|Dozent/in||Prof. Dr.-Ing. Jochen Schiller|
|Institution||Freie Universität Berlin|
Institute of Computer Science
Computer Systems and Telematics
Students have to meet all deadlines listed in the following schdule. Otherwise s/he will lose the right to take part in the final presentation. Attention at all final presentations is mandatory.
19.10.2012: Choose topics as described below, send the short list to firstname.lastname@example.org.
26.10.2012: All participants will have a topic assigned and contact their advisor.
16.11.2012: Hand in a preliminary outline and reference list to your advisor and email@example.com.
11.01.2013: Hand in your final report to your advisor and firstname.lastname@example.org.
01.02.2013: Hand in your slides to your advisor and email@example.com.
The seminar will take place February, 14th and 15th, in room 137 according to the following schedule:
BSc bzw. Vordiplom, Telematik
We offer the following topics for this course. You have to choose 3 topics currently not assigned to anyone (N.N.) and send your short list to firstname.lastname@example.org. Place the topic you are most interested in at position one etc.We will assign you to a topic - if possible - according to your preferences. In case of a collision and your list is exhausted the date/time of your registration determines the order (FCFS). We will tell you the topic or the failure of the assignment.
Public spaces bring together large numbers of people on limited space to use security-sensitive infrastructure. Monitoring large crowds can provide guidance in case of unexpected events (e.g., a mass panic). The challenge is to detect mass panic or abnormal behaviour of the crowd before dangerous situation occurs. In this topic the state of the art in mass panic detection should be presented. Existing crowd monitoring systems should be analysed in such aspects as covered scenarios, used sensors, collected data, general algorithmic idea. The focus lies on the analysis of distributed sensing event detection systems.
Infrared imaging is becoming increasingly important and widely used in surveillance systems for pedestrian and crowd monitoring. In this topic existing algorithms to monitor crowds and estimate their density for classical colour image processing should be analysed in terms of applicability in thermal domain. The algorithms for infared image processing for crowd monitoring should be evaluated with respect to collected, processed and stored data, complexity in time and space.
With the rise of the Internet of Things, the number of wireless devices is dramatically increasing. Several of these devices will be powerful enough to sport a WiFi chip and run a variant of the Linux OS. An example of such a WiFi System on a Chip (Soc) is the Carambola open hardware platform. This particular SoC runs the OpenWRT operating system, which is tailored to embedded networking devices. Other projects and guides have been initiated, such as the embedded versions of Debian and Gentoo, and Raspbian for the popular Raspberry Pi platform. Another interesting approach is the Damn Small Linux (DSL) project. For this topic, you will assess the current state of the art of Linux variants for embedded systems. You will compare the existing approaches based on the technical capabilities, their user base, level of support, and liveliness of the project. The goal of your report should be, that you can give a profound opinion which Linux variant is most suitable for a certain embedded systems.
The rise of the Internet of Things is characterized with an immense increase of wireless devices. Self-organization provides a new paradigm for such massively distributed systems. However, there is no clear definition of self-organization and a standard approach to build these types of systems. It is your task to study what approaches for self-organizing network systems exist and what the basic principles of these systems are. Hence, for your report various publications have to be read and the important parts extracted. The following questions have to be answered: What definitions of self-organization (within respect to the area of computer science) exist? How can self-organizing systems modeled and studied? What algorithms/architectures for self-organizing systems exist?
Wireless Sensor Networks are optimized to transmit only few data, caused by high cost of energy per bit. Corresponding transceiver are optimized for idle time. What happens in the use-case of distributing new firmware within the network? In networks larger ten sensor nodes spread in a small area a 'wired' updating is an exhausting process. Present current approaches and optimizations for over-the-air updates in Wireless Sensor Networks.
Successful data delivery within the Internet is based on a correct mapping of the receiver's IP prefix to the legitimate autonomous system of the prefix. The only protocol that is deployed to dynamically exchange routing data between Internet domains, namely the Border Gateway Protocol (BGP), follows a loosely coupled trust model. In the original BGP specification there is no mechanism to verify the ownership of an advertised IP prefix. This makes it quite vulnerable against prefix hijacking, i.e., an illegitimate BGP peer claims to own the prefix. However, also misconfiguration may result in the same effect. The goal of this work is to review and discuss related work that tries to identity prefix hijacking. A good starting point is the paper by Shi, which will be published at IMC 2012.
To provide reliable event detection in an eStadiums a distributed sensor network needs to gather environmental data in an efficient way to prolong the networks lifetime. These include detecting: the occurrence of loud cheering or booing, which indicates which video clips of plays will be most frequently downloaded by fans; excessive delays in lines outside concession stands and restrooms. Distributed event detection in multi-hop clusters of sensor motes have to deal with both measurement and communication errors. Analyze a given optimization tool that enables the operator to find the tradeoffs between detection-performance, delay, energy-usage, etc.
Paper: Low-Complexity Algorithms for Event Detection in Wireless Sensor Networks Xusheng Sun and Edward J. Coyle
Find scientific publications covering systems used for structural health monitoring on buildings with a focus on bridges. Find out which type of sensors are used, which data is collected, and how the data is analyzed. Try to focus on systems which act autonomously to a certain extend and use some kind of network to communicate.
The Cramér Rao bound (CRB) is mathematical method to find the lower bound of the variance of estimators using deterministic parameters. In Computer Science with a special focus on localization techniques, such as GPS, this method is used to calculate the worst-case accuracy of localization algorithms. Understand what the CRB delivers and how it is calculated and when it can be applied. Try to explain your findings in simple terms.
Note: You should enjoy doing math/statistics when you choose that topic.
Maschinelles Lernen (Machine Learning) hat ein breites Anwendungsspektrum in vielen Bereichen der maschinellen Sprachverarbeitung, der Suchmaschinen und der Robotik. Man unterscheidet zwischen Maschinelles Lernen und Data Mining. Der Schwerpunkt bei Data Mining liegt bei der Entdeckung von Mustern in großen Datenmengen, die nützlich sein können. Bei umfangreichen Datensätzen existieren meistens viele Beziehungen zwischen den Daten. Das Maschinelle Lernen ist der algorithmische Teil von einem Data Mining Prozess und stellt eine Teildisziplin der künstlichen Intelligenz dar. Es befasst sich mit Techniken und Algorithmen, die den Lernenden (Rechnern) aus Beispieldaten das Lernen ermöglicht. Die wichtigsten Lernklassen sind: überwachtes Lernen (supervised learning), semi-überwachtes Lernen (Semi-Supervised Learning), unüberwachtes Lernen (unsupervised learning) und bestärkendes Lernen (Reinforcement-Learning). Ziel dieser Arbeit ist die Untersuchung von Algorithmen für Klassifikationsprobleme, die auf semi-überwachtes und bestärkendes Lernen basieren. Die Algorithmen sollen auf Rechenaufwand, Spei-cherbedarf, sowie Abhängigkeit von Trainingsdaten untersucht werden. Auf Wunsch kann eine Einführung in die Mustererkennung angeboten werden.