AI-driven Knee Posture Detection in Cycle Training using IMUsAndreas Emrich; Janaki Viswanathan; Michael Frey; Peter Fettke; Peter Loos
In: Theresa Züger; Hadi Asghari (Hrsg.). AI Systems for the Public Interest. AI Systems for the Public Interest, located at 46th German Conference on Artificial Intelligence (KI 2023), September 26, Berlin, Germany, Humboldt Institute for Internet and Society (HIIG), 9/2023.
Knee injuries are a wide-spread medical issue with about 20% of the adult population affected. Knee surgery patients often suffer from serious instability of the knee that may affect their safety while moving. At the same time physical exercise is crucial in post-rehabilitation, in order to prevent further damage to the knee. To address this problem, we have designed an AI-driven, sensorbased solution. In collaborative research with partners from the medicine and sports domain, we identified three major postures that could lead to undesired results and that are likely bound to knee phenomena: (i) knee extension angle, (ii) (inner / internal) knee rotation angle and (iii) foot rotation angle. Inertial measurement units (IMUs) at the knee measure the knee positions, and based on history-based machine learning models, we detect the malpositions. Here, we present the system approach, the underlying algorithms and a laboratory evaluation that serves as a proof of concept that these knee malpositions can indeed be captured correctly. Future work hence can provide recommendations for the user in order to prevent potentially harmful situations for them while exercising.