HPI Digital Health Center

Nursing Activity Recognition Using Data Augmentation

Nurses in Germany are called to document the care they provide. A considerable amount of the working time is spent on documentation. Currently available and proposed solutions do not show a significant reduction in time, which would not itself involve a certain amount of effort. We are working on an approach based on motion sensors and machine learning to reduce the time spent on documentation. Nursing staff would no longer have to worry about documentation and could devote themselves fully to the patient.

In order to achieve a good classification performance of the different activities we are working on techniques, which are well suited to transform the raw sensor signal into images. The transformation allows us the usage of image classification methods.

Since data collection plays an important role in machine learning tasks, we are also doing research on data augmentation techniques with the help of simulation software like Unity or Blender.