The research department „Artificial Intelligence in Biomedical Signal Processing“ (BioSP) develops innovative techniques for different areas of signal processing. These include the analysis of biosignals acquired from humans, the analysis and processing of acoustic signals, for example for hearing aids, and the evaluation of physical measurement signals for medical imaging. Beyond the medical context, the work extends to many other fields of application, such as wearables and human-machine interfaces, smart home and assisted living, as well as production monitoring and predictive maintenance.
The requirements for the systems are as diverse as the applications. Artificial Intelligence (AI) methods are used to learn complex relationships in a data-driven manner in order to overcome the limitations of classical mathematical models. Robustness to perturbations and incomplete data as well as interpretability of the algorithms are essential. Three examples will demonstrate the power and versatility of AI solutions.
Hand gesture recognition
The example of hand gesture recognition illustrates the high adaptability of AI methods. One can train one‘s own gesture recognition and then play games. It is shown how well AI systems can be adapted to new requirements by a short training phase and experience how powerful they are even in the context of embedded systems.
Acoustic event detection
Audio event detection plays an important role in many areas, like smart home, surveillance (also for production processes), and hearing aids. One can see how well a variety of different events can be classified in real-time thanks to sophisticated AI.
AI must function robustly despite many unknown circumstances. Sleep staging involves the analysis of EEG data recorded during sleep. One of the challenges is to provide reliable results for all patients despite different conditions and measurement devices. Countless EEG measurements are used to show how reliable and robust AI can be.