Research topics

Our research work

3D reconstruction of the Martian surface with different scale and line drawings.
Martian surface reconstructed from Mastcam-Z images of the Perseverance rover from Sol 256 (Martian mission day) with VRVis software PRo3D. (© NASA/JPL/ASU/MSSS/JOANNEUM RESEARCH/VRVis/ICL)

Remote Sensing

REMOTE SENSING pertains to information about places or objects that are physically difficult to access. Scenarios include rock walls or building facades photographed by drone flight, and continents or distant planets reconstructed from satellite data. We create solutions adapting remote sensing methods to different requirements. Our research results range from interactive visualization of the surface of Mars for planetary research/space travel, to the generation of BIM models from scans of existing structures. The innovative bridging between data acquisition and semantic modelling is important to us.

Aus Punkten eines Laserscans rekonstruiertes Haus.
A house reconstructed from points of a laser scan.

Semantic reconstruction

REMOTE SENSING pertains to information about places or objects that are physically difficult to access. Scenarios include rock walls or building facades photographed by drone flight, and continents or distant planets reconstructed from satellite data. We create solutions adapting remote sensing methods to different requirements. Our research results range from interactive visualization of the surface of Mars for planetary research/space travel, to the generation of BIM models from scans of existing structures. The innovative bridging between data acquisition and semantic modelling is important to us.

View from above: a woman with glasses is touching a tactile relief with her hands.
The painting "Noche de frio espeso" by Aurelio Suárez as a tactile relief.

Inklusive digitization

In order to make the museums of the future barrier-free, we have been conducting intensive research for many years in the field of INCLUSIVE DIGITIZATION to find IT-based solutions that can make the fine arts "visible" to all target groups. To this end, we have developed, among other things, special visualization software that creates a digital 2.5D model from 3D scans or photographs of museum objects. Using an innovative milling process, this model can be transformed into a tactile relief that makes heights, depths, surface structures, etc. haptically perceptible - thus equipping museums for the 21st century.

A researcher points towards two computer screens on which you can see the semnatic annotation of a human spine.Eine Forscherin zeigt auf zwei Bildschirme, wo annotierte Wirbelsäulen zu sehen sind.
Our fully automated semantic annotation of the spine facilitates the work of digital radiology.

Reconstruction in medicine

For the efficient further use and ANALYSIS OF DATA OBTAINED BY IMAGING METHODS in the medical sector, we are researching innovative ways of automatically segmenting specific image content such as vertebrae, tumors or organs using machine learning and specially developed algorithms. This is an important supporting measure in the diagnostic process and also serves as a basis for reconstructions of volume data, for example.

A researcher at her desk in front of her two screens.
The combination of (geo)data visualisation with artificial intelligence, for example for the automatic recognition of platforms or pavements in point clouds.

Photogrammetry

We maintain a high precision, large-scale PHOTOGRAMMETRY pipeline to enrich image-based tools and workflows. Purpose-built for terrain reconstruction from drone flights, our state-of-the-art algorithms find application in a range of use cases. We augment laser scans with precisely registered detail photos, use photo stitching to generate high-resolution orthophotos, and provide comparison views of a reconstructed area over time. Our modular and reusable implementation meshes well with upcoming innovations in low-cost scanning and AI-assisted reconstruction.

Visualization of a reinforcement cage, individual bars are colored.
Geometric reconstruction of single rebars in a rebar cage. The point cloud is reconstructed from images with NeRFs.

NeRFs

Neural Radiance Fields (NeRFs) are a recent AI method for creating novel views from a set of images of an object. Since they could be used as basis for semantic reconstruction, we conducted a series of experiments. We evaluated the reconstruction results of different capturing procedures, image sizes, lighting conditions and the influence of tracking errors.  We will further explore various state-of-the-art methods and develop extensions that address specific problems of our use cases.

A woman and a man are sitting next to each other in front of two computer screens, the woman is pointing at one of the screens
The GeoSMAQ research team combines image-based AI methods and guided optical character recognition to extract valuable information from engineering drawings.

Understanding of engineering drawings

Engineering drawings (ED) hold valuable information for a number of use cases in manufacturing. Our goal is the development of a fully automated pipeline for ED understanding and entity extraction since our partners need to process large quantities of digitized EDs to analyze the additive manufacturing (AM) potential of the represented parts. We employ image-based AI methods to extract entities including object boundaries, text fields, and dimensions, and perform guided optical character recognition (OCR) to find important textual information like materials, scales, or numeric values. The logical combination of these steps allows us to reconstruct the necessary information from EDs to analyze the AM potential of the respective parts in an interpretable way.

Projects

Publications