Human operators of modern surveillance systems are confronted with an increasing amount of trajectory data from moving objects, such as people, vehicles, vessels, and aircraft. A large majority of these trajectories reflect routine traffic and are uninteresting. Nevertheless, some objects are engaged in dangerous, illegal or otherwise interesting activities, which may manifest themselves as unusual and abnormal trajectories. These anomalous trajectories can be difficult to detect by human operators due to cognitive limitations. In this thesis, we study algorithms for the automated detection of anomalous trajectories in surveillance applications. The main results and contributions of the thesis are two-fold. Firstly, we propose and discuss a novel approach for anomaly detection, called conformal anomaly detection, which is based on conformal prediction (Vovk et al.). In particular, we propose two general algorithms for anomaly detection: the conformal anomaly detector (CAD) and the computationally more efficient inductive conformal anomaly detector (ICAD). A key property of conformal anomaly detection, in contrast to previous methods, is that it provides a well-founded approach for the tuning of the anomaly threshold that can be directly related to the expected or desired alarm rate. Secondly, we propose and analyse two parameter-light algorithms for unsupervised online learning and sequential detection of anomalous trajectories based on CAD and ICAD: the sequential Hausdorff nearest neighbours conformal anomaly detector (SHNN-CAD) and the sequential sub-trajectory local outlier inductive conformal anomaly detector (SSTLO-ICAD), which is more sensitive to local anomalous sub-trajectories. We implement the proposed algorithms and investigate their classification performance on a number of real and synthetic datasets from the video and maritime surveillance domains. The results show that SHNN-CAD achieves competitive classification performance with minimum parameter tuning on video trajectories. Moreover, we demonstrate that SSTLO-ICAD is able to accurately discriminate realistic anomalous vessel trajectories from normal background traffic.

With facial recognition technology becoming a greater part of our everyday lives the ethicalimplications it may bring is something worth exploring, and according to scientists the technology could already be incorporated in video surveillance. Previous studies have shown thatwomen and men see things differently within this context, something that will be explored inthis paper. Through focus group interviews and a survey study the students could not confirmthis statement, and found that in only one specific context women and men had significantdifferent standpoints. Further the study found that there is a fear of what this technology couldresult in, in the form of abuse of the information gathered about people and how society couldchange with the incorporation of said technology in video surveillance. / Eftersom ansiktsigenkänningsteknologi blir en större del av vår vardag är de etiska konsekvenserna det kan ge något som är värt att utforska, och enligt forskare kan tekniken redanfinnas implementerad i videoövervakning. Tidigare studier har visat att kvinnor och män uppfattar saker annorlunda inom detta sammanhang, något som kommer att undersökas i dennauppsats. Genom fokusgruppsintervjuer och en enkätstudie kunde studenterna inte bekräftadetta uttalande och fann endast att kvinnor och män i ett specifikt sammanhang hade märkbartolika uppfattningar. Vidare fann studien att det finns en rädsla för vad denna teknik kan resultera i, i form av missbruk av den information som samlats in om människor och hur samhälletkan förändras genom införandet av nämnd teknik i videoövervakning.

No description available.

Bürger weisen eine positive Einstellung gegenüber Videoüberwachung auf, obwohl sie sehr wenig über Videoüberwachung nachdenken, wenig über die Zahl und Verteilung der Videokameras in ihrer Stadt wissen, Videoüberwachung nicht mit ihren Bürgerrechten in Beziehung bringen sowie dem Staat „blind\\\\\\\\\\\\\\\" vertrauen. Klocke resümiert: Das Unwissen über die Kamerawirklichkeit ist als ein Anzeichen für bürgerrechtliche Unmotiviertheit und mangelnde Freiheitssensibilität anzusehen. Daraus ergibt sich die Forschungsfrage dieses Aufsatzes, welche darauf abzielt nicht die Einstellung zur Videoüberwachung, sondern die (geringe) Aufmerksamkeit gegenüber Videoüberwachung zu erklären: Warum sind Menschen gleichgültig gegenüber Videoüberwachung, obwohl dadurch eines ihrer Grundrechte beeinträchtigt wird?:Vorbemerkung; Fragestellung; Theoretischer Ansatz und Annahmen; Daten und Operationalisierung; Ergebnisse; Zusammenfassung und Diskussion der Ergebnisse; Ausblick

With the increasing demand of intelligent video surveillance systems, person re-identification (re-ID) plays an important role in intelligent video analysis, which aims at matching person images across non-overlapping camera views. It has gained increasing attention in computer vision community. With the advanced deep neural networks, existing methods have achieved promising performance on the widely-used re-ID benchmarks, even outperform the human-level rank-1 matching accuracy. However, most of the research efforts are conducted on the closed-world settings, with large-scale well annotated training data and all the person images are from the same visible modality. As a prerequisite in practical video surveillance application, there is still a large gap between the closed-world research-oriented setting and the practical open-world settings. In this thesis, we try to narrow the gap by studying three important issues in open-world person re-identification, including 1) unsupervised learning with large-scale unlabelled training data; 2) learning robust re-ID model with label corrupted training data and 3) cross-modality visible-thermal person re-identification with multi-modality data. For unsupervised learning with unlabelled training data, we mainly focus on video-based person re-identification, since the video data is usually easily obtained by tracking algorithms and the video sequence provides rich weakly labelled samples by assuming the image frames within the tracked sequence belonging to the same person identity. Following the cross-camera label estimation approach, we formulate the cross-camera label estimation as a one-to-one graph matching problem, and then propose a novel dynamic graph matching framework to estimate cross-camera labels. However, in a practical wild scenario, the unlabelled training data usually cannot satisfy the one-to-one matching constraint, which would result in a large proportion of false positives. To address this issue, we further propose a novel robust anchor embedding method for unsupervised video re-ID. In the proposed method, some anchor sequences are firstly selected to initialize the CNN feature representation. Then a robust anchor embedding method is proposed to measure the relationship between the unlabelled sequences and anchor sequences, which considers both the scalability and efficiency. After that, a top-{dollar}k{dollar} counts label prediction strategy is proposed to predict the labels of unlabelled sequences. With the newly estimated sequences, the CNN representation could be further updated. For robust re-ID model learning with label corrupted training data, we propose a two-stage learning method to handle the label noise. Rather than simply filtering the falsely annotated samples, we propose a joint learning method by simultaneously refining the falsely annotated labels and optimizing the neural networks. To address the limited training samples for each identity, we further propose a novel hard-aware instance re-weighting strategy to fine-tune the learned model, which assigns larger weights to hard samples with correct labels. For cross-modality visible-thermal person re-identification, it addresses an important issue in night-time surveillance applications by matching person images across different modalities. We propose a dual-path network to learn the cross-modality feature representations, which learns the multi-modality sharable feature representations by simultaneously considering the modality discrepancy and commonness. To guide the feature representation learning process, we propose a dual-constrained top-ranking loss, which contains both cross-modality and intra-modality top-ranking constraints to reduce the large cross-modality and intra-modality variations. Besides the open-world person re-identification, we have also studied the unsupervised embedding learning problem for general image classification and retrieval. Motivated by supervised embedding learning, we propose a data augmentation invariant and instance spread-out feature. To learn the feature embedding, we propose a instance feature-based softmax embedding, which optimizes the embedding directly on top of the real-time instance features. It achieves much faster learning speed and better accuracy than existing methods. In short, the major contributions of this thesis are summarized as follows. l A dynamic graph matching framework is proposed to estimate cross-camera labels for unsupervised video-based person re-identification. l A robust anchor embedding method with top-{dollar}k{dollar} counts label prediction is proposed to efficiently estimate the cross-camera labels for unsupervised video-based person re-identification under wild settings. l A two-stage PurifyNet is introduced to handle the label noise problem in person re-identification, which jointly refines the falsely annotated labels and mines hard samples with correct labels. l A dual-constrained top-ranking loss with a dual-path network is proposed for cross-modality visible-thermal person re-identification, which simultaneously addresses the cross-modality and intra-modality variations. l A data augmentation invariant and instance spread-out feature is proposed for unsupervised embedding learning, which directly optimizes the learned embedding on top of real-time instance features with softmax function

Désormais, l'analyse des scènes denses s'impose incontestablement comme une tâche importante pour contrôler et gérer les foules. Notre recherche a pour objectifs d'apporter des solutions à l'estimation de la densité de la foule et de prouver l'utilité de cette estimation comme préalable pour d'autres applications. Concernant le premier objectif, afin de cerner les difficultés de la détection de personnes dans une foule, on se focalise sur l'estimation de la densité basée sur un niveau d'analyse bas. Dans un premier temps, on démontre que nos approches sont plus adéquates que les méthodes de l’état de l’art que ce soit pour compter les individus ou pour estimer le niveau de la foule. Dans un second temps, nous proposons une approche innovante dans laquelle une estimation locale au niveau des pixels remplace l'estimation au niveau global de la foule ou le nombre des personnes. Elle est basée sur l’utilisation des suivis de caractéristiques visuelles dans une fonction de densité. Notre recherche a également pour objectif d'utiliser la densité comme information supplémentaire pour affiner d'autres tâches. D'abord, nous avons utilisé la mesure de la densité qui comporte une description pertinente à la répartition spatiale des individus afin d'améliorer leur détection et leur suivi dans les foules. Ensuite, en prenant en compte la notion de la protection de la vie privée, nous ajustons le niveau de floutage en fonction de la densité de la foule. Enfin, nous nous appuyons sur l’estimation locale de la densité ainsi que sur le mouvement en tant qu'attributs pour des applications de haut niveau telles que la détection des évolutions et la reconnaissance des événements. / Crowd analysis has recently emerged as an increasingly important problem for crowd monitoring and management in the visual surveillance community. In this thesis, our objectives are to address the problems of crowd density estimation and to investigate the usefulness of such estimation as additional information to other applications. Towards the first goal, we focus on the problems related to the estimation of the crowd density using low level features in order to avert typical problems in detection of high density crowd. We demonstrate in this dissertation, that the proposed approaches perform better than the baseline methods, either for counting people, or alternatively for estimating the crowd level. Afterwards, we propose a novel approach, in which local information at the pixel level substitutes the overall crowd level or person count. It is based on modeling time-varying dynamics of the crowd density using sparse feature tracks as observations of a probabilistic density function. The second goal is to use crowd density as additional information to complement other tasks related to video surveillance in crowds. First, we use the proposed crowd density measure which conveys rich information about the local distributions of persons to improve human detection and tracking in videos of high density crowds. Second, we investigate the concept of crowd context-aware privacy protection by adjusting the obfuscation level according to the crowd density. Finally, we employ additional information about the local density together with regular motion patterns as crowd attributes for high level applications such as crowd change detection and event recognition.

The wide usage of surveillance cameras provides data that can be used in various areas, such as security and urban planning. An important stepping stone for useful information extraction is matching the seen object across different points in time or different cameras. In this work, we focus specifically on this part of the video processing, usually referred to as re-identification. We split our work into two stages. In the first part, we focus on the spatial and temporal information regarding the detected objects. In the second part, we combine this metadata with the visual information. For the extraction of useful descriptors from the images, we use methods based on the color distribution as well as state-of-the-art deep neural networks. We also annotate a dataset to provide a comprehensive evaluation of our approaches. Additionally, we provide a custom tool we used to annotate the dataset. 1

Video surveillance systems are a rapidly growing industry. As with most systems, this technology presents both opportunities and threats. The wide adoption of video surveillance systems by various businesses and individuals has raised some vital security issues.  Appropriately addressing these security issues is of great importance for video surveillance systems, as these systems may capture sensitive personal data and may attract numerous attacks. As of today nearly all devices have become networked (or are on their way to being connected to networks), hence eavesdropping is a common attack which can exploit a breach of a system’s security and result in data disclosure to unauthorised parties, video stream alterations, interference, and reduction of a system’s performance. Moreover, it is important that video surveillance systems are standardized by appropriate standardization organizations in order to assure high quality of the security services that utilize these systems and to facilitate interoperability. In this master thesis project rules and regulations concerning personal data protection were studied in order to define the requirements of the proposed robust and high quality security scheme that is to be integrated into video surveillance systems. This security scheme provides United States Federal Information (FIPS)* compliant security services by securing the communication channel between the system’s devices. The authentication of the system’s devices is established by using certificates and key exchanges. The proposed security scheme has been scrutinized in order to analyze its performance (and efficiency) in terms of overhead, increased jitter, and one-way delay variations.<p> Our implementation of the proposed security scheme utilized OpenVPN to provide privacy, integrity and authentication to the video streaming captured by Veracity’s clients and stored in Veracity’s proprietary NAS device (COLDSTORE). Utilization of OpenSSL FIPS Object module develops our security scheme in a FIPS compliant solution. For testing purposes, we created different test scenarios and collected data about the total delivery time of a video file, delivered from the IPCamera/NVR/DVR devices to the COLDSTORE device, the network overhead and lastly the one-way delay between the two endpoints. Another area of interest that we focus on is how to deploy certificates to new, existing, and replacement devices; and how this deployment may affect the system’s security design. In addition, we investigate the problems arising when a secured video stream needs to be played back via another device outside of our system’s network.The results of the thesis will be used as an input for product development activities by the company that hosted this thesis project. / Videoövervakningssystem är en växande industri. Precis som med de flesta systemen, har denna teknologi både möjligheter och risker. Den stora utspridningen av videoövervarkningssystemen har lett till essentiella säkerhetsrisker. Det ligger en stor vikt i att hantera säkerhetsrisker för videoövervakningssystem i och med att dessa system kan eventuellt fånga upp personlig data och därav attrahera attacker. Idag har nästan alla enheter blivit nätverksanslutna (eller är påväg att bli), vilket har lett till att avlyssning har blivit en vanlig attack. En avlyssnare kan exploatera en säkerhetsrisk och resultera i informationsläckor till obehöriga, videomanipulering, störningar, och reducerad prestanda i systemet. Det viktigt att videoövervakningssystem är standardiserade av lämpliga standardiseringsorganisationer för att säkra en hög kvalité i säkerhetstjänsterna som använder sig av dessa system och för att försäkra sig om kompatibilitet.<p> I den här examensarbetet studerade man regler och förordningar som har att göra med säkrandet av personlig data, för att kunna definiera kraven för det föreslagna robusta och högkvalitativa säkerhetsarkitekturen som skall integreras med videoövervakningssystemen.  Säkerhetsarkitekturen erbjuder United States Federal Information (FIPS)* kompatibla säkerhetstjänster genom att säkra kommunikationskanalen mellan systemets enheter.  Autentiseringen av systemets enheter sker genom att använda certifikat och nyckelutbyten.  Det föreslagna säkerhetsarkitekturen har granskats för att analysera dess prestanda vad gäller ineffektiviteter, ökade störningar och fördröjningar i envägs variationer. Vår genomförandet av den föreslagna systemet utnyttjas OpenVPN att tillhandahålla sekretess, integritet och autentisering till strömmande video fångades av Veracity kunder och lagras i Veracity egenutvecklade NAS-enhet (COLDSTORE). Utnyttjande av OpenSSL FIPS Objekt modulen utvecklar vår trygghet i ett FIPS-kompatibel lösning. För teständamål, skapade vi olika testscenarier och insamlade data om den totala leveranstiden för en videofil, som levereras från IPCamera / NVR / DVR-enheter till fryshus enhet, nätverket overhead och slutligen den enkelriktad fördröjning mellan de två ändpunkterna. Ett annat område av intresse som vi fokuserar på är certifikat för nya, existerande och ersättningsenheter; och hur det kan påverka systemets säkerhetsarkitektur. Utöver detta undersöker vi problemen som uppstår när en säkrad videoström behöver spelas upp i en enhet utanför systemets nätverk. Insatsen gjord i det här examensarbetet kommer användas som grund för produktutvecklingen av företaget där examensarbetet gjordes.

La problématique de l’observance de la biosécurité est présente dans tous les types de production. Il est essentiel de définir des stratégies pour améliorer l’application des mesures de biosécurité. Cette étude décrit l’application des mesures de biosécurité à l’entrée et à la sortie de 24 bâtiments d’élevages avicoles au Québec, Canada. L’effet des audits et de caméras visibles sur l’observance a été étudié, de même que les déterminants de l’observance. De plus, la relation entre l’observance et les profils de personnalité, l’expérience et l’éducation a été décrite. L’application des mesures de biosécurité a été évaluée à l’aide de caméras cachées. L’observance à court terme (deux premières semaines) et à moyen terme (six mois plus tard) a été déterminée. Basés sur les résultats du groupe contrôle, 44 différentes erreurs lors de l’application des mesures de biosécurité ont été observées à l’entrée et la sortie des bâtiments. La plupart étaient reliées à la délimitation des zones (propre versus contaminée). La nature et la fréquence des erreurs suggèrent un manque de compréhension des principes associés aux mesures de biosécurité. Le visionnement des vidéos a révélé 3055 visites par 277 individus différents (136 employés, 123 visiteurs, 3 superviseurs et 15 éleveurs). Les résultats ont démontré que les audits n’avaient pas d’impact sur l’observance des employés. Les caméras visibles ont eu un impact, à court terme, sur le port de bottes et le respect des zones durant la visite. Par contre, six mois plus tard, l’observance avait significativement diminué, au point de ne plus être statistiquement plus élevée que le groupe contrôle. La durée et le moment de la visite, la présence de l’éleveur ou d’un observateur, la conception de l’entrée, le nombre de bâtiments, le nombre de mesures de biosécurité exigé, le type de bottes, le genre et être membre de la famille de l’éleveur étaient significativement associés à l’observance de certaines mesures. Finalement, trois traits de la personnalité étaient associés à l’observance: responsabilité, orienté vers l’action et complexité, de même que le nombre d’années d’expérience et le niveau d’éducation. Il est nécessaire d’améliorer la formation en matière de biosécurité en fournissant du matériel de formation à tous les intervenants qui démontrent pourquoi et comment appliquer les mesures de biosécurité. La formation continue devrait également aborder les problématiques reliées aux caractéristiques de visites et de fermes. Améliorer la conception des entrées de bâtiments devrait contribuer à augmenter et à maintenir l’observance. L’identification de traits de personnalité associés à l’observance peut avoir des implications sur la sélection des candidats à l’embauche ou sur l’attribution de tâches et sur la planification des programmes de formation. / Biosecurity compliance issue is present in all types of animal productions, Therefore, it is essential to define strategies to improve the implementation of biosecurity measures. This study described the application of biosecurity measures when entering and exiting 24 poultry barns in Québec, Canada. The effect of audits and visible cameras on compliance was investigated, as well as determinants of compliance. Also, the relationship between compliance and personality profiles, experience and education has been described. Application of biosecurity measures was evaluated using hidden cameras. Short term (first two weeks) and medium term (six months later) compliance were determined. Based on the control group, 44 different biosecurity breaches were observed when getting in and out of poultry barns. Most were related to area delimitation (clean versus contaminated). The nature and frequency of errors suggest a lack of understanding of biosecurity principles. Overall, video viewing revealed 3055 visits done by 277 different individuals (136 employees, 123 visitors, 3 supervisors and 15 growers). Results showed that audits did not have any impact on employee compliance. Visible cameras had a significant impact on changing boots and respecting areas during the visit for the short term period. However, six months later, compliance significantly declined and was no longer statistically higher compared to the control group. Duration and moment of the visit, presence of the grower or an observer, barn entrance design, number of barns, number of biosecurity measures requested, type of boots, gender and being a member of a grower’s family were significantly associated with compliance with some biosecurity measures. Finally, three personality traits were significantly associated with compliance: responsibility, action-oriented and complexity, as well as the number of years of experience and the level of education. There is a need to improve biosecurity training by making training material available to all poultry personnel demonstrating why and how to apply biosecurity measures. Educational meetings should also address issues related to visit and farm characteristics. Improving barn entrance design should contribute to enhance and maintain compliance. The identification of personality traits associated with compliance may have implications for the selection of job applicants or task attribution, and for developing educational materials and training programs.

Cette thèse est effectuée en collaboration entre le LaBRI (Laboratoire bordelais de recherche en informatique) et MIRANE S.A.S., le leader français en Publicité sur Lieu de Vente (PLV) Dynamique. Notre but est d'analyser des comportements humains dans un point de vente. Le long de cette thèse, nous présentons un système d'analyse vidéo composé de plusieurs procédés de divers niveaux. Nous présentons, dans un premier temps, l'analyse vidéo de bas niveau composée de la détection de mouvement et du suivi d'objets. Puis nous analysons le comportement de ces objets suivis, lors de l'analyse de niveau moyen. Finalement, l'analyse de haut-niveau est composée d'une interprétation sémantique de ces comportements et d'une détection de scenarios de haut-niveau. / Along this thesis various subjects are studied, from the lowest to the higher level of video analysis. We first present motion detection and object tracking that compose the low-level processing part of our system. Motion detection aims at detecting moving areas, which correspond to foreground, of an image. The result of motion detection is a foreground mask that is used as input for the object tracking process. Tracking matches and identifies foreground regions across frames. Then, we analyze the behavior of the tracked objects, as the mid-level analysis. At each frame, we detect the current state of action of each tracked object currently in the scene. Finally, the system generates a semantic interpretation of these behaviors and we analyze high-level scenarios as the high-level part of our system. These two processes analyze the series of states of each object. The semantic interpretation generates sentences when state changes occur. Scenario recognition detect three different scenarios by analyzing the temporal constraints between the states.