· · · Institutional Repository

In this work we propose a template-based tracker for simultaneously tracking multiple targets such as facial features. Our work falls within the particle fil- tering framework and more specifically it extends the Particle Filtering with Factorized Likelihoods (PFFL) tracking scheme. In this work we introduce the idea of using a distribution per constellation of facial features that. These distri- butions are updated in a two-stage observation process in which appearance and shape likelihoods are evaluated separately. Information from all distributions is used to estimate the head pose, which is needed to apply the shape prior onto the observations. In this work we propose the use of a hierarchical shape prior that is modeled to have at its root the shape described by the centroids of the feature constellations and its leaf nodes the shapes for the feature constellations.

In Japan, most cell phones have the capability of receiving GPS signals. Besides the cell phone there are many other devices that can keep track of GPS data, including the GPS device made by Sony. If all this GPS data is collected and organized, one could get a good overview of peoples whereabouts. That is why the goal of this project was to develop a tool that keeps track of peoples movements, share a visualization of these movements with other people and allow for other means of social interaction, such as chatting, messaging, finding other people and so on. To realize this goal we created the Sococa system. The name stems from the full name we had devised at first, Socialocalize. Additionally, the Japanese phrase 'sokoka' can be taken to mean something akin to 'over there?'. The system consists of a client part and a server part. The client program can parse and upload users' locations to a server in real time. The server has an interface which also allows other programs to upload their location data. Because the user doesnt always have an internet connection available, it is also possible to upload files with location data at a later time. This can be done on the website or with the client program. The server hosts a website on which a map with all positions of the users are shown. When a user uploads a new location to the server, this is directly updated on the map. The map also has the capability to show animated routes the user took in the past. The website has some social networking features. It is possible to create an account and add some personal information to it. It is possible to look up ones friends using the search option and add them as friend on the website. After they are added, users can chat in the global chat or send private messages to each other. To ensure the users' privacy, the user can choose to limit the availability of his location data to his friends or only to himself. The most important things the current system is lacking are full OpenId support and support for the Opensocial API. OpenId is a service which allows users to have only one account for the whole internet. This technology is supported by the website, but not by the client program. The OpenSocial API links social networking sites to each other, i.e. friends can be easily imported from one site to another. Because of the complexity of this technology, we have chosen not to implement this.

Deze thesis beschrijft de ontwikkelling van software die op basis van FMCW-radargegevens een voetbal kan detecteren en volgen. De problemen die zich daarbij voordoen zijn onder andere het maken van onderscheid tussen voetballen en andere objecten, het volgen van een voetbal en het omgaan met complexe situaties zoals die zich in het voetbalspel voordoen. De software is opgesplitst in drie delen, namelijk objectdetectie, objecttracking en objectclassificatie. Goede objectdetectie is essentieel voor het kunnen tracken van objecten, en goede tracking is belangrijk om objecten juist te kunnen classificeren. Tot slot vindt er besluitvorming plaats op basis van de door de objecttracking en objectclassificatie verzamelde gegevens. Een studie is gedaan naar deelconcepten die mogelijke oplossingen vormen voor de verschillende delen in het eindontwerp. Tevens zijn er enkele deelconcepten onderzocht en getoetst. Op basis van het gedane onderzoek en de gevonden resultaten is er software voor een prototype in Matlab ontwikkeld en gelueerd. Het prototype is vooral bedoeld om verschillende oplossingen voor de verschillende deelsystemen te kunnen evalueren. Het detecteren van objecten vindt in het prototype plaats door middel van piekdetectie. De drempel die hierbij toegepast wordt, kan vast ingesteld worden maar kan ook adaptief bepaald worden met behulp van een CFAR detector. Voor het volgen van objecten kan gebruik gemaakt worden van zowel de dichtstbijzijnde nabuurmethode als een methode die gebruik maakt van een Kalman filter. Objectclassificatie vindt in het prototype plaats op basis van de mate van constantheid van de reflectie van een object. Andere alternatieven zoals het detecteren van een verschil in reflectie bij verschillende polarisatierichtingen, of het gebruik van RFID-chips zijn niet toegepast, omdat deze niet geschikt leken in de gebruikte opstelling of niet beschikbaar waren. Uit de evaluatie van het prototype blijkt dat in sterk vereenvoudigde spelsituaties het systeem aan van de twee gestelde nauwkeurigheidseisen voldoet. Het beeldverwerkingsysteem wordt in meer dan 75% van de gevallen terecht voor een naderende bal ingeschakeld en niet voor een ander naderend object. Het komt echter te vaak voor dat het beeldverwerkingsysteem niet ingeschakeld wordt terwijl er wel een bal aan komt. De beste combinatie om objecten te volgen bestaat uit het Kalman filter met een CML-CFAR detector. De gebruikte classificatiemethode blijkt niet goed genoeg te zijn om gebruikt te worden in de uiteindelijke implementatie. Het prototype classificeert 80% van de ballen juist in situaties waarin de bal duidelijk voor de radar zichtbaar is. Wanneer de bal zich echter in de buurt van een ander object bevindt, treden er classificatiefouten op.

Since the stone age the human race seeks for strategies to extend its viewing range. With the rise of technology in the twentieth century, cameras are found to be a very useful tool to survey a large area with limited resources. With an increasing numbers of cameras, it becomes more difficult to watch every monitor and prevent incidents in the surveillance area. For the last decades, research seeks for possibilities to automatize the process of video surveillance. For this thesis, we approach the surveillance task from the human perspective: we try to emulate what human operators do when they watch the monitors. To perform this task, state-of-the-art techniques from Computer Vision and Artificial Intelligence are applied. An object tracking technique called P-N Learning is used that enables the tracker to learn from its mistakes. The Java Agent Development Framework (JADE) is used to enable communication between agents in the FIPA Agent Communication Language standard. A surveillance system model is designed that detects suspicious behavior in a non-public area. Its task is to alert the operators about suspicious events to give them the chance to investigate and take action. Two prototype applications are implemented and experiments are conducted to show the performance. We showed the proof-of-concept of a system which is able to emulate operators and can potentially outperform a human being. Once the system knows what is considered suspicious behavior it can be automatically detected.

This paper introduces an alternative approach to following persons for robotic systems in a domestic environment; Flood Fill on Depth (FFOD) data. Other related work is shown: tracking using a Lucas Kanade tracker, tracking using OpenTLD (Tracking, Learning, Detecting) and tracking using a state of the art particle filter. These algorithms are all processing RGB images, while this paper suggests an alternative: image processing on depth data. The FFOD algorithm is evaluated by checking the introduced requirements and benchmarking using the Bonn Benchmark on Tracking (BoBoT). FFOD passed all critical requirements but lacks a detection system to regain a person when lost. The BoBoT benchmark resulted in a precision of 80:09%, which is almost 10% higher than a state of the art particle filter algorithm. Hereby showing that person following using depth data is an interesting approach and should be further researched.

Object Tracking has been a very active area in the field of C omputer Vision. Over the years, a variety of approaches have been put forth to solve this problem and though many of them have demonstrate considerable success none of them have been completely successful. With more methods being written each day, the evaluation of such systems becomes a very important task. If an evaluation system exists that is able to point out specific flaws in the stage of development, it can lead to a very robust and improved algorithm. This work attempts to create such an evaluation framework. Given an algorithm that detects people and simultaneously tracks them, we evaluate its output by considering the complexity of the input scene. Some videos used for the evaluation are recorded using the Kinect sensor and a benchmark dataset from the PETS workshop is also used. To analyze the performance of the tracking system,the reasons due to which the algorithm might fail are investigated and quantified over the entire video sequence. A set of features called Scene C omplexity Measures are obtained for each input frame. The variability in the algorithm performance is modeled by these complexity measures using various regression models. From the regression statistics, we show that we can compare the performance of two different algorithms and also quantify the relative influence of the scene complexity measures on a given algorithm.

Tracking technologies such as GPS, mobile phone tracking, video and RFID monitoring are rapidly becoming part of daily life. Technological progress offers huge possibilities for studying human activity patterns in time and space in new ways. Delft University of Technology (TU Delft) held an international expert meeting in early 2007 to investigate the current and future possibilities and limitations of the application of tracking technologies in urban design and spatial planning. This book is the result of that expert meeting. Urbanism on Track introduces the reader to the basics of tracking research and provides insight into its advantages above other research techniques. But it also shows the bottlenecks in gathering and processing data and applying research results to real-life problems. Urbanism on Track showcases tracking experiments in urban studies, planning and design – from pedestrian navigation in Austria to Danish field tests, from TU Delft's Spatial Metro project to MIT's Real Time Rome and last but not least the Sense of the City project realised in Eindhoven. Urbanism on Track discusses the relevance of tracking for policy making, the possibilities of a new cartography and the implementation of tracking technologies in urban design and planning. This makes Urbanism on Track a unique book, setting the agenda for the structural embedment of research using tracking technologies in urbanism.

By many, code cloning is nowadays recognized as a threat to the maintainability of source code. Many clone detection strategies have been proposed and a considerable number of removal strategies, mostly based on refactoring techniques, has been shown. However, recent research has showed that clones can often not be removed easily and other strategies, based on clone management need to be developed. In this thesis, a clone management strategy based on dynamic inferring of clone relations based on monitored clipboard activity is described. A tool is introduced that is able to track live changes to clones and offers several resolution strategies for inconsistently modified clones. The adequacy, usability and effectiveness of this Eclipse plug-in have been studied in an experiment, the results of which show that developers actually do see the added value of such a tool but have very strict requirements with respect to its usability.

This research is focussed on a project management technique called fast tracking. Fast tracking is the integration of design and construction phases by arranging work packages of the total project and overlapping the design and the construction of these packages as well as overlapping the total work packaging. The relocation of the faculty of architecture, after the fire in May 2008, was the case study used for this research.

High speed feature point detection and tracking is very demanding for many realtime computer vision applications. In existing work, the commonly used feature point detection algorithms like Harris and KLT (Kanade-Lucas-Tomasi) and feature tracking algorithm (Pyramidal-KLT) were redesigned to increase the performance by reducing the algorithmic complexity, resulting in the Low Complexity Corner detector (LOCOCO) and Robust Low Complexity Feature tracking (RLCT) algorithms. To attain further speedup, this report proposes the implementation of these low complexity detection and tracking algorithms on a massively parallel architecture of the modern graphics processing units (GPUs) using Compute Unified Device Architecture (CUDA). In the computing domain, due to semiconductor scaling limits and associated power and thermal challenges, combined with the difficulty of exploiting greater levels of instruction level parallelism, a paradigm shift is happening from a single core to many-core processors and massively multi-processing platforms. High performance is now available on single-chip commodity GPUs. Moreover, GPUs are no longer limited to graphics applications, but are emerging as usable general purpose computing devices. Advancement in such platforms, are making many computational intensive problems that were solvable only on supercomputing systems, to be computed on desktop systems, at a reduced price, and lower power requirements. The arrival of this new generation of low-cost high performance computing platforms presents both numerous opportunities and challenges. In this report, we present the use of such high performance many-core GPU platforms to obtain speedup by mapping general purpose computations to massively parallel architectures. It is observed, when properly executed, GPU adaptation of algorithms can result in significant savings in computation times. For an image size of 960x960 pixels, the low complexity corner detector and robust low complexity feature tracking algorithms are factor of 16 and 25 times faster on a GeForce 280 GTX GPU than the corresponding implementation on an Intel Core 2 Duo, 2.66 GHz, and 2GB RAM CPU.

In spite of numerous road management schemes and developed infrastructure the society nowadays still faces the problem of highly congested roads due to the increasing traffic demand. The focus of this thesis is to develop a complex and integrated system that addresses the challenges of dynamic traffic assignment in modern times. To reach this goal we had to study the theory behind such systems, come with new ideas and implement them into a real world human-centered traveler information system. Taking into account state of the art systems in this field we built a design for a Personal Advanced Traveler Assistant (PTA). The main purpose of PTA is to give routing advices depending on the users preferences and the available capacity in the network. The system incorporates a dynamic traffic algorithm that employs a prediction model. The prediction model should use current traffic updates and the routes of the guided cars in order to achieve an accurate prediction of the future traveling times. Next step was to implement a prototype. This is available through multiple interfaces like smart phones or desktop applications. The prediction model that the algorithm uses is based on historical data. Given the time constraints the prototype uses now only historical data but it can be easily extended to include live updates. This allowed us to build a prediction model that considers typical traffic dynamics in order to estimate the traveling time. The performance measure of the traffic assignment algorithm is the shortest traveling time. The algorithm that we implemented is a time dimensional-extended version of the Dijkstra algorithm. The algorithm is tested on real data collected from the highways in The Netherlands. We evaluate the performance of the algorithm by comparing it to two versions of the static Dijkstra algorithm. To conclude, we successfully implemented a working prototype that uses various technologies such as Java, the Open Street Map API for rendering the map or J2ME for the mobile phone client. The prototype that we have built represents a working proof of concept for a dynamic routing assistant. One advantage of the structure that we chose to implement is that each component can be further on extended independently. In this way we showed that such a system is feasible but we also left the possibility for different parts of the system to be extended into more advanced applications.

In recent years, interest in correspondence between virtual environments and realworld events has grown considerably. We investigate different methods of integrating position and orientation data of moving real-world objects and individuals into a virtual environment, and of simulating their interaction with virtual objects and individuals. This simulation requires accurate measurements of positional data of dynamic real-world objects. This data is then represented in a virtual copy of the real-world environment. One or more user-controlled virtual objects are added and interaction with this environment is made possible. We investigate different ways of manipulating the avatars of real-world objects to simulate this interaction. Each such method must be guaranteed to smoothly recover to an exact copy of the real-world situation in all but the most extreme circumstances. An accurate prediction of both the future state of the real world and the users avatar is needed to plan these manipulations ahead of time. Different prediction methods are researched, for both the real-world objects and the user object. We give an overview of the flow and manipulation of information through a system that allows for such simulated interaction, and conclude that integration of and interaction with real-world objects in a virtual environment is very well possible, but that each application may have specific needs based on parameters such as relative velocity and maneuverability. This results in the necessity of selecting a specific solution for each individual stage of the information flow for each application.

In onderhavige studie is de meettechniek "Particle Tracking Velocimetry" (PTV) toegepast ten behoeve van het onderzoek naar ondiepe waterstromingen. Het doel van het onderzoek was de meettechniek beschikbaar te krijgen voor studies aan ondiep-waterstromingen en de geschiktheid van de methode voor het meten van grootschalige structuren aan te tonen. Eerst is er een literatuuronderzoek naar de PTV-methode geweest. De methode meet verplaatsingen van deeltjes in een medium over een bepaalde periode. Hiermee kunnen watersnelheden worden bepaald. Vervolgens is de PTV-methode gebruiksklaar gemaakt voor de metingen. Na het opzetten van de PTV-methode is een literatuuronderzoek gedaan naar de werve1s in een menglaag. Hieruit is gekomen dat wervels zijn aan te tonen door samenhang in fluctuaties van snelheden in ruimte en tijd. Daama is er een post processing methode opgezet om de PTV metingen om te zetten naar fluctuaties ten opzichte van een gemiddelde. Om deze methode uit te voeren is er een post processing programma geschreven. Als laatste is bij de opzet van de methode een uitgebreide foutenanalyse geformuleerd. Hiermee is aangetoond dat aanwezige fluctuaties gemeten kunnen worden. Er zijn in twee opstellingen metingen met behulp van de PTV-methode verricht. Eerst is er in een havenmodel gemeten en vervolgens in een ondiep-watermenglaag. De opstelling in het havenmodel is gebruikt om de methode op te zetten en bekend te raken met de methode. De opstelling met de ondiep-watermenglaag is gebruikt als test voor het meten van de structuren. Hierin zijn eveneens LDA metingen verricht waarmee de PTV-methode is vergeleken. Uit de resultaten van de metingen is gebleken dat: het opzetten van de meetmethode is gelukt; het meten van turbulente structuren in de menglaag niet is gelukt. Dit werd veroorzaakt doordat de afstand tussen de deeltjes groter is dan de grootte van de samenhang in de fluctuaties. De belangrijkste oorzaak hiervan was te ondiep water. Hierdoar konden de structuren zich niet in voldoende mate ontwikkelen om deze te kunnen waamemen met deze methode. Aanbevelingen voar de meting aan wervels: voer de experimenten uit met deeltjes die kleiner zijn ell dichter bij elkaar liggen, zodat de wervels meetbaar zijn. Aanbevelingen voor de metingen: er zijn interessante effecten in de metingen zichtbaar geweest, deze kunnen met deze methode goed worden onderzocht.

During the surgery, the lighting unit often needs to be manually adapted. However, the manual manipulation may cause loss of concentration, loss of time and increased risk of infections. To improve the manipulation, a semi-automated lighting system commanded by the surgeon is suggested. The commands are given via a pointer device to indicate the desired location and orientation of the light beam. The pointer device is tracked in space with a low-cost infrared 6DOF real-time system based on Nintendo Wii-Remotes. In this study, the precision and accuracy of the tracking system was tested with static markers, dynamic markers and with the pointer device. Moreover, the wound reconstruction was evaluated. The results showed that the tracking system is sufficiently precise and repeatable. On the other hand, the position accuracy was lower than expectations, especially with the pointer device. It is concluded that the method of collecting relevant parameters with a pointer device using a low-cost tracking system is promising for the surgical lighting control application when a proper mapping of the system coordinates to the real world coordinates is achieved.

Peer-to-peer (P2P) technology allows us to create self-organising and scalable systems. The distributed nature of these systems requires a solution for finding interesting peers. In the context of P2P file sharing, finding peers who are downloading the same file is referred to as the swarm discovery problem. In BitTorrent, this problem is solved using a central server, called a tracker. This central component hinders the scalability of BitTorrent. We have designed a distributed swarm discovery algorithm, called 2-Hop TorrentSmell. It is composed of two parts. The first part builds on top of an existing keyword search system to find peers who have recently downloaded a certain file. The second part consists of an algorithm called RePEX, which allows a peer to stay in touch with swarms it is no longer in by periodically recontacting previously encountered swarm members. Our RePEX algorithm leverages a widely used BitTorrent extension for swarm discovery called PEX. We have conducted a study to understand the reliability and usability of PEX to optimize the design of our solution. We have implemented the RePEX part of 2-Hop TorrentSmell as an addition to the Tribler P2P network. For the evaluation of our RePEX algorithm, we have tested it on the 10 largest swarms on a public tracker. In addition, we have deployed the algorithm in a beta version of Tribler and let it run on swarms the user has downloaded from. Evaluation results show our algorithm is scalable and effective in popular swarms.

How can Twitter be exploited to fulfill information needs of users on Twitter during incidents? In this Thesis this question wil be investigated and a strategy to locate tweets which fulfill information needs will be introduced. First, techniques are proposed to automatically detect incidents given an unstructured incident data source and subsequently tweets which report on these incidents are tracked. An interpretative analysis is carried out to derive and classify typical information needs during incidents. Furthermore, a metric is proposed to compute a tweet's relevance by measuring its informativeness and the user's trustworthiness. Finally, tweets are ranked according to this metric such that for each information need at a particular time the most relevant tweets can be found.

The Nuna 6 is a solar racing vehicle that solely relies on solar energy from the sun. Every year, the team seeks ways to improve the performance of their car. One way to improve the performance is to maximize the power output of the solar panels on top of the car. Maximizing the power output can be done with a 'Maximum Power Point Tracker'. The aim of this thesis was to develop an improved, distributed 'Maximum Power Point Tracking'-system which optimizes the power efficiency of the solar panel array of the Nuna 6 solar racing vehicle. To prove that the proposed distributed topology is more power effificient when compared to a central tracking topology, simulations of the total Nuna 6 electrical system were performed. Based on the simulation results, together with Nuna 6 specifications, a DC-DC boost converter was designed. Validation of the design was done by simulation with the Nuna 6 model. After validation, a breadboard proof-of-concept was built. The proof-of-concept was successfully tested and compared with earlier simulations. The system design process was evaluated and recommendations for further study and future real-life implementations were formulated. The simulation results prove that the proposed distributed tracking system is as much as 40% more efficient in large insolation differences and 10% in small insolation differences. The system excels when insolation differs, however it is slightly less efficient when used with equal insolation on every panel. The developed proof of concept demonstrates a functioning maximum power point tracker and DC-DC boost converter. The power efficiency of the boost converter was found to be between 95:8% and 98:5%, with an efficiency of 97:1% for the rated input power of 200W.

Currently, the Nuon Solar Team is building their new solar car, which makes use of solar panels. Solar panels have a Maximum Power Point (MPP), which, when operated at that point, ensures the maximum available power is obtained from them. In this thesis, different options were explored to solve the problem of tracking the MPP of a solar panel. The focus of this thesis was the software part of tracking the MPP and the goal of this thesis was to implement the most efficient algorithm that works in fast changing levels of irradiance and when thesolar panels are partially shaded. In order to realize this goal, we first did a literature survey to learn about the available algorithms and their respective advantages and disadvantages. Subsequently, we chose the algorithms which had potential and we could realistically implement. Those algorithms were P&O, In ond and their adaptive variants. We simulated those algorithms for their efficiencies in Simulink and implemented them onto a microcontroller. Lastly, we made an experimental setup and measured the algorithms for their efficiencies. The results showed that based on the simulations, the adaptive InCond algorithm is the most efficient algorithm, also in fast changing levels of irradiance. As the simulation did not simulate partially shaded solar panels, we can not make any conclusions about the performance of the different algorithms in that case. In the experimental setup, we verified that the controller and all implemented algorithms worked correctly. However, we were not able to verify all the simulation results, as we could only only test the sudden shading condition. The MPPT was able to reliably track the MPP of a solar panel, depending on what algorithm was used. Some algorithms were more susceptible to noise than others, and eventually we concluded that the adaptive P&O algorithm performed best in the experimental setup, because it is least susceptible to noise and has the advantages of an adaptive algorithm. We did not measure the efficiency of the MPPT. However, based on the simulations and the measured efficiencies of the other subsystems of the MPPT, we are confident that we succeeded in designing and implementing an MPPT algorithm with an efficiency of at least 95%.