Abstract. In a groundbreaking study at TH Köln, a cutting-edge artificial intelligence (AI) system developed by Prof. Dr. Thomas Bartz-Beielstein and his team at TH Köln has demonstrated an impressive 98% accuracy in detecting falls among care-dependent individuals. Part of the „FallKI“ project, this innovative system utilizes a specialized measurement setup with vibration sensors placed near beds to monitor and identify fall events. The AI was trained using data collected from over 1,000 simulated falls conducted with a life-sized dummy. While laboratory results were promising, further real-world testing is needed to validate its practical application in nursing homes. The project, a collaboration with tecfor care GmbH, received financial support from the Federal Ministry for Economic Affairs and Climate Action. Future developments aim to enhance the system’s reliability and explore efficient methods to alert caregivers in real-time while safeguarding data privacy.
The artificial intelligence (AI) for fall detection in the „FallKI“ project was developed by Prof. Dr. Thomas Bartz-Beielstein, Prof. Dr. Olaf Mersmann, Noah Pütz, Jens Brandt, Alexander Hinterleitner, Richard Schulz, and Richard Scholz from the Institute for Data Science, Engineering and Analytics at TH Köln. This innovative AI is part of a measurement system capable of potentially detecting falls of care-dependent individuals near their beds with an impressive accuracy of 98 percent, as demonstrated by laboratory tests at TH Köln. Project leader Prof. Dr. Axel Wellendorf from the Institute for Mechanical Engineering at TH Köln highlights the significant risks that falls pose for care patients, as they often go unnoticed and can lead to health impairments.
Because continuous monitoring is not feasible, the system aims to inform caregivers as quickly as possible. In the research project, a measurement system with vibration sensors was developed, which can be placed next to the bed. These sensors capture mechanical vibrations and could potentially send this information to a central evaluation unit that decides whether an alarm should be triggered. Leonard Klemenz, a research associate, explains that the team staged over 1,000 falls with a dummy to adequately train the AI. This dummy is modeled after the human body to generate realistic data. To distinguish fall detection from other vibrations, the team conducted tests with various objects and events.
Despite the promising results from the lab experiments, the laboratory measurement technology used is not suitable for continuous use in nursing homes. Therefore, a robust and cost-effective sensor system was developed and tested over six months in a nursing home. However, the data collected was not sufficient to make a valid statement about its practical suitability. The researchers aim to further develop both the sensor system and measurement system and to apply them in future real-world tests. A key aspect remains the real-time information of the nursing staff, with data protection priorities needing to be considered.
The project was conducted in collaboration between TH Köln and tecfor care GmbH, and it was supported by the Federal Ministry for Economic Affairs and Climate Action as part of the Central Innovation Program for SMEs (ZIM) with a grant of 670,000 euros.
