UAV's are mostly equipped with low-cost navigation sensors (GPS or IMU), allowing a moderate positioning accuracy. In this project we analyze the potential of differential post-processing of GPS data from an UAV in order to improve the positioning accuracy for applications based on direct georeferencing. [PDF]
This project deals with the improvement of various steps in image processing in order to achieve an automated approach for glacier and glacial lake monitoring. In particular, we focus on: automated ground control and tie point measurement, glacial lake detection for volumetric calculations, glacier surface modelling and flow estimation. [ICVSS 2013: Poster]
The Drapham Dzong is a ruin from the 15th/16th century situated in central Bhutan. As a part of the first archaelogical project in the kingdom, the aim of this work is to document the citadel with UAV and terrestrial images, and to generate a DSM and an orthophoto of the area. The results were presented at the Bhutan exhibition at the Rietberg Museum in Zurich (2010). [PDF]
© 2016 Maroš Bláha
In this work, we aim to jointly reason about 3D geometry and semantics, allowing one to take into account class-specific shape priors (e.g. building walls should be vertical, and vice versa vertical surfaces are likely to be building walls). Exploiting the synergies between the 3D shape and semantics of a scene leads to superior and interpreted 3D models. As test bed, we focus on urban areas.
Classified as UNESCO heritage site, the lake dwellings on the ground of Hallwilersee represent important artifacts from the Stone Age. For the reason of precaution, the goal of this work is to generate a virtual 3D model of this underwater archaeological site. [Video, "Aufwändige Forschung"]
Urban scene modelling from images has reached remarkable achievements. While many techniques focus on visualization, the utility of such models extends well beyond. Thus, these models should enable advanced scene study. In this project, we analyze different types of representations for their utility to perform a diverse set of reasoning tasks such as navigation.
Contemporary research advances in image-based mapping have focused on automatic reconstruction of both, the 3D geometry and semantic meaning (e.g. ground, building, etc.) of a documented scene. The resulting models extend the line-up of utility and enable a more advanced reasoning, e.g. in the domain of digital urban simulation.The aim of this work is to analyze how suitable semantic 3D models are in terms of digital urban simulation tools such as shadow or solar radiation analysis.
During my past years at ETH, I was involved in several UAV projects such as DSM/orthophoto generation of gravel pits for various analyses (e.g. volume calculations) or modeling of residential parcels for a cadastral map.