Fraunhofer Institute for Computer Graphics Research IGD
Fraunhofer Institute for Computer Graphics Research IGD

ML for image and point cloud analysis in the geo and environmental sector

Due to the significant increase in computing capacity and the volume of data available in recent years, AI-based evaluation methods and machine learning are becoming increasingly important. Among other things, they can be used to recognize and classify objects in images or point clouds, detect sensitive information and personal data in photographs and make it unrecognizable, and process very large amounts of data automatically and intelligently without the need for human intervention.

We have extensive expertise in the field of machine learning, which we can use to provide our customers with application-related advice and support for their projects. We have also developed various software components that, for example, improve the automation of neural network training or apply machine learning methods to various use cases.

Aerial and satellite images: classification of cities and forests

The analysis of aerial images and orthophotos using machine learning methods offers a wide range of potential applications, particularly in the context of urban environments and urban planning. For example, different types of sealed surfaces such as roads, parking lots and buildings can be precisely identified and used for urban planning activities or infrastructure projects. In addition, by capturing and analyzing different environmental pressures such as overheating, air pollution and the loss of green spaces, cities can better understand and take appropriate action to reduce these pressures.

Another application is the classification of forested areas. A particular problem here is the accurate identification of tree species. Based on our workflow management system Steep, we are able to train and evaluate various machine learning approaches on Sentinel 2 data.

 

go to the Steep Website

 

Publications:

Kocon, K., Krämer, M., and Würz, H. M.: Comparison of CNN-based segmentation models for forest type classification, AGILE GIScience Ser., 3, 42, https://doi.org/10.5194/agile-giss-3-42-2022, 2022.

Würz, H. M., Kocon, K., Pedretscher, B., Klien, E., and Eggeling, E.: A Scalable AI Training Platform for Remote Sensing Data, AGILE GIScience Ser., 4, 53, https://doi.org/10.5194/agile-giss-4-53-2023, 2023.

Showcases

 

ML-based classification of forest types in Copernicus data

Zum Showcase

Panoramic images: anonymizing vehicles and persons

Geodata is generated for a variety of use cases and made available to a wide range of users. It is important here to ensure compliance with the legal framework. For example, images with a spatial reference (such as 360° panoramic images, aerial images, textures or other photographs) often contain sensitive and personal information (such as faces or license plates).

To use these images within a secure legal framework, we have developed a software component that automatically recognizes and blurs (“pixelates”) people and vehicles. The anonymizer uses machine learning methods and can process a very large number of images in a short time through GPU acceleration. The software is extremely reliable and has a high detection rate. The anonymizer has already been in productive use in the industry for several years.

 

Publications:

Krämer, M., Bormann, P., Würz, H. M., Kocon, K., Frechen, T., & Schmid, J. (2024). A cloud-based data processing and visualization pipeline for the fibre roll-out in Germany. Journal of Systems and Software, 211, 112008. https://doi.org/10.1016/j.jss.2024.112008