Sebastian Höffner: Gaze Tracking Using Common Webcams, Master's thesis. Osnabrück University, February 1, 2018.
pdf download, approx. 12.5 MB
Eye and gaze tracking have long been methods to study visual attention. Many devices for gaze tracking are expensive and require specific setup and calibration procedures. For many gaze tracking setups, it is even mandatory to use multiple computers, for showing stimuli and for tracking gaze, respectively.
Today, modern laptops are equipped with enough processing power to process a video stream live. Additionally, many laptops come with a built-in webcam for teleconferencing and video chats. In this thesis, the possibility of performing gaze tracking using a calibration free, feature-based approach on laptops using built-in webcams is explored. To try the model, the free and open source software library Gaze is implemented and evaluated.
It is shown that Gaze reaches very good eye tracking capabilities and manages to be easily usable and extendable. Its gaze tracking abilities are still to be improved, but because of its modular structure existing solutions like pre-trained neural networks can be integrated to leverage their strengths.
Sebastian Höffner: Probabilistic Robot Localization in Continuous 3D Maps. Bachelor's thesis. Osnabrück University, November 10, 2014.
pdf download, approx. 1.02 MB
Robot localization in known environments is an important topic, especially in the field of mobile robotics. Whenever a robot is turned on or relocated, it has to (re)locate itself in its environment. Several approaches exist for robots to find their three-dimensional pose in two dimensional raster maps. A popular algorithm for the localization task is the Augmented Monte Carlo Localization (AMCL) which is described in details by Thrun et. al. [TBF05]. A localization is successful when the robot found its current three dimensional pose, that is its two dimensional position and its one dimensional orientation.
Mobile robots have to work not only in two, but also in three dimensional environments. This means they can have six dimensional poses because their position consists of three dimensions and so does their orientation. In this thesis I will derive an algorithm based on AMCL to solve the localization problem in such three dimensional environments and find the six dimensional pose. Therefor I will no longer use raster maps but polygon meshes as continuous maps. For the sensor model I will use a ray tracer, which simulates sensor data comparable to real world distance sensors like laser scanners or depth cameras.
After deriving the algorithm and running some simulations with it, I will take a look out about how the algorithm can be developed further.
- Summer 2017: Lecturer for Basic Programming in Python
- Winter 2016/17: Tutor for Computer Vision
- Summer 2016: Tutor for Machine Learning (Creation of exercise sheets)
- Winter 2015/16: Tutor for Computer Vision
- 2016-2018: Osnabrück University
- Institute of Cognitive Science: M. Sc. in Cognitive Science (awaiting grading)
- 2014-2016: Osnabrück University
- Institute for Computer Science: M. Sc. in Computer Science (discontinued)
- 2013-2014: Middle East Technical University, Ankara
- Computer Engineering Department: Erasmus Exchange Semester
- 2011-2014: Osnabrück University
- Institute of Cognitive Science: B. Sc. in Cognitive Science
- 2002-2011: Städtisches Gymnasium Sundern
- Higher education entrance qualification (Abitur)