Teaching

Generating Evacuation Tours through Deep Reinforcement Learning

[Type: Master Thesis, Focus: Computer Graphics, Machine Learning]

​

Digital twins of indoor scenes are today more and more available in several forms such as 3D models, games and through building information modelling tools. In this project we want to leverage 3D building models for evacuation safety where evacuation tours are automatically generated for the scope of training people on evacuation paths in case ofemergency. The computed evacuation tours are used to guide and train people in case of emergency evacuation. Such tours could also be enhanced with inclusive features (e.g., an automated generation of tours that include a vocal description of what is seen, so that could be used to help people with visual impairments learning about evacuation paths). This project will focus on developing a methodology for automating the generation of trajectories of evacuation paths based on deep reinforcement learning. An agent is trained to learn how to automatically control a camera to generate an evacuation path that can be used for safety purposes.

​

Wildfire Evacuation Planning

[Type: Bachelor or Master Thesis, Focus: Computer Vision, Machine Learning]

​

This project aims to develop a system for estimating population distribution from satellite imagery to improve wildfire evacuation planning. Using machine learning and geospatial data, the goal is to extract features such as buildings, roads, and land use to infer population densities. The challenge is in processing satellite imagery to differentiate the involved object categories. We aim to jointly use data sources like satellite imagery and census information to provide accurate predictions to support emergency response and optimize evacuation strategies in wildfire-prone regions. The project will leverage computer vision, geospatial analysis, and data fusion techniques to enhance the accuracy and usability of the predictions.

​

Speech-based Robot Control

[Type: Bachelor or Master Thesis, Focus: Machine Learning, Robotics]

​

This project aims to integrate a large language model (LLM) with our Boston Dynamics Spot robot to enable intuitive speech-based control. By leveraging speech recognition and natural language processing, the system will interpret spoken commands, translate them into executable actions, and interface with Spot’s API for real-time operation. The research focuses on optimizing command interpretation, ensuring safe execution, and evaluating responsiveness in real-world tasks like navigation and object interaction. This work enhances human-robot interaction, making advanced robotics more accessible for applications in automation, security, and emergency response.

​

​Modeling and Rendering of Underwater Environments

[Type: Bachelor or Master Thesis, Focus: Computer Graphics]

​

This project aims to create photorealistic underwater scenes that can be explored with a robotic agent. The core focus is on developing sophisticated computer graphics techniques to accurately simulate underwater optical properties like light refraction, scattering, and caustics, ensuring detailed and lifelike visuals. The project also involves programming a robotic agent to move through these scenes, interact with underwater terrain, and engage with marine flora and fauna, emulating real-life underwater exploration.

​​

Procedural Leaf Placement and Rendering

[Type: Bachelor or Master Thesis, Focus: Computer Graphics]

​

This project is dedicated to the procedural generation of leaf placement on trees, aiming to achieve realistic and diverse foliage through algorithmic methods. The goal is to develop a system that automatically determines the optimal positioning of leaves on branches, reflecting the natural growth patterns and density variations found in different tree species. This involves creating algorithms that simulate biological growth processes and environmental factors, ensuring that leaves are placed in a manner that enhances the overall realism and aesthetic appeal of the tree.

​

​​​​Evaluation of Algorithms for the Segmentation of Medical CT Volumes 

[Type: Bachelor or Master Thesis, Focus: Computer Vision, Medical]

​

CT scans offer a 3D representation of the human body. In order to reconstruct a 3D mesh of those organs, they need to be segmented in the greyscale axial images, either with conventional segmentation methods (such as thresholding or watershed) or with modern AI models. The outcome as well as required effort can be compared and evaluated.

​​

Uncertainty Visualization for AI-based Segmentations of Medical CT Volumes

[Type: Bachelor or Master Thesis, Focus: Computer Vision, Medical]

​

Various organs of the human patient can be segmented from CT scans and the corresponding greyscale axial views. AI models already achieve a high accuracy in the segmentation, but still need to be corrected. Instead of a binary classification mask as output, a heatmap may display the certainty of the segmentation and improve the understandability of the model.

​

Removal of Smoke and Spray from Medical Endoscopy Videos

[Type: Bachelor or Master Thesis, Focus: Computer Vision, Medical]

​

In robotic surgeries, an endoscope provides the view into the inflated belly of the human patient. While the organs are being operated on, smoke and spray often obscure the view for the surgeon. Conventional methods or modern AI models may be able to segment and inpaint the occluding particles and provide a cleared-up view across multiple video frames.​

​

Effect of the Patient's Rotation Angle on Tissue Deformation in CT Volumes

[Type: Bachelor or Master Thesis, Focus: Computer Vision, Medical]

​

While medical CT scans are generally taken in a supine position (lying on the back), during the actual surgery, patients are often positioned on their side. This leads to a deformation of internal organs so that their position does not match the CT scan anymore. The elasticity of various organ types may be estimated from tissue characteristics and used to predict absolute movement under the influence of gravity.

​

Semantic Segmentation of Outdoor Scenes

[Type: Master Thesis, Focus: Computer Vision]

​

This project aims to develop a deep learning model for semantic segmentation in outdoor environments, with a specific focus on identifying and classifying various types of vegetation such as trees, grass, shrubs, and other plant species. The student will work with datasets of outdoor images and leverage convolutional neural networks (CNNs) or other machine learning models to differentiate vegetation from non-vegetation areas. The outcome of the project will be a tool that can accurately map vegetation in outdoor scenes, potentially useful for applications like environmental monitoring, agriculture, and urban planning.

​

​Point of Interest Identification with a Robotic Agent

[Type: Master Thesis, Focus: Computer Vision, Robotics]

​

This project aims to develop a system that enables a robotic agent to autonomously explore an environment and identify points of interest (POIs) using advanced representation learning techniques. The core objective is to combine computer vision, representation learning, and robotic path planning to allow the agent to efficiently detect, represent, and categorize POIs, such as landmarks, objects, or regions of interest, based on sensory inputs like cameras or LiDAR. Representation learning will be used to extract meaningful, compact features from high-dimensional sensory data, enabling the robot to generalize across different environments and POIs with minimal supervision. The focus will be on emoloying the learned representations to enhance real-time identification and interaction with the surrounding environment. Applications include autonomous exploration, mapping, environmental monitoring, and search-and-rescue operations.​

​

Randomized Look-Up Data Structures for Spatial Data

[Type: Bachelor or Master Thesis, Focus: Computer Graphics]

​

The goal of this project is to investigate how randomization can be used to engineer efficient data structures for queries on spatial data. Initially, a literature research should be conducted to compile an overview of what the state of the art is. Based on this, data structures that have only been proposed theoretically should be implemented or entirely new data structures should be developed and implemented. Experiments should then be conducted comparing the performance to that of conventional data structures (e. g. Kd-trees, octrees). Optionally, these newly developed data structures could make use of learned biases to make random decisions which might yield further performance improvements.​