Theses

Theses

The Human-Computer Interaction group offers various theses (e.g. Bachelor's or Master's theses) and projects (e.g. Bachelor's projects). All topics include the conceptual development, implementation and evaluation of a human-computer interaction problem and are presented in the HCI colloquium. If you would like to start a project or thesis with us, please contact Marvin Strauß and state the topics you are interested in, the type of work and give a brief overview of previous projects and experiences.

Jonathan Liebers: Novel Authentication and Identification Schemes using Machine Learning & Deep Learning, Biometric Authentication in Virtual and Mixed Reality, Wearable and Embedded Tools for User Identification.

Alia Saad: Usable Security and Privacy, Behavioral Biometrics in Mixed Reality and on Mobile Devices, Spoofing Authentication.

Jonas Keppel: Interactive Health Applications, Activity Motivating Application, Explainable AI, Exergames, Indoor Cycling Gamification.

Max Pascher: Human-Robot Collaboration, Intervention Strategies/ Interfaces, Multimodal Input & Feedback Technologies, Augemented/Mixed/Virtual Reality, Assistive Technologies.

Carina Liebers: Control and Agency in Human-AI Interaction, Machine Learning & Deep Learning in Generative AI, often with Virtual Reality/ AR.

Nick Wittig: Augmented Reality Technologies, Interaction with Augmented Reality, Learning and Education.

Marvin Strauß: Designing and Evaluating Scalable Privacy Awareness and Control User Interfaces for Mixed Reality, Development of mechanisms to protect users' and bystanders' privacy.

Niklas Pfützenreuter: Enhancing user control for Generative AI, Explicit and implicit interaction with Generative AI.

Roman Heger: Hardware prototyping and haptic feedback with focus on pneumatics.

Please click on the symbols at the entries of the theses to get further information: 

Filter:
  • DeepFake Anonymization - A Novel Approach to Privacy Preservation in Video ContentDetails

    Abstract:

    The advent of deep learning technologies has significantly impacted fields such as computer vision and digital content creation. An innovative application of these technologies lies in the realm of privacy preservation in video content. This master's thesis explores the use of deep fake technology as a tool for anonymization in videos. The study focuses on developing a deep learning-based framework capable of replacing or modifying identifiable features in video content, like faces or voices, with realistic synthetic alternatives. The primary objective is to ensure the privacy and anonymity of individuals in video data while preserving the integrity and continuity of the visual content. The research encompasses technical aspects of deep fake creation and potential applications in various domains, including media, surveillance, and online content creation. Importantly, the thesis includes a comprehensive user study to evaluate the effectiveness and perception of these anonymization techniques. The study will also address the challenges and limitations of using deep fakes for anonymization and propose solutions to mitigate potential risks associated with this technology.

    Research Objectives:

    1. Development of a deep learning framework capable of generating high-quality deep fakes for anonymization in videos.

    2. Evaluation of the effectiveness of deep fake anonymization in preserving privacy without compromising the authenticity of the video content.

    Methodology:

    - Literature Review: Conduct a comprehensive review of existing literature on deep fakes, privacy preservation techniques in videos, and relevant deep learning models.

    - Framework Development: Development of a deep learning framework for creating deep fakes specifically tailored for anonymization purposes. This would involve training models on diverse datasets to ensure generalizability.

    - Evaluation: Test the framework on various video datasets to assess the quality of anonymization, the preservation of content integrity, and the robustness of the anonymization against re-identification attacks. An additional critical step in the methodology will be conducting a user study. This study will gather qualitative and quantitative data on user perceptions of the anonymized videos. The study aims to understand how viewers perceive the anonymized content in terms of realism, the impact on the video's message or content, and their overall acceptance of using deep fakes for privacy preservation. This feedback will be crucial in refining the framework and understanding its potential impact on viewers.

    Expected Outcomes:

    - A fully functional deep learning framework for video anonymization using deep fakes.

    - A comprehensive evaluation of the framework's performance and perception in a user study and its limitations.

    This thesis aims to contribute to the field by providing a novel approach to privacy preservation in videos and sparking discussions on the responsible use of deep fake technology in safeguarding personal privacy.


    Master Thesis, Computer Science, Tutor: Marvin Strauß, M.Sc.
  • Designing HandNet: A Deep Learning Architecture for User Identification via Hand Movement in Augmented and Virtual RealityDetails

    In the fast-evolving world of augmented (AR) and virtual reality (VR), identifying users accurately and effortlessly through their digital interactions is crucial. This thesis should introduce HandNet, a pioneering deep learning architecture designed to leverage hand-tracking data for implicit user identification. Unlike traditional authentication methods that rely on explicit inputs such as passwords, HandNet identifies individuals based on their unique hand movement behavior during interactions with universal interface elements like buttons and sliders. Through a precise design and extensive optimization, including hyperparameter tuning and the evaluation of various hand-tracking preprocessing functions, HandNet should achieve remarkable user identification accuracy. This thesis not only showcases the potential of behavioral biometrics as a seamless authentication method in AR and VR but also lays the groundwork for further exploration into personalized digital interactions. Therefore, the thesis primarily involves creating and testing a deep learning architecture based on an available data set.

    Details:

    • This thesis is suitable for students pursuing a bachelor's thesis; conducting the work and writing the thesis can be performed in German or English.
    • Students should be proficient with Python and have an inherent interest in and skills in machine learning.
    • This thesis will primarily use Python, Pandas, Jupyter Notebooks, Git, and Tensorflow or Torch.
    • Before assigning the thesis, there will be a trial task. The trial task essentially covers the IJHCI 2022 publication. Here, a simple deep-learning-based classification algorithm should be created to identify users (i.e., predict the participant id as y). In turn, x is the hand-tracking data. The training data comes from the first session and the validation/test data from the second session. The data set can be found online.
    • Please contact Jonathan Liebers via e-mail to schedule an appointment if you are interested in this thesis.

    Bachelor Thesis, Computer Science, Tutor: Jonathan Liebers, M.Sc.