Human Activity Recognition (HAR) for early detection: from training data collection to real world feasibility testing

23-11-18 Rongbo Hu 0 comment

Phase 1: Joint training data collection in a structured environment of a test flat by Fraunhofer, TUM, and SK. In a joint workshop, three Touchpoint2 partners, including FIAIS, TUM and SK, executed the collection of training data and achieved the following targets:

  1. Initial implementation, integration and functional test of the sensing systems
  2. Collection of initial data for testing data annotation software to start the development of machine learning algorithms
  3. Performing a pre-integration of REACH sub-systems to prepare for the data collection in SK´s clinical settings

news 3a

Phase 2: Activity detection through close-to-body and ambient sensors – generation of data sets for empirical validation with neurological patients at the Schön Klinik Bad Aibling

In a previous testing at the TUM laboratory, 19 wireless sensors were placed on the body (activPAL™, SmartCardia, Myo-Armband, iPhone®, BPMS™) and in the environment (cameras on the wall, and hand-hold camera) to create a heterogeneous sensor system for the recognition of activities, e.g., sleeping, sitting, standing, walking, and eating etc.

The clinical trial will evaluate whether the sensor system also works with neurological patients in the clinical setting. Measurements will focus on the recognition of activities of daily life (ADL runs) and repeated sequences (Drill runs).

news 3b.png



Leave a reply