A new algorithm for detection of longitudinal crack-like indications in radiographic images is developed in this work. Conventional local detection techniques give unsatisfactory results for this task due to the low signal to noise ratio (SNR ~ 1) of crack-like indications in radiographic images. The usage of global features of crack-like indications provides the necessary noise resistance, but this is connected with prohibitive computational complexities of detection and difficulties in a formal description of the indication shape. Conventionally, the excessive computational complexity of the solution is reduced by usage of heuristics. The heuristics to be used, are selected on a trial and error basis, are problem dependent and do not guarantee the optimal solution. Not following this way is a distinctive feature of the algorithm developed here. Instead, a global characteristic of crack-like indication (the estimation function) is used, whose maximum in the space of all possible positions, lengths and shapes can be found exactly, i.e. without any heuristics. The proposed estimation function is defined as a sum of a posteriori information gains about hypothesis of indication presence in each point along the whole hypothetical indication. The gain in the information about hypothesis of indication presence results from the analysis of the underlying image in the local area. Such an estimation function is theoretically justified and exhibits a desirable behaviour on changing signals. The developed algorithm is implemented in the C++ programming language and testet on synthetic as well as on real images. It delivers good results (high correct detection rate by given false alarm rate) which are comparable to the performance of trained human inspectors. / In dieser Arbeit wurde ein neuer Algorithmus zur Detektion rissartiger Anzeigen in der digitalen Radiographie entwickelt. Klassische lokale Detektionsmethoden versagen wegen des geringen Signal-Rausch-Verhältnisses (von ca. 1) der Rissanzeigen in den Radiographien. Die notwendige Resistenz gegen Rauschen wird durch die Benutzung von globalen Merkmalen dieser Anzeigen erzielt. Das ist aber mit einem undurchführbaren Rechenaufwand sowie Problemen bei der formalen Beschreibung der Rissform verbunden. Üblicherweise wird ein übermäßiger Rechenaufwand bei der Lösung vergleichbarer Probleme durch Anwendung von Heuristisken reduziert. Dazu benuzte Heuristiken werden mit der Versuchs-und-Irrtums-Methode ermittelt, sind stark problemangepasst und können die optimale Lösung nicht garantieren. Das Besondere dieser Arbeit ist anderer Lösungsansatz, der jegliche Heuristik bei der Suche nach Rissanzeigen vermeidet. Ein globales wahrscheinlichkeitstheoretisches Merkmal, hier Schätzfunktion genannt, wird konstruiert, dessen Maximum unter allen möglichen Formen, Längen und Positionen der Rissanzeige exakt (d.h. ohne Einsatz jeglicher Heuristik) gefunden werden kann. Diese Schätzfunktion wird als die Summe des a posteriori Informationsgewinns bezüglich des Vorhandenseins eines Risses im jeden Punkt entlang der hypothetischen Rissanzeige definiert. Der Informationsgewinn entsteht durch die Überprüfung der Hypothese der Rissanwesenheit anhand der vorhandenen Bildinformation. Eine so definierte Schätzfunktion ist theoretisch gerechtfertigt und besitzt die gewünschten Eigenschaften bei wechselnder Anzeigenintensität. Der Algorithmus wurde in der Programmiersprache C++ implementiert. Seine Detektionseigenschaften wurden sowohl mit simulierten als auch mit realen Bildern untersucht. Der Algorithmus liefert gute Ergenbise (hohe Detektionsrate bei einer vorgegebenen Fehlalarmrate), die jeweils vergleichbar mit den Ergebnissen trainierter menschlicher Auswerter sind.

Visual recognition research develops algorithms and representations to autonomously recognize visual entities such as objects, actions, and attributes. The traditional protocol involves manually collecting training image examples, annotating them in specific ways, and then learning models to explain the annotated examples. However, this is a rather limited way to transfer human knowledge to visual recognition systems, particularly considering the immense number of visual concepts that are to be learned. I propose new forms of active learning that facilitate large-scale transfer of human knowledge to visual recognition systems in a cost-effective way. The approach is cost-effective in the sense that the division of labor between the machine learner and the human annotators respects any cues regarding which annotations would be easy (or hard) for either party to provide. The approach is large-scale in that it can deal with a large number of annotation types, multiple human annotators, and huge pools of unlabeled data. In particular, I consider three important aspects of the problem: (1) cost-sensitive multi-level active learning, where the expected informativeness of any candidate image annotation is weighed against the predicted cost of obtaining it in order to choose the best annotation at every iteration. (2) budgeted batch active learning, a novel active learning setting that perfectly suits automatic learning from crowd-sourcing services where there are multiple annotators and each annotation task may vary in difficulty. (3) sub-linear time active learning, where one needs to retrieve those points that are most informative to a classifier in time that is sub-linear in the number of unlabeled examples, i.e., without having to exhaustively scan the entire collection. Using the proposed solutions for each aspect, I then demonstrate a complete end-to-end active learning system for scalable, autonomous, online learning of object detectors. The approach provides state-of-the-art recognition and detection results, while using minimal total manual effort. Overall, my work enables recognition systems that continuously improve their knowledge of the world by learning to ask the right questions of human supervisors. / text

The safety of Powered Two Wheelers (PTWs) is important for public authorities and roadadministrators around the world. Recent official figures show that PTWs are estimated to represent only 2% of the total traffic but represent 30% of total deaths on French roads. However, as these estimated figures are obtained by simply counting the number plates registered, they do not give a true picture of the PTWs on the road at any given moment. This dissertation comes under the project METRAMOTO and is a technical applied research work and deals with two problems: detection of PTWsand the use of a laser scanner to count PTWs in the traffic. Traffic generally contains random vehicles of unknown nature and behaviour such as speed,vehicle interaction with other users on the road etc. Even though there are several technologies that can measure traffic, for example radars, cameras, magnetometers etc, as the PTWs are small-sized vehicles, they often move in between lanes and at quite a high speed compared to the vehicles moving in the adjacent lanes. This makes them difficult to detect. the proposed solution in this research work is composed of the following parts: a configuration to install the laser scanner on the road is chosen and a data coherence method is introduced so that the system is able to detect the road verges and its own height above the road surface. This is validated by simulator. Then the rawd ata obtained is pre-processed and is transform into the spatial temporal domain. Following this, an extraction algorithm called the Last Line Check (LLC) method is proposed. Once extracted, the objectis classified using one of the two classifiers either the Support Vector Machine (SVM) or the k-Nearest Neighbour (KNN). At the end, the results given by each of the two classifiers are compared and presented in this research work. The proposed solution in this research work is a propototype that is intended to be integrated in a real time system that can be installed on a highway to detect, extract, classify and counts PTWs in real time under all traffic conditions (traffic at normal speeds, dense traffic and even traffic jams).

Recent advances in telecommunications, collaborated with the development of image and video processing and acquisition devices has lead to a spectacular growth of the amount of the visual content data stored, transmitted and exchanged over Internet. Within this context, elaborating efficient tools to access, browse and retrieve video content has become a crucial challenge. In Chapter 2 we introduce and validate a novel shot boundary detection algorithm able to identify abrupt and gradual transitions. The technique is based on an enhanced graph partition model, combined with a multi-resolution analysis and a non-linear filtering operation. The global computational complexity is reduced by implementing a two-pass approach strategy. In Chapter 3 the video abstraction problem is considered. In our case, we have developed a keyframe representation system that extracts a variable number of images from each detected shot, depending on the visual content variation. The Chapter 4 deals with the issue of high level semantic segmentation into scenes. Here, a novel scene/DVD chapter detection method is introduced and validated. Spatio-temporal coherent shots are clustered into the same scene based on a set of temporal constraints, adaptive thresholds and neutralized shots. Chapter 5 considers the issue of object detection and segmentation. Here we introduce a novel spatio-temporal visual saliency system based on: region contrast, interest points correspondence, geometric transforms, motion classes' estimation and regions temporal consistency. The proposed technique is extended on 3D videos by representing the stereoscopic perception as a 2D video and its associated depth

Object detection is the initial step in any image analysis procedure and is essential for the performance of object recognition and augmented reality systems. Research concerning the detection of edges and blobs is particularly rich and many algorithms or methods have been proposed in the literature. This master‟s thesis presents 4 most common blob and edge detectors, proposes method for detected numbers separation and describes the experimental setup and results of object detection and detected numbers separation performance. Finally, we determine which detector demonstrates the best results for mobile augmented reality system. / Objektų aptikimas yra pagrindinis žingsnis vaizdų analizės procese ir yra pagrindinis veiksnys apibrėžiantis našumą objektų atpažinimo ir papildytosios realybės sistemose. Literatūroje gausu metodų ir algoritmų aprašančių sričių ir ribų aptikimą. Šiame magistro laipsnio darbe aprašomi 4 dažniausiai naudojami sričių ir ribų aptikimo algoritmai, pasiūlomas metodas aptiktų skaičių atskyrimo problemai išspręsti. Pateikiami atliktų eksperimentų rezultatai, palyginmas šių algoritmų našumas. Galiausiai yra nustatoma, kuris iš jų yra geriausias.

A device capable of continuously monitoring an individual’s levels of alertness in real-time is highly desirable for preventing drowsiness and lapse related accidents. This thesis presents the development of a non-intrusive and light-insensitive video-based system that uses computer-vision methods to localize face, eyes, and eyelids positions to measure level of eye closure within an image, which, in turn, can be used to identify visible facial signs associated with drowsiness and behavioural microsleeps. The system was developed to be non-intrusive and light-insensitive to make it practical and end-user compliant. To non-intrusively monitor the subject without constraining their movement, the video was collected by placing a camera, a near-infrared (NIR) illumination source, and an NIR-pass optical filter at an eye-to-camera distance of 60 cm from the subject. The NIR-illumination source and filter make the system insensitive to lighting conditions, allowing it to operate in both ambient light and complete darkness without visually distracting the subject. To determine the image characteristics and to quantitatively evaluate the developed methods, reference videos of nine subjects were recorded under four different lighting conditions with the subjects exhibiting several levels of eye closure, head orientations, and eye gaze. For each subject, a set of 66 frontal face reference images was selected and manually annotated with multiple face and eye features. The eye-closure measurement system was developed using a top-down passive feature-detection approach, in which the face region of interest (fROI), eye regions of interests (eROIs), eyes, and eyelid positions were sequentially localized. The fROI was localized using an existing Haar-object detection algorithm. In addition, a Kalman filter was used to stabilize and track the fROI in the video. The left and the right eROIs were localized by scaling the fROI with corresponding proportional anthropometric constants. The position of an eye within each eROI was detected by applying a template-matching method in which a pre-formed eye-template image was cross-correlated with the sub-images derived from the eROI. Once the eye position was determined, the positions of the upper and lower eyelids were detected using a vertical integral-projection of the eROI. The detected positions of the eyelids were then used to measure eye closure. The detection of fROI and eROI was very reliable for frontal-face images, which was considered sufficient for an alertness monitoring system as subjects are most likely facing straight ahead when they are drowsy or about to have microsleep. Estimation of the y- coordinates of the eye, upper eyelid, and lower eyelid positions showed average median errors of 1.7, 1.4, and 2.1 pixels and average 90th percentile (worst-case) errors of 3.2, 2.7, and 6.9 pixels, respectively (1 pixel 1.3 mm in reference images). The average height of a fully open eye in the reference database was 14.2 pixels. The average median and 90th percentile errors of the eye and eyelid detection methods were reasonably low except for the 90th percentile error of the lower eyelid detection method. Poor estimation of the lower eyelid was the primary limitation for accurate eye-closure measurement. The median error of fractional eye-closure (EC) estimation (i.e., the ratio of closed portions of an eye to average height when the eye is fully open) was 0.15, which was sufficient to distinguish between the eyes being fully open, half closed, or fully closed. However, compounding errors in the facial-feature detection methods resulted in a 90th percentile EC estimation error of 0.42, which was too high to reliably determine extent of eye-closure. The eye-closure measurement system was relatively robust to variation in facial-features except for spectacles, for which reflections can saturate much of the eye-image. Therefore, in its current state, the eye-closure measurement system requires further development before it could be used with confidence for monitoring drowsiness and detecting microsleeps.

Šiame darbe tiriami algoritminiai sprendimai mobiliam robotui, leidžiantys aptikti ieškomą objektą bei įvertinti jo poziciją ir orientaciją erdvėje. Atlikus šios srities technologijų analizę surasta įvairių realizacijai tinkamų metodų, tačiau bendro jų efektyvumo palyginimo trūko. Siekiant užpildyti šią spragą realizuota programinė ir techninė įranga, kuria atliktas labiausiai roboto sistemoms tinkamų metodų vertinimas. Algoritmų analizė susideda iš algoritmų tikslumo ir jų veikimo spartos vertinimo panaudojant tam paprastus bei efektyvius metodus. Darbe analizuojamas objektų orientacijos nustatymas iš Kinect kameros gylio duomenų pasitelkiant ICP algoritmą. Atliktas dviejų gylio sistemų spartos ir tikslumo tyrimas parodė, jog Kinect kamera spartos atžvilgiu yra efektyvesnis bei 2-5 kartus tikslesnis sprendimas nei įprastinė stereo kamerų sistema. Objektų aptikimo algoritmų efektyvumo eksperimentuose nustatytas maksimalus aptikimo tikslumas apie 90% bei pasiekta maksimali 15 kadrų/s veikimo sparta analizuojant standartinius VGA 640x480 raiškos vaizdus. Atliktas objektų pozicijos ir orientacijos nustatymo ICP metodo efektyvumo tyrimas parodė, jog vidutinė absoliutinė pozicijos ir orientacijos nustatymo paklaida yra atitinkamai apie 3.4cm bei apie 30 laipsnių, o veikimo sparta apie 2 kadrai/s. Tolesnis optimizavimas arba duomenų kiekio minimizavimas yra būtinas norint pasiekti geresnius veikimo rezultatus mobilioje ribotų resursų roboto sistemoje. Darbe taip pat buvo sėkmingai... [toliau žr. visą tekstą] / This work presents a performance analysis of the state-of-the-art computer vision algorithms for object detection and pose estimation. Initial field study showed that many algorithms for the given problem exist but still their combined comparison was lacking. In order to fill in the existing gap a software and hardware solution was created and the comparison of the most suitable methods for a robot system were done. The analysis consists of detector accuracy and runtime performance evaluation using simple and robust techniques. Object pose estimation via ICP algorithm and stereo vision Kinect depth sensor method was used in this work. A conducted two different stereo system analysis showed that Kinect achieves best runtime performance and its accuracy is 2-5 times more superior than a regular stereo setup. Object detection experiments showcased a maximum object detection accuracy of nearly 90% and speed of 15 fps for standard size VGA 640x480 resolution images. Accomplished object position and orientation estimation experiment using ICP method showed, that average absolute position and orientation detection error is respectively 3.4cm and 30 degrees while the runtime speed – 2 fps. Further optimization and data size minimization is necessary to achieve better efficiency on a resource limited mobile robot platform. The robot hardware system was also successfully implemented and tested in this work for object position and orientation detection.

[pt] A detecção e o subsequente rastreamento de objetos em sequencias de vídeo é um desafio no que tange o processamento de vídeos em tempo real. Nesta tese propomos um método de detecção em tempo real adequado para o processamento de vídeos de alta definição. Neste método utilizamos um procedimento de segmentação de quadros usando as imagens integrais de frente, o que permite o rápido descarte de várias partes da imagem a cada quadro, desta maneira atingindo uma alta taxa de quadros processados por segundo. Estendemos ainda o algoritmo proposto para que seja possível detectar múltiplos objetos em paralelo. Além disto, através da utilização de uma GPU e técnicas que podem ter seu desempenho aumentado por meio de paralelismo, como o operador prefix sum, conseguimos atingir um desempenho ainda melhor do algoritmo, tanto para a detecção do objeto, como na etapa de treinamento de novas classes de objetos. / [en] The detection and subsequent tracking of objects in video sequences is a challenge in terms of video processing in real time. In this thesis we propose an detection method suitable for processing high-definition video in real-time. In this method we use a segmentation procedure through integral image of the foreground, which allows a very quick disposal of various parts of the image in each frame, thus achieving a high rate of processed frames per second. Further we extend the proposed method to be able to detect multiple objects in parallel. Furthermore, by using a GPU and techniques that can have its performance enhanced through parallelism, as the operator prefix sum, we can achieve an even better performance of the algorithm, both for the detection of the object, as in the training stage of new classes of objects.

Ce mémoire présente les travaux réalisés dans le cadre de ma thèse. Celle-ci a été menée dans le groupe ISPR (ImageS, Perception systems and Robotics) de l’Institut Pascal au sein de l’équipe ComSee (Computers that See). Ces travaux s’inscrivent dans le cadre du projet Bio Rafale initié par la société clermontoise Vesalis et financé par OSEO. Son but est d’améliorer la sécurité dans les stades en s’appuyant sur l’identification des interdits de stade. Les applications des travaux de cette thèse concernent la détection de visages. Elle représente la première étape de la chaîne de traitement du projet. Les détecteurs les plus performants utilisent une cascade de classifieurs boostés. La notion de cascade fait référence à une succession séquentielle de plusieurs classifieurs. Le boosting, quant à lui, représente un ensemble d’algorithmes d’apprentissage automatique qui combinent linéairement plusieurs classifieurs faibles. Le détecteur retenu pour cette thèse utilise également une cascade de classifieurs boostés. L’apprentissage d’une telle cascade nécessite une base d’apprentissage ainsi qu’un descripteur d’images. Cette description des images est ici assurée par des matrices de covariance. La phase d’apprentissage d’un détecteur d’objets détermine ces conditions d’utilisation. Une de nos contributions est d’adapter un détecteur à des conditions d’utilisation non prévues par l’apprentissage. Les adaptations visées aboutissent à un problème de classification avec données manquantes. Une formulation probabiliste de la structure en cascade est alors utilisée pour incorporer les incertitudes introduites par ces données manquantes. Cette formulation nécessite l’estimation de probabilités a posteriori ainsi que le calcul de nouveaux seuils à chaque niveau de la cascade modifiée. Pour ces deux problèmes, plusieurs solutions sont proposées et de nombreux tests sont effectués pour déterminer la meilleure configuration. Enfin, les applications suivantes sont présentées : détection de visages tournés ou occultés à partir d’un détecteur de visages de face. L’adaptation du détecteur aux visages tournés nécessite l’utilisation d’un modèle géométrique 3D pour ajuster les positions des sous-fenêtres associées aux classifieurs faibles. / This thesis has been realized in the ISPR group (ImageS, Perception systems and Robotics) of the Institut Pascal with the ComSee team (Computers that See). My research is involved in a project called Bio Rafale. It was created by the compagny Vesalis in 2008 and it is funded by OSEO. Its goal is to improve the security in stadium using identification of dangerous fans. The applications of these works deal with face detection. It is the first step in the process chain of the project. Most efficient detectors use a cascade of boosted classifiers. The term cascade refers to a sequential succession of several classifiers. The term boosting refers to a set of learning algorithms that linearly combine several weak classifiers. The detector selected for this thesis also uses a cascade of boosted classifiers. The training of such a cascade needs a training database and an image feature. Here, covariance matrices are used as image feature. The limits of an object detector are fixed by its training stage. One of our contributions is to adapt an object detector to handle some of its limits. The proposed adaptations lead to a problem of classification with missing data. A probabilistic formulation of a cascade is then used to incorporate the uncertainty introduced by the missing data. This formulation involves the estimation of a posteriori probabilities and the computation of new rejection thresholds at each level of the modified cascade. For these two problems, several solutions are proposed and extensive tests are done to find the best configuration. Finally, our solution is applied to the detection of turned or occluded faces using just an uprigth face detector. Detecting the turned faces requires the use of a 3D geometric model to adjust the position of the subwindow associated with each weak classifier.

La compréhension d'image est un problème majeur de la robotique moderne, la vision par ordinateur et l'apprentissage automatique. En particulier, dans le cas des systèmes avancés d'aide à la conduite, la compréhension de scènes routières est très importante. Afin de pouvoir reconnaître le grand nombre d’objets pouvant être présents dans la scène, plusieurs capteurs et algorithmes de classification doivent être utilisés. Afin de pouvoir profiter au mieux des méthodes existantes, nous traitons le problème de la compréhension de scènes comme un problème de fusion d'informations. La combinaison d'une grande variété de modules de détection, qui peuvent traiter des classes d'objets différentes et utiliser des représentations distinctes, est faites au niveau d'une image. Nous considérons la compréhension d'image à deux niveaux : la détection d'objets et la segmentation sémantique. La théorie des fonctions de croyance est utilisée afin de modéliser et combiner les sorties de ces modules de détection. Nous mettons l'accent sur la nécessité d'avoir un cadre de fusion suffisamment flexible afin de pouvoir inclure facilement de nouvelles classes d'objets, de nouveaux capteurs et de nouveaux algorithmes de détection d'objets. Dans cette thèse, nous proposons une méthode générale permettant de transformer les sorties d’algorithmes d'apprentissage automatique en fonctions de croyance. Nous étudions, ensuite, la combinaison de détecteurs de piétons en utilisant les données Caltech Pedestrian Detection Benchmark. Enfin, les données du KITTI Vision Benchmark Suite sont utilisées pour valider notre approche dans le cadre d'une fusion multimodale d'informations pour de la segmentation sémantique. / Image understanding is a key issue in modern robotics, computer vison and machine learning. In particular, driving scene understanding is very important in the context of advanced driver assistance systems for intelligent vehicles. In order to recognize the large number of objects that may be found on the road, several sensors and decision algorithms are necessary. To make the most of existing state-of-the-art methods, we address the issue of scene understanding from an information fusion point of view. The combination of many diverse detection modules, which may deal with distinct classes of objects and different data representations, is handled by reasoning in the image space. We consider image understanding at two levels : object detection ans semantic segmentation. The theory of belief functions is used to model and combine the outputs of these detection modules. We emphazise the need of a fusion framework flexible enough to easily include new classes, new sensors and new object detection algorithms. In this thesis, we propose a general method to model the outputs of classical machine learning techniques as belief functions. Next, we apply our framework to the combination of pedestrian detectors using the Caltech Pedestrain Detection Benchmark. The KITTI Vision Benchmark Suite is then used to validate our approach in a semantic segmentation context using multi-modal information