to in order to make these height measurements. The thesis describes how the large amount of data generated by the stereo camera can be scaled down to a more effective representation in the form of an elevation map. The report discusses a method for relating data from different instances in time using information from the vehicle's motion sensors and shows how this method can be used for temporal filtering of the elevation map. For estimating the road surface two different methods are compared, one that uses a RANSAC-approach to iterate for a good surface model fit and one that uses conditional random fields for modeling the probability of different parts of the elevation map to be part of the road. A way to detect curb lines and how to use them to improve the road surface estimate is shown. Both methods for road classification show good results with a few differences that are discussed towards the end of the report. An example of how the road surface estimate can be used to detect obstacles is also included.

L'algorithme RANSAC est l'approche la plus commune pour l'estimation robuste des paramètres d'un modèle en vision par ordinateur. C'est principalement sa capacité à traiter des données contenant potentiellement plus d'erreurs que d'information utile qui fait son succès dans ce domaine où les capteurs fournissent une information très riche mais très difficilement exploitable. Depuis sa création, il y a trente ans, de nombreuses modifications ont été proposées pour améliorer sa vitesse, sa précision ou sa robustesse. Dans ce travail, nous proposons d'accélérer la résolution d'un problème par RANSAC en utilisant plus d'information que les approches habituelles. Cette information, calculée à partir des données elles-même ou provenant de sources complémentaires de tous types, nous permet d'aider RANSAC à générer des hypothèses plus pertinentes. Pour ce faire, nous proposons de distinguer quatre degrés de qualité d'une hypothèse: la "non contamination", la "cohésion", la "cohérence" et enfin la "pertinence". Puis nous montrons à quel point une hypothèse non contaminée par des données erronées est loin d'être pertinente dans le cas général. Dès lors, nous nous attachons à concevoir un algorithme original qui, contrairement aux méthodes de l'état de l'art, se focalise sur la génération d'échantillons "pertinents" plutôt que simplement "non contaminés". Notre approche consiste à commencer par proposer un modèle probabiliste unifiant l'ensemble des méthodes de réordonnancement de l'échantillonnage de RANSAC. Ces méthodes assurent un guidage du tirage aléatoire des données tout en se prémunissant d'une mise en échec de RANSAC. Puis, nous proposons notre propre algorithme d'ordonnancement, BetaSAC, basé sur des tris conditionnels partiels. Nous montrons que la conditionnalité du tri permet de satisfaire des contraintes de cohérence des échantillons formés, menant à une génération d'échantillons pertinents dans les premières itérations de RANSAC, et donc à une résolution rapide du problème. L'utilisation de tris partiels plutôt qu'exhaustifs, quant à lui, assure la rapidité et la randomisation, indispensable à ce type de méthodes. Dans un second temps, nous proposons une version optimale de notre méthode, que l'on appelle OABSAC (pour Optimal and Adaptative BetaSAC), faisant intervenir une phase d'apprentissage hors ligne. Cet apprentissage a pour but de mesurer les propriétés caractéristiques du problème spécifique que l'on souhaite résoudre, de façon à établir automatiquement le paramétrage optimal de notre algorithme. Ce paramétrage est celui qui doit mener à une estimation suffisamment précise des paramètres du modèle recherché en un temps (en secondes) le plus court. Les deux méthodes proposées sont des solutions très générales qui permettent d'intégrer dans RANSAC tout type d'information complémentaire utile à la résolution du problème. Nous montrons l'avantage de ces méthodes pour le problème de l'estimation d'homographies et de géométries épipolaires entre deux photographies d'une même scène. Les gains en vitesse de résolution du problème peuvent atteindre un facteur cent par rapport à l'algorithme RANSAC classique.

Ένα σημαντικό πρόβλημα είναι η αντιστοίχιση εικόνων με σκοπό τη δημιουργία πανοράματος. Στην παρούσα εργασία έχουν χρησιμοποιηθεί αλγόριθμοι που βασίζονται στη χρήση σημείων κλειδιών. Αρχικά στην εργασία βρίσκονται σημεία κλειδιά για κάθε εικόνα που μένουν ανεπηρέαστα από τις αναμενόμενες παραμορφώσεις με την βοήθεια του αλγορίθμου SIFT (Scale Invariant Feature Transform). Έχοντας τελειώσει αυτή τη διαδικασία για όλες τις εικόνες, προσπαθούμε να βρούμε το πρώτο ζευγάρι εικόνων που θα ενωθεί. Για να δούμε αν δύο εικόνες μπορούν να ενωθούν, ακολουθεί ταίριασμα των σημείων κλειδιών τους. Όταν ένα αρχικό σετ αντίστοιχων χαρακτηριστικών έχει υπολογιστεί, πρέπει να βρεθεί ένα σετ που θα παράγει υψηλής ακρίβειας αντιστοίχιση. Αυτό το πετυχαίνουμε με τον αλγόριθμο RANSAC, μέσω του οποίου βρίσκουμε το γεωμετρικό μετασχηματισμό ανάμεσα στις δύο εικόνες, ομογραφία στην περίπτωσή μας. Αν ο αριθμός των κοινών σημείων κλειδιών είναι επαρκής, δηλαδή ταιριάζουν οι εικόνες, ακολουθεί η ένωσή τους. Αν απλώς ενώσουμε τις εικόνες, τότε θα έχουμε σίγουρα κάποια προβλήματα, όπως το ότι οι ενώσεις των δύο εικόνων θα είναι πολύ εμφανείς. Γι’ αυτό, για την εξάλειψη αυτού του προβλήματος, χρησιμοποιούμε τη μέθοδο των Λαπλασιανών πυραμίδων. Επαναλαμβάνεται η παραπάνω διαδικασία μέχρι να δημιουργηθεί το τελικό πανόραμα παίρνοντας κάθε φορά σαν αρχική την τελευταία εικόνα που φτιάξαμε στην προηγούμενη φάση. / Stitching multiple images together to create high resolution panoramas is one of the most popular consumer applications of image registration and blending. At this work, feature-based registration algorithms have been used. The first step is to extract distinctive invariant features from every image which are invariant to image scale and rotation, using SIFT (Scale Invariant Feature Transform) algorithm. After that, we try to find the first pair of images in order to stitch them. To check if two images can be stitched, we match their keypoints (the results from SIFT). Once an initial set of feature correspondences has been computed, we need to find the set that is will produce a high-accuracy alignment. The solution at this problem is RANdom Sample Consensus (RANSAC). Using this algorithm (RANSAC) we find the motion model between the two images (homography). If there is enough number of correspond points, we stitch these images. After that, seams are visible. As solution to this problem is used the method of Laplacian Pyramids. We repeat the above procedure using as initial image the ex panorama which has been created.

The evolution of technology has led the research into new emerging wearable devices such as the Smart Glasses. This technology provides with new visualization techniques. Augmented Reality is an advanced technology that could significantly ease the execution of much complex operations. Augmented Reality is a combination of both Virtual and Actual Reality, making accessible to the user new tools to safeguard in the transfer of knowledge in several environments and for several processes. This thesis explores the development of an android based image recognition application. The feature point detectors and descriptors are used as they can deal great with the correspondence problems. The selection of best image recognition technique on the smart glasses is chosen based on the time taken to retrieve the results and the amount of power consumed in the process. As the smart glasses are equipped with the limited resources, the selected approach should use low computation on it by making the device operations uninterruptable. The effective and efficient method for detection and recognition of the safety signs from images is selected. The ubiquitous SIFT and SURF feature detectors consume more time and are computationally complex and require very high-level hardware components for processing. The binary descriptors are taken into account as they are light weight and can support low power devices in a much effective style. A comparative analysis is being done on the working of binary descriptors like BRIEF, ORB, AKAZE, FREAK, etc., on the smart glasses based on their performance and the requirements. ORB is the most efficient among the binary descriptors and has been more effective for the smart glasses in terms of time measurements and low power consumption.

Dans les examens médicaux et les actes de thérapie, les techniques minimalement invasives sont de plus en plus utilisées. Des instruments comme des aiguilles de biopsie, ou des électrodes sont utilisés pour extraire des échantillons de cellules ou pour effectuer des traitements. Afin de réduire les traumatismes et de faciliter le suivi visuelle de ces interventions, des systèmes d’assistance par imagerie médicale, comme par exemple, par l’échographie 2D, sont utilisés dans la procédure chirurgicale. Nous proposons d’utiliser l’échographie 3D pour faciliter la visualisation de l’aiguille, mais en raison de l’aspect bruité de l’image ultrasonore (US) et la grande quantité de données d’un volume 3D, il est difficile de trouver l’aiguille de biopsie avec précision et de suivre sa position en temps réel. Afin de résoudre les deux principaux problèmes ci-dessus, nous avons proposé une méthode basée sur un algorithme RANSAC et un filtre de Kalman. De même l’étude est limitée à une région d’intérêt (ROI) pour obtenir une localisation robuste et le suivi de la position de l’aiguille de biopsie en temps réel. La méthode ROI-RK se compose de deux étapes: l’étape d’initialisation et l’étape de suivi. Dans la première étape, une stratégie d’initialisation d’une ROI en utilisant le filtrage de ligne à base de matrice de Hesse est mise en œuvre. Cette étape permet de réduire efficacement le bruit de granularité du volume US, et de renforcer les structures linéaires telles que des aiguilles de biopsie. Dans la deuxième étape, après l’initialisation de la ROI, un cycle de suivi commence. L’algorithme RK localise et suit l’aiguille de biopsie dans une situation dynamique. L’algorithme RANSAC est utilisé pour estimer la position des micro-outils et le filtrage de Kalman permet de mettre à jour la région d’intérêt et de corriger la localisation de l’aiguille. Une stratégie d’estimation de mouvement est également appliquée pour estimer la vitesse d’insertion de l’aiguille de biopsie. Des volumes 3D US avec un fond inhomogène ont été simulés pour vérifier les performances de la méthode ROI-RK. La méthode a été testée dans des conditions variables, telles que l’orientation d’insertion de l’aiguille par rapport à l’axe de la sonde et le niveau de contraste (CR). La précision de la localisation est de moins de 1 mm, quelle que soit la direction d’insertion de l’aiguille. Ce n’est que lorsque le CR est très faible que la méthode proposée peut échouer dans le suivi d’une structure incomplète de l’aiguille. Une autre méthode, utilisant l’algorithme RANSAC avec apprentissage automatique a été proposée. Cette méthode vise à classer les voxels en se basant non seulement sur l’intensité, mais aussi sur les caractéristiques de la structure de l’aiguille de biopsie. Les résultats des simulations montrent que l’algorithme RANSAC avec apprentissage automatique peut séparer les voxels de l’aiguille et les voxels de tissu de fond avec un CR faible. / In medical examinations and surgeries, minimally invasive technologies are getting used more and more often. Some specially designed surgical instruments, like biopsy needles, or electrodes are operated by radiologists or robotic systems and inserted in human’s body for extracting cell samples or delivering radiation therapy. To reduce the risk of tissue injury and facilitate the visual tracking, some medical vision assistance systems, as for example, ultrasound (US) systems can be used during the surgical procedure. We have proposed to use the 3D US to facilitate the visualization of the biopsy needle, however, due to the strong speckle noise of US images and the large calculation load involved as soon as 3D data are involved, it is a challenge to locate the biopsy needle accurately and to track its position in real time in 3D US. In order to solve the two main problems above, we propose a method based on the RANSAC algorithm and Kalman filter. In this method, a region of interest (ROI) has been limited to robustly localize and track the position of the biopsy needle in real time. The ROI-RK method consists of two steps: the initialization step and the tracking step. In the first step, a ROI initialization strategy using Hessian based line filter measurement is implemented. This step can efficiently reduce the speckle noise of the ultrasound volume, and enhance line-like structures as biopsy needles. In the second step, after the ROI is initialized, a tracking loop begins. The RK algorithm can robustly localize and track the biopsy needles in a dynamic situation. The RANSAC algorithm is used to estimate the position of the micro-tools and the Kalman filter helps to update the ROI and auto-correct the needle localization result. Because the ROI-RK method is involved in a dynamic situation, a motion estimation strategy is also implemented to estimate the insertion speed of the biopsy needle. 3D US volumes with inhomogeneous background have been simulated to evaluate the performance of the ROI-RK method. The method has been tested under different conditions, such as insertion orientations angles, and contrast ratio (CR). The localization accuracy is within 1 mm no matter what the insertion direction is. Only when the CR is very low, the proposed method could fail to track because of an incomplete ultrasound imaging of the needle. Another methodology, i.e. RANSAC with machine learning (ML) algorithm has been presented. This method aims at classifying the voxels not only depending on their intensities, but also using some structure features of the biopsy needle. The simulation results show that the RANSAC with ML algorithm can separate the needle voxels and background tissue voxels with low CR.

This thesis presents the novel use of Coherent Point Drift in improving the robustness of a number of computer vision applications. CPD approach includes two methods for registering two images - rigid and non-rigid point set approaches which are based on the transformation model used. The key characteristic of a rigid transformation is that the distance between points is preserved, which means it can be used in the presence of translation, rotation, and scaling. Non-rigid transformations - or affine transforms - provide the opportunity of registering under non-uniform scaling and skew. The idea is to move one point set coherently to align with the second point set. The CPD method finds both the non-rigid transformation and the correspondence distance between two point sets at the same time without having to use a-priori declaration of the transformation model used. The first part of this thesis is focused on speaker identification in video conferencing. A real-time, audio-coupled video based approach is presented, which focuses more on the video analysis side, rather than the audio analysis that is known to be prone to errors. CPD is effectively utilised for lip movement detection and a temporal face detection approach is used to minimise false positives if face detection algorithm fails to perform. The second part of the thesis is focused on multi-exposure and multi-focus image fusion with compensation for camera shake. Scale Invariant Feature Transforms (SIFT) are first used to detect keypoints in images being fused. Subsequently this point set is reduced to remove outliers, using RANSAC (RANdom Sample Consensus) and finally the point sets are registered using CPD with non-rigid transformations. The registered images are then fused with a Contourlet based image fusion algorithm that makes use of a novel alpha blending and filtering technique to minimise artefacts. The thesis evaluates the performance of the algorithm in comparison to a number of state-of-the-art approaches, including the key commercial products available in the market at present, showing significantly improved subjective quality in the fused images. The final part of the thesis presents a novel approach to Vehicle Make & Model Recognition in CCTV video footage. CPD is used to effectively remove skew of vehicles detected as CCTV cameras are not specifically configured for the VMMR task and may capture vehicles at different approaching angles. A LESH (Local Energy Shape Histogram) feature based approach is used for vehicle make and model recognition with the novelty that temporal processing is used to improve reliability. A number of further algorithms are used to maximise the reliability of the final outcome. Experimental results are provided to prove that the proposed system demonstrates an accuracy in excess of 95% when tested on real CCTV footage with no prior camera calibration.