Tesi sul tema "Traffic pattern recognition"

Designing smarter vehicles, aiming to minimize the number of driverbased wrong decisions or accidents, which can be faced with during the drive, is one of hot topics of today&rsquo
s automotive technology. In the design of smarter vehicles, several research issues can be addressed
one of which is Traffic Sign Recognition (TSR). In TSR systems, the aim is to remind or warn drivers about the restrictions, dangers or other information imparted by traffic signs, beforehand. Since the existing signs are designed to draw drivers&rsquo
attention by their colors and shapes, processing of these features is one of the crucial parts in these systems. In this thesis, a Traffic Sign Recognition System, having ability of detection and identification of traffic signs even with bad visual artifacts those originate from some weather conditions or other circumstances, is developed. The developed algorithm in this thesis, segments the required color influenced by the illumination of the environment, then reconstructs the shape of partially occluded traffic sign by its remaining segments and finally, identifies it. These three stages are called as &ldquo
Segmentation&rdquo
, &ldquo
Reconstruction&rdquo
and &ldquo
Identification&rdquo
respectively, within this thesis. Due to the difficulty of analyzing partial segments to construct the main frame (a whole sign), the main complexity of the algorithm takes place in the &ldquo
Reconstruction&rdquo
stage.

This paper explores the potential applications of existing spectral clustering algorithms to real life problems through experiments on existing road traffic data. The analysis begins with an overview of previous unsupervised machine learning techniques and constructs an effective spectral clustering algorithm that demonstrates the analytical power of the method. The paper focuses on the spectral embedding method’s ability to project non-linearly separable, high dimensional data into a more manageable space that allows for accurate clustering. The key step in this method involves solving a normalized eigenvector problem in order to construct an optimal representation of the original data. While this step greatly enhances our ability to analyze the relationships between data points and identify the natural clusters within the original dataset, it is difficult to comprehend the eigenvalue representation of the data in terms of the original input variables. The later sections of this paper will explore how the careful framing of questions with respect to available data can help researchers extract tangible decision driving results from real world data through spectral clustering analysis.

Within the fields of network operation and performance measurement, it is a common requirement that the technologies involved must provide the basis for an effective, reliable, measurable and controllable service. In order to comply with the service performance criteria, the constrains often lead to very complex techniques and methodologies for the simulation, control, test, and measurement processes. This thesis addresses some of the factors that contribute to the overall spectrum of statistical performance measurements in telecommunication services. Specifically, it is concerned with the development of three low complexity and effective techniques for real-time traffic generation, control and measurement. These techniques have proved to be accurate and near optimum. In the three cases the work starts with a literature survey of known methodologies, and later new techniques are proposed and investigated by simulating the processes involved. The work is based on the use of high-speed Asynchronous Transfer Mode (ATM) networks. The problem of developing a fast traffic generation technique for the simulation of Variable Bit Rate traffic sources is considered in the first part of this thesis. For this purpose, statistical measures are obtained from the analysis of different traffic profiles or from the literature. With the aid of these measures, a model for the fast generation of Variable Bit Rate traffic at different time resolutions is developed. The simulated traffic is then analysed in order to obtain the equivalent set of statistical measures and these are compared against those observed in real traffic traces. The subject of traffic control comprises a very wide area in communication networks. It refers to the generalised classification of actions such as Connection Admission and Flow Control, Traffic Policing and Shaping. In the second part of this thesis, a method to modify the instantaneous traffic profile of a variable rate source is developed. It is particularly useful for services which have a hard bound on the cell loss probability, but a soft bound on the admissible delay, matching the characteristics of some of the services provided by ATM networks. Finally, this thesis is also concerned with a particular aspect of the operation and management of high speed networks, or OAM functions plane, namely with the monitoring of network resources. A monitoring technique based on numerical approximation and statistical sampling methods is developed and later used to characterise a particular traffic stream, or a particular connection, within a high speed network. The resulting algorithms are simple and computationally inexpensive, but effective and accurate at the same time, and are suitable for real-time processing.

Thesis (Ph. D.)--University of California, Riverside, 2009.
Includes abstract. Title from first page of PDF file (viewed March 8, 2010). Includes bibliographical references. Issued in print and online. Available via ProQuest Digital Dissertations.

Travel-time information is essential in Advanced Traveler Information Systems (ATISs) and Advanced Traffic Management Systems (ATMSs). A key component of these systems is the prediction of the spatiotemporal evolution of roadway traffic state and travel time. From the perspective of travelers, such information can result in better traveler route choice and departure time decisions. From the transportation agency perspective, such data provide enhanced information with which to better manage and control the transportation system to reduce congestion, enhance safety, and reduce the carbon footprint of the transportation system. The objective of the research presented in this dissertation is to develop a framework that includes three major categories of methodologies to predict the spatiotemporal evolution of the traffic state. The proposed methodologies include macroscopic traffic modeling, computer vision and recursive probabilistic algorithms. Each developed method attempts to predict traffic state, including roadway travel times, for different prediction horizons. In total, the developed multi-tool framework produces traffic state prediction algorithms ranging from short – (0~5 minutes) to medium-term (1~4 hours) considering departure times up to an hour into the future. The dissertation first develops a particle filter approach for use in short-term traffic state prediction. The flow continuity equation is combined with the Van Aerde fundamental diagram to derive a time series model that can accurately describe the spatiotemporal evolution of traffic state. The developed model is applied within a particle filter approach to provide multi-step traffic state prediction. The testing of the algorithm on a simulated section of I-66 demonstrates that the proposed algorithm can accurately predict the propagation of shockwaves up to five minutes into the future. The developed algorithm is further improved by incorporating on- and off-ramp effects and more realistic boundary conditions. Furthermore, the case study demonstrates that the improved algorithm produces a 50 percent reduction in the prediction error compared to the classic LWR density formulation. Considering the fact that the prediction accuracy deteriorates significantly for longer prediction horizons, historical data are integrated and considered in the measurement update in the developed particle filter approach to extend the prediction horizon up to half an hour into the future. The dissertation then develops a travel time prediction framework using pattern recognition techniques to match historical data with real-time traffic conditions. The Euclidean distance is initially used as the measure of similarity between current and historical traffic patterns. This method is further improved using a dynamic template matching technique developed as part of this research effort. Unlike previous approaches, which use fixed template sizes, the proposed method uses a dynamic template size that is updated each time interval based on the spatiotemporal shape of the congestion upstream of a bottleneck. In addition, the computational cost is reduced using a Fast Fourier Transform instead of a Euclidean distance measure. Subsequently, the historical candidates that are similar to the current conditions are used to predict the experienced travel times. Test results demonstrate that the proposed dynamic template matching method produces significantly better and more stable prediction results for prediction horizons up to 30 minutes into the future for a two hour trip (prediction horizon of two and a half hours) compared to other state-of-the-practice and state-of-the-art methods. Finally, the dissertation develops recursive probabilistic approaches including particle filtering and agent-based modeling methods to predict travel times further into the future. Given the challenges in defining the particle filter time update process, the proposed particle filtering algorithm selects particles from a historical dataset and propagates particles using data trends of past experiences as opposed to using a state-transition model. A partial resampling strategy is then developed to address the degeneracy problem in the particle filtering process. INRIX probe data along I-64 and I-264 from Richmond to Virginia Beach are used to test the proposed algorithm. The results demonstrate that the particle filtering approach produces less than a 10 percent prediction error for trip departures up to one hour into the future for a two hour trip. Furthermore, the dissertation develops an agent-based modeling approach to predict travel times using real-time and historical spatiotemporal traffic data. At the microscopic level, each agent represents an expert in the decision making system, which predicts the travel time for each time interval according to past experiences from a historical dataset. A set of agent interactions are developed to preserve agents that correspond to traffic patterns similar to the real-time measurements and replace invalid agents or agents with negligible weights with new agents. Consequently, the aggregation of each agent's recommendation (predicted travel time with associated weight) provides a macroscopic level of output – predicted travel time distribution. The case study demonstrated that the agent-based model produces less than a 9 percent prediction error for prediction horizons up to one hour into the future.
Ph. D.

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).

Diese Dissertation soll einen Beitrag zur Weiterentwicklung der luftgestützten Verkehrslageerfassung leisten. Als Plattform dafür dient ein flugzeuggetragenes Kamerasystem, welches mit einem Inertialsystem gekoppelt ist. Vorgestellt werden hauptsächlich bildverarbeitende Algorithmen, welche an die Bildaufnahme anschließend bis hin zur Ermittlung der verkehrstechnischen Kenngrößen zum Einsatz kommen. Nach kurzer Skizzierung der verwendeten Hardware wird die Kalibrierung der Kameraeinbauwinkel durch Testflüge erläutert und auf ihre Genauigkeit hin untersucht. Es wird gezeigt, dass die Orientierungsdaten nicht die vom Hersteller angegebene Genauigkeit erreichen, was jedoch für die Verkehrslageerfassung nur von geringer Bedeutung ist. Anschließend an die Bildaufbereitung, welche die Orthobildgenerierung sowie die Eingrenzung der verkehrsaktiven Flächen beinhaltet, wird zur Ermittlung der Fahrzeugdichte ein zweistufiger Fahrzeugerkennungsalgorithmus entwickelt, welcher zunächst auf Kantenfilterbasis möglichst schnell Hypothesen erstellt. Diese werden in einer zweiten Phase durch eine Support Vector Machine überprüft, wobei ein Großteil der Fehlhypothesen verworfen wird. Die Erkennung erreicht bei guten Voraussetzungen Vollständigkeiten bis zu 90 Prozent bei sehr geringem Anteil von Fehldetektionen. Anschließend wird ein auf Singulärwertzerlegung basierender Tracking-Algorithmus verwendet, um Fahrzeughypothesen in benachbarten Bildern zu assoziieren und die mittleren Geschwindigkeiten zu ermitteln. Die erhaltenen Geschwindigkeiten unterscheiden sich um weniger als zehn km/h von den manuell erhobenen. Abschließend wird eine alternative Orientierungsmethode vorgestellt, welche auf Basis von GPS-Positionen und Bildinformationen automatisch die Fluglage ermittelt. Dies geschieht durch die Extraktion und das Matching von Straßensegmenten sowie zusätzliche Passpunktverfolgung. Die Ergebnisse weisen Genauigkeiten von etwa 0,1 bis 0,2 Grad auf.
This dissertation should make a contribution to the further development of airborne traffic detection. The used hardware is an airborne camera system combined with an inertial measurement unit for orientation determination. Mainly computer vision algorithms are presented, which are applied afterwards the image acquisition up to the determination of the most important traffic data. After a short presentation of the used hardware the calibration of the camera''s alignment angles during test flights is explained and its accuracy is analyzed. It is shown that the orientation data doesn''t reach the specified accuracy, which is fortunately less important for traffic detection. After the image preparation, which contains the ortho image generation as well as the clipping of traffic areas, a two-stage vehicle detection algorithm is implemented, which at first rapidly creates hypotheses based on edge filters. In the second stage those hypotheses are verified by a Support Vector Machine which rejects most of the False Posititves. At good conditions the detection reaches completeness rates of up to 90 percent with a low contingent of FP detections. Subsequently a tracking algorithm based on singular value decomposition is applied to associate vehicle hypotheses in adjacent images and determine the average speed. The achieved velocities differ less than ten kph from the manually obtained data. Concluding an orientation method is presented, that automatically determines the airplane''s attitude based on GPS and image information. This is realized by extraction and matching of street segments and additional tracking of ground control points. The results have accuracies of around 0.1 to 0.2 degrees.

Le thème du travail présenté dans ce mémoire traite de la détection automatique des véhicules sur autoroute par vision artificielle. Plusieurs projets et expérimentations ont été menés dans le monde autour de ce thème de recherche, ce qui prouve l'intérêt de l'information image pour la détection des véhicules et la gestion du trafic routier. Notre contribution à pour objectif de prouver la faisabilité d'un système de détection des véhicules par vision artificielle qui soit automatique et valable sur une journée complète. Ainsi des solutions originales sont apportées dans le domaine de la modélisation de la route, la calibration automatique du système de vision et la détection des véhicules sur une journée complète. La phase de modélisation de la route permet au système de s'adapter automatiquement à la diversité des sites étudiés à travers la formation de zones de traitement connexes, permettant ainsi de limiter les traitements aux seules zones utiles de roulement. Une procédure de calibration à base de technique d'identification vient compléter cette procédure de modélisation pour former ainsi la phase d'initialisation du processus de détection des véhicules. La phase de détection est scindée en trois parties: la première traite du problème de la détection diurne. Trois méthodes de détection sont réalisées et une contribution originale pour la séparation des images des véhicules rapprochés est apportée. La seconde présente la méthode utilisée pour la détection nocturne de véhicules grâce à des techniques de reconnaissance de formes. Enfin, la détection au crépuscule (et à l'aube) permet de réaliser le lien entre les deux derniers types de détection grâce à des techniques de mise en correspondance. La validité des méthodes utilisées est prouvée sur différents sites réels et la faisabilité d'un système automatique, valable sur une journée complète est maintenant acquise

La mise en oeuvre de systèmes avancés d’aide à la conduite (ADAS) basée vision, est une tâche complexe et difficile surtout d’un point de vue robustesse en conditions d’utilisation réelles. Une des fonctionnalités des ADAS vise à percevoir et à comprendre l’environnement de l’ego-véhicule et à fournir l’assistance nécessaire au conducteur pour réagir à des situations d’urgence. Dans cette thèse, nous nous concentrons sur la détection et la reconnaissance des objets mobiles car leur dynamique les rend plus imprévisibles et donc plus dangereux. La détection de ces objets, l’estimation de leurs positions et la reconnaissance de leurs catégories sont importants pour les ADAS et la navigation autonome. Par conséquent, nous proposons de construire un système complet pour la détection des objets en mouvement et la reconnaissance basées uniquement sur les capteurs de vision. L’approche proposée permet de détecter tout type d’objets en mouvement en fonction de deux méthodes complémentaires. L’idée de base est de détecter les objets mobiles par stéréovision en utilisant l’image résiduelle du mouvement apparent (RIMF). La RIMF est définie comme l’image du mouvement apparent causé par le déplacement des objets mobiles lorsque le mouvement de la caméra a été compensé. Afin de détecter tous les mouvements de manière robuste et de supprimer les faux positifs, les incertitudes liées à l’estimation de l’ego-mouvement et au calcul de la disparité doivent être considérées. Les étapes principales de l’algorithme sont les suivantes : premièrement, la pose relative de la caméra est estimée en minimisant la somme des erreurs de reprojection des points d’intérêt appariées et la matrice de covariance est alors calculée en utilisant une stratégie de propagation d’erreurs de premier ordre. Ensuite, une vraisemblance de mouvement est calculée pour chaque pixel en propageant les incertitudes sur l’ego-mouvement et la disparité par rapport à la RIMF. Enfin, la probabilité de mouvement et le gradient de profondeur sont utilisés pour minimiser une fonctionnelle d’énergie de manière à obtenir la segmentation des objets en mouvement. Dans le même temps, les boîtes englobantes des objets mobiles sont générées en utilisant la carte des U-disparités. Après avoir obtenu la boîte englobante de l’objet en mouvement, nous cherchons à reconnaître si l’objet en mouvement est un piéton ou pas. Par rapport aux algorithmes de classification supervisée (comme le boosting et les SVM) qui nécessitent un grand nombre d’exemples d’apprentissage étiquetés, notre algorithme de boosting semi-supervisé est entraîné avec seulement quelques exemples étiquetés et de nombreuses instances non étiquetées. Les exemples étiquetés sont d’abord utilisés pour estimer les probabilités d’appartenance aux classes des exemples non étiquetés, et ce à l’aide de modèles de mélange de gaussiennes après une étape de réduction de dimension réalisée par une analyse en composantes principales. Ensuite, nous appliquons une stratégie de boosting sur des arbres de décision entraînés à l’aide des instances étiquetées de manière probabiliste. Les performances de la méthode proposée sont évaluées sur plusieurs jeux de données de classification de référence, ainsi que sur la détection et la reconnaissance des piétons. Enfin, l’algorithme de détection et de reconnaissances des objets en mouvement est testé sur les images du jeu de données KITTI et les résultats expérimentaux montrent que les méthodes proposées obtiennent de bonnes performances dans différents scénarios de conduite en milieu urbain
Vision-based Advanced Driver Assistance Systems (ADAS) is a complex and challenging task in real world traffic scenarios. The ADAS aims at perceiving andunderstanding the surrounding environment of the ego-vehicle and providing necessary assistance for the drivers if facing some emergencies. In this thesis, we will only focus on detecting and recognizing moving objects because they are more dangerous than static ones. Detecting these objects, estimating their positions and recognizing their categories are significantly important for ADAS and autonomous navigation. Consequently, we propose to build a complete system for moving objects detection and recognition based on vision sensors. The proposed approach can detect any kinds of moving objects based on two adjacent frames only. The core idea is to detect the moving pixels by using the Residual Image Motion Flow (RIMF). The RIMF is defined as the residual image changes caused by moving objects with compensated camera motion. In order to robustly detect all kinds of motion and remove false positive detections, uncertainties in the ego-motion estimation and disparity computation should also be considered. The main steps of our general algorithm are the following : first, the relative camera pose is estimated by minimizing the sum of the reprojection errors of matched features and its covariance matrix is also calculated by using a first-order errors propagation strategy. Next, a motion likelihood for each pixel is obtained by propagating the uncertainties of the ego-motion and disparity to the RIMF. Finally, the motion likelihood and the depth gradient are used in a graph-cut-based approach to obtain the moving objects segmentation. At the same time, the bounding boxes of moving object are generated based on the U-disparity map. After obtaining the bounding boxes of the moving object, we want to classify the moving objects as a pedestrian or not. Compared to supervised classification algorithms (such as boosting and SVM) which require a large amount of labeled training instances, our proposed semi-supervised boosting algorithm is trained with only a few labeled instances and many unlabeled instances. Firstly labeled instances are used to estimate the probabilistic class labels of the unlabeled instances using Gaussian Mixture Models after a dimension reduction step performed via Principal Component Analysis. Then, we apply a boosting strategy on decision stumps trained using the calculated soft labeled instances. The performances of the proposed method are evaluated on several state-of-the-art classification datasets, as well as on a pedestrian detection and recognition problem.Finally, both our moving objects detection and recognition algorithms are tested on the public images dataset KITTI and the experimental results show that the proposed methods can achieve good performances in different urban scenarios

L'Internet des objets à bande étroite (NB-IoT) est une technologie de réseau étendu à faible consommation d'énergie, qui a été normalisée dans 3GPP, spécifie une nouvelle procédure d'accès aléatoire et un nouveau schéma de transmission pour l'IoT. Les avantages du réseau NB-IoT sont une couverture étendue, une faible consommation d'énergie et la prise en charge d'un grand nombre de connexions. En particulier, le réseau NB-IoT peut connecter efficacement jusqu'à 50 000 dispositifs par cellule de réseau NB-IoT. Nous concentrons notre travail sur l'étude du dimensionnement des réseaux NB-IoT. À cet égard, nous utilisons des techniques de géométrie stochastique et d'apprentissage automatique et cette thèse caractérise les indicateurs de performance clés du réseau NB-IoT, tels que la probabilité de couverture, le nombre de blocs de ressources radio nécessaires, ainsi que la reconnaissance et la prédiction des modèles de trafic sur la base des informations de contrôle en liaison descendante. La thèse est divisée en trois études principales. Premièrement, nous dérivons les performances de la probabilité de couverture de la liaison montante dans un réseau NB-IoT à cellule unique et à cellules multiples. Les expressions analytiques de la couverture et des probabilités d'accès réussi dans un réseau NB-IoT mono-cellulaire sont présentées en considérant la distribution d'arrivée des paquets. Dans le scénario multi-cellules, une prédiction de la probabilité de couverture est déterminée directement à partir des paramètres du réseau en utilisant un réseau neuronal profond. L'analyse suivante consiste en un modèle analytique permettant de calculer les blocs de ressources radio nécessaires dans le réseau NB-IoT multi-cellules et de déterminer la probabilité de panne du réseau. Ce modèle est bénéfique pour les opérateurs car il clarifie la façon dont ils doivent gérer le spectre disponible. Enfin, la thèse aborde les problèmes de reconnaissance et de prédiction du type de trafic en utilisant les données collectées à partir des informations de contrôle de la liaison descendante. Un large groupe d'algorithmes d'apprentissage automatique est mis en œuvre et comparé pour identifier celui avec les meilleures performances. L'analyse menée dans cette thèse démontre que la géométrie stochastique et les techniques d'apprentissage automatique peuvent servir d'outils puissants pour analyser les performances du réseau NB-IoT. Les framewoks développés dans ce travail fournissent des outils analytiques généraux qui peuvent être facilement étendus pour faciliter d'autres recherches sur les réseaux 5G
Narrowband Internet of Things (NB-IoT) is a Low Power Wide Area technology, which was standardized in the Third Generation Partnership Project release, specifies a new random access procedure and a new transmission scheme for IoT. The advantages of the NB-IoT network are providing deep coverage, low power consumption, and support of a huge number of connections. Especially, NB-IoT can efficiently connect up to 50,000 devices per NB-IoT network cell.We focus our work on the study of NB-IoT network dimensioning. In this regard, we use stochastic geometry and machine learning techniques along with the thesis to characterize key performance indicators of the NB-IoT network, such as coverage probability, the number of required radio resource blocks, and the traffic pattern recognition and prediction based on the downlink control information. The thesis is divided into three major studies. Firstly, we derive the performance of uplink coverage probability in single-cell and multi-cell of NB-IoT network. The analytical expressions of the coverage and successful access probabilities in a single-cell NB-IoT network are presented by considering the packet arrival distribution. In the multi-cell scenario, a prediction of coverage probability is determined directly from the network parameters by using a Deep Neural Network. The subsequent analysis consists of an analytical model to calculate the required radio resource blocks in the multi-cell NB-IoT network and determine the network outage probability. This model is beneficial for operators because it clarifies how they should manage the available spectrum. Finally, the thesis addresses the recognition and prediction traffic type problems using the data collected from the Downlink Control Information. A wide group of machine learning algorithms are implemented and compared to identify the highest performances.The analysis conducted in this thesis demonstrates that stochastic geometry and machine learning techniques can serve as powerful tools to analyze the performance of the NB-IoT network. The frameworks developed in this work provide general analytical tools that can be readily extended to facilitate other research in 5G networks

Tato práce se zabývá problematikou popisu síťového provozu pomocí automaticky vytvořeného modelu komunikace. Hlavním zaměřením jsou komunikace v řídicích systémech , které využívají speciální protokoly, jako je například IEC 60870-5-104 . V této práci představujeme metodu charakteristiky síťového provozu z pohledu obsahu komunikace i chování v čase. Tato metoda k popisu využívá deterministické konečné automaty , prefixové stromy  a analýzu opakovatelnosti. Ve druhé části této diplomové práce se zaměřujeme na implementaci programu, který je schopný na základě takového modelu komunikace verifikovat síťový provoz v reálném čase.

In law enforcement, tools and techniques are required that enable forensic analysts to uncover electronic evidence about the communication activities of possible criminal or terrorist suspects. This is needed in order to better understand the actions of criminal or terrorist groups and to also understand the communication patterns of suspected individuals. The extraction of useful information from electronic communication data is a difficult task, due to the large amounts of data and also due to the difficulty in making sense of unusual activities in the data. This thesis considers the problem of aiding the analyst to provide a better understanding about the communication behaviour of suspected individuals. The type of data considered for the thesis is e-mail traffic, which is based on information obtained from e-mail message headers but not the content of e-mails. This thesis proposes a "computational intelligence" approach for analysing email traffic, by using a set of computational techniques to provide different perspectives for examining the communication behaviour of suspect e-mail accounts. This is considered important, since a range of views on e-mail traffic behaviour can provide the user/analyst a more overall understanding about the behaviour of suspect e-mail accounts. The purpose of using a set of computational techniques is to utilise the capabilities of each technique, so that the combined effect of using those techniques present useful information to the user/analyst about a suspect e-mail account's traffic behaviour. The computational techniques used for the research in this thesis are visualisation and feature extraction techniques, which each provide different ways of examining e-mail traffic behaviour. Visualisation is used to provide a visual method of interpreting, exploring, and understanding the communication patterns present in e-mail traffic data. The two visualisation techniques used for visualization are social network visualisation and time-series visualisation. Feature extraction techniques are another type of technique used to analyse e-mail traffic behaviour, by providing information that locate features in the data, indicating where unusual changes in communication activity are occurring. The two techniques used for feature extraction in the research are decision tree classification and hierarchical fuzzy inference. Two case studies are provided in this thesis. The first case study explores the detection of unusual variations in traffic behaviour from simulated e-mail traffic data, while the second case study explores the rating of abnormal communication changes from the Enron e-mail corpus dataset. Both case studies demonstrate that computational intelligence is a useful approach for providing the user/analyst a better understanding about the traffic behaviour of suspect e-mail accounts.

近年來，在公開的大規模人工標注數據集上訓練通用行人檢測器的方法有了顯著的進步。然而，當通用行人檢測器被應用到一個特定的，未公開過的場景中時，它的性能會不如預期。這是由待檢測的數據（源樣本）與訓練數據（目標樣本）的不匹配，以及新場景中視角、光照、分辨率和背景噪音的變化擾動造成的。
在本論文中，我們提出一個新的自動將通用行人檢測器適應到特定場景中的框架。這個框架分為兩個階段。在第一階段，我們探索監控錄像場景中提供的特定表征。利用這些表征，從目標場景中選擇正負樣本並重新訓練行人檢測器，該過程不斷迭代直至收斂。在第二階段，我們提出一個新的機器學習框架，該框架綜合每個樣本的標簽和比重。根據這些比重，源樣本和目標樣本被重新權重，以優化最終的分類器。這兩種方法都屬於半監督學習，僅僅需要非常少的人工干預。
使用提出的方法可以顯著提高通用行人檢測器的准確性。實驗顯示，由方法訓練出來的檢測器可以和使用大量手工標注的目標場景數據訓練出來的媲美。與其它解決類似問題的方法比較，該方法同樣好於許多已有方法。
本論文的工作已經分別於朲朱朱年和朲朱朲年在杉杅杅杅計算機視覺和模式識別會議（权杖材杒）中發表。
In recent years, significant progress has been made in learning generic pedestrian detectors from publicly available manually labeled large scale training datasets. However, when a generic pedestrian detector is applied to a specific, previously undisclosed scene where the testing data (target examples) does not match with the training data (source examples) because of variations of viewpoints, resolutions, illuminations and backgrounds, its accuracy may decrease greatly.
In this thesis, a new framework is proposed automatically adapting a pre-trained generic pedestrian detector to a specific traffic scene. The framework is two-phased. In the first phase, scene-specific cues in the video surveillance sequence are explored. Utilizing the multi-cue information, both condent positive and negative examples from the target scene are selected to re-train the detector iteratively. In the second phase, a new machine learning framework is proposed, incorporating not only example labels but also example confidences. Source and target examples are re-weighted according to their confidence, optimizing the performance of the final classifier. Both methods belong to semi-supervised learning and require very little human intervention.
The proposed approaches significantly improve the accuracy of the generic pedestrian detector. Their results are comparable with the detector trained using a large number of manually labeled frames from the target scene. Comparison with other existing approaches tackling similar problems shows that the proposed approaches outperform many contemporary methods.
The works have been published on the IEEE Conference on Computer Vision and Pattern Recognition in 2011 and 2012, respectively.
Detailed summary in vernacular field only.
Detailed summary in vernacular field only.
Detailed summary in vernacular field only.
Detailed summary in vernacular field only.
Wang, Meng.
Thesis (M.Phil.)--Chinese University of Hong Kong, 2012.
Includes bibliographical references (leaves 42-45).
Abstracts also in Chinese.
Chapter 1 --- Introduction --- p.1
Chapter 1.1 --- PedestrianDetection --- p.1
Chapter 1.1.1 --- Overview --- p.1
Chapter 1.1.2 --- StatisticalLearning --- p.1
Chapter 1.1.3 --- ObjectRepresentation --- p.2
Chapter 1.1.4 --- SupervisedStatisticalLearninginObjectDetection --- p.3
Chapter 1.2 --- PedestrianDetectioninVideoSurveillance --- p.4
Chapter 1.2.1 --- ProblemSetting --- p.4
Chapter 1.2.2 --- Challenges --- p.4
Chapter 1.2.3 --- MotivationsandContributions --- p.5
Chapter 1.3 --- RelatedWork --- p.6
Chapter 1.4 --- OrganizationsofChapters --- p.9
Chapter 2 --- Label Inferring by Multi-Cues --- p.10
Chapter 2.1 --- DataSet --- p.10
Chapter 2.2 --- Method --- p.12
Chapter 2.2.1 --- CondentPositiveExamplesofPedestrians --- p.13
Chapter 2.2.2 --- CondentNegativeExamplesfromtheBackground --- p.17
Chapter 2.2.3 --- CondentNegativeExamplesfromVehicles --- p.17
Chapter 2.2.4 --- FinalSceneSpecicPedestrianDetector --- p.19
Chapter 2.3 --- ExperimentResults --- p.20
Chapter 3 --- Transferring a Detector by Condence Propagation --- p.24
Chapter 3.1 --- Method --- p.25
Chapter 3.1.1 --- Overview --- p.25
Chapter 3.1.2 --- InitialEstimationofCondenceScores --- p.27
Chapter 3.1.3 --- Re-weightingSourceSamples --- p.27
Chapter 3.1.4 --- Condence-EncodedSVM --- p.30
Chapter 3.2 --- Experiments --- p.33
Chapter 3.2.1 --- Datasets --- p.33
Chapter 3.2.2 --- ParameterSetting --- p.35
Chapter 3.2.3 --- Results --- p.36
Chapter 4 --- Conclusions and Future Work --- p.40

For the last years, constant news about information and data leaks are raising public discussion of the safety of the systems that we all nowadays depend on. Communications are increasingly more private; hence next-generation security systems rely on pattern recognition techniques to detect and infer the safety without the need for scrapping its content. This dissertation proposes methodologies to infer entity patterns and their nature according to their network traffic: if they are running according to their previously known safe pattern or if its behavior is uncommon, an indication of a possible breach. There is a strong indication that behavioral pattern recognition will continue to lead the research of security solutions, not only for the network traffic but also for other measurable activities. Other examples are identity access management or programs running on a computer. This dissertation proposes modeling network OSI layers 3 to 5 metadata in features that are later processed by machine learning algorithms to classify the network activity. The classification itself is divided into two groups: the first level is recognizing active entities operating within a network domain and the second if each entity is acting according to each known pattern. The presented methods of inferring if something is acting according to known patterns are transversal to other domains. Although aggregation of metadata and modeling differ, the described process of solving the problem of inferring patterns is generic and can be applied to user use cases rather than to the network, or combined with more complex scenarios. The last chapter includes a proof of concept with a few evaluation metrics using synthetic data, to evaluate if the classification algorithms can successfully distinguish different patterns. The tests showed promising results, ranging from 99% for entity classification and 77% to 98% (depending on the entity nature) for abnormality detection.
Nos últimos anos notícias sobre roubos e perdas de informação e de dados têm sido constante, levantando discussão sobre a segurança dos sistemas dos quais hoje dependemos. As comunicações são também cada vez mais privadas, pelo que os sistemas de segurança de última geração têm desenvolvido técnicas de reconhecimento de padrões para detetar e inferir a segurança sem a necessidade de processar conteúdos. Esta dissertação propõe metodologias para inferir os padrões de entidades considerando o seu tráfego de rede: se está enquadrado no comportamento de tráfego previamente conhecido, ou se a atividade gerada é incomum e, por isso, ser indicação de um possível problema. Há uma forte indicação de que o reconhecimento de padrões de comportamento continuará a liderar a investigação no domínio de soluções de segurança, não só para o tráfego de rede, mas também para outras atividades mensuráveis. Outros exemplos englobam a gestão de acesso de identidade ou programas em execução em um computador. As metodologias propõem a modelação de metadados da camada de rede OSI 3 a 5 em contagens que são posteriormente processadas por algoritmos de aprendizagem automática para classificar a atividade da rede. Esta classificação baseia-se em dois níveis: no primeiro o reconhecimento entidades ativas dentro de um domínio de rede e o segundo, se cada entidade corresponde ao padrão conhecido. As metodologias apresentadas para inferir se algo está de acordo com padrões conhecidos são transversais a outros domínios. Embora a agregação de metadados e modelação seja diferente, o processo descrito para inferir padrões é genérico o suficiente para ser aplicado a outros casos de uso, de rede ou não, ou ainda combinado em cenários mais complexos. O último capítulo inclui uma prova de conceito com dados sintéticos e algumas métricas de avaliação, para perceber se os algoritmos de classificação podem distinguir com sucesso padrões diferentes. Os testes mostraram resultados promissores, variando de 99% para classificação de entidades e 77% para 98% (dependendo da natureza da entidade) para deteção de anormalidades.
Mestrado em Engenharia de Computadores e Telemática

Maintaining correct lateral position on roadways is critical for safe and efficient operation. The driving task comprises control, navigation, and guidance, all relying predominantly upon the sense of vision; however, not all traffic control devices rely on vision. Drowsy, distracted, or inattentive drivers may not have their "eyes on the road" and, are therefore, more reliant on other senses. One way that other senses have been used to assist drivers in maintaining a proper lateral position is through the use of rumble strips. Up until the late 1990's, rumble strips were primarily used on roadway shoulders; more recently, they are also being used in centerline applications as transportation officials attempt to address the high number of cross-over-centerline crashes that occur annually. In light of the effectiveness of shoulder rumble strips in reducing the number of roadway departure crashes, many states are using the same rumble strip pattern on both the shoulder and centerline, anticipating reductions in cross-over-centerline crashes. Findings of previously conducted research indicate that this duplication of signals may violate driver expectancy. This research focuses on determining if a different pattern is beneficial to the driver, one that would allow a driver to determine their lateral position based on a rumble strip's audible and haptic cues. It was anticipated that if the pattern is discernable to drivers, then after repeated exposure, the strips would elicit a conditioned, correct response. A number of hypotheses were enumerated to assist in determining whether or not a unique rumble strip pattern would prove to be easily identified and would ultimately benefit the driver. To support or refute the hypotheses, research was performed in two evaluation environments, a static and a dynamic. Results from the evaluations indicate that a unique rumble strip pattern can aid drivers in maintaining their correct lateral position. Based on the opposite lane and shoulder incursions experienced in the dynamic evaluation environment, every driver responded properly. In instances where the incursion was accompanied with audible and haptic feedback, a result of the rumble strips, drivers were able to respond faster when compared to those incursions lacking multi-modality.

Shi, Xingang.
Thesis (Ph.D.)--Chinese University of Hong Kong, 2011.
Includes bibliographical references (leaves 135-152).
Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web.
Abstract also in Chinese.