Group 3D Geodata Analytics

The group’s research focus is on the analysis, classification and modeling/vectorization of spatial measurement data. We use machine learning methods such as deep learning for the fully automated analysis of 2D and 3D measurement data. This involves training artificial neural networks (ANNs) to recognize and pinpoint objects – for instance from urban infrastructure – in comprehensive mobile measurement system data sets. In addition to manual annotation, we use our own tools and synthetic training data to efficiently train KNN.

To ensure the highest level of precision and reliability in data analysis, we combine artificial intelligence (AI) with traditional heuristics. We optimize runtime to achieve real-time capability, and we integrate analysis directly into measurement systems. This allows measurement results to be tracked in real time and measurement parameters to be adjusted as needed. Also, time-consuming post-processing and storage of raw data can be avoided.

Automated data analysis

• Fully automated analysis of 2D and 3D measurement data, e.g. by means of deep learning
• Implementation of local or cloud-based solutions for data processing
• Compilation of comprehensive training datasets for the automated training of algorithms

Synthetic training data

• Creation of 3D scenes including material properties, lighting conditions, weather phenomena and dynamic properties
• Algorithmic generation of 3D models from parameterizable components
• Creation of simulated measurement data: photo-realistic images and 3D point clouds

Real-time analysis

• Optimizing analytical AI for real-time evaluation based on the computing power available on mobile platforms (e.g., Nvidia Jetson)
• Smart data handling: real-time AI for data reduction or regulatory tasks (e.g. anonymization), complex analytical tasks in postprocessing or in the cloud