Leonie Pflaum

  • Job title: Bachelor's Thesis
  • Working group: Development and Evaluation of Embedded Classification and Intelligent Sensor Management for Power Saving in Wearable Sensor Systems

Nils Roth, M. Sc., Dr. Heiko Gaßner, Martin Ullrich, M. Sc., Prof. Dr. Björn Eskofier

05/2018 – 10/2018

To assess motor and physical functioning of patients, wearable sensor systems are becoming increasingly interesting for medical care as well as diagnostics [1]. Especially for patients with chronic diseases like Parkinson’s disease (PD), such sensors can be used to objectively assess motor fluctuations and other related symptoms. To provide measures under free living condition and over extended periods of time they need to be integrated in an unobtrusive and non-stigmatizing way into a patient’s everyday life [2]. This demands extremely small and light-weight sensor solutions which in turn only provide limited battery power and computational resources. To enable continuous data acquisition without frequent battery recharging or battery replacement, a low power consumption and long runtimes are crucial for long-term and home-monitoring applications.
State-of-the-art systems meet these requirements by restricting sensor nodes to only ultra-low-power accelerometers and at the same time minimizing the used sampling rate to some tens of Hertz [3] or utilize fully embedded processing instead of raw data recording [4]. However, for gait analysis in PD higher sampling frequencies in a range of 100Hz as well as additional sensors with an increased energy consumption like gyroscopes are needed. This is necessary to assess high resolution gait parameters like stride length or foot angles [5]. Nevertheless, these factors increase energy consumption dramatically and make runtimes of more than one day difficult to achieve [6].

The aim of this bachelor thesis is to determine the optimum balance between conserving all important information while consuming minimal energy and thereby maximizing the runtime of wearable sensor systems. This shall be achieved by utilizing embedded processing and on-board, real-time classification of sensor data instead of a naïve continuous recording or streaming approach. The acquired classification results can be further used to dynamically adapt the sampling rate or to select the required sensor channels according to the respective situation as mentioned by T. Rault et al. [7]. For example, during complex activities such as gait, where feature calculation requires advanced and computational demanding signal post-processing, raw and high-resolution sensor data is desired. In turn, for static activities like e.g. sitting or standing, pre-processed activity labels or only low-resolution data is sufficient. As the quantity of complex activities is expected to be considerably lower compared to static activities during whole day measurements this approach will not only help to minimize power consumption and boost the runtime of wearable sensor systems but also lower the overall amount of recorded raw data. Due to the limited memory in low-power wearable sensor systems, this states another important objective for continuous home-monitoring applications.

This bachelor thesis will include the development of different classification algorithms which will be optimized for embedded systems and followed by an implementation on an already existing, fully embedded and wearable sensor insole. Finally, the proposed system will be evaluated regarding power consumption and accuracy in a small study focusing on long-term gait monitoring applications.


  1. Allet, Lara, et al. “Wearable systems for monitoring mobility-related activities in chronic disease: a systematic review.” Sensors 10.10 (2010): 9026-9052.
  2. Ossig, Christiana, et al. “Wearable sensor-based objective assessment of motor symptoms in Parkinson’s disease.” Journal of neural transmission 123.1 (2016): 57-64.
  3. Aadland, Eivind, and Einar Ylvisåker. “Reliability of the Actigraph GT3X+ accelerometer in adults under free-living conditions.” PLoS One 10.8 (2015): e0134606.
  4. Wang, Ning, et al. “Energy and accuracy trade-offs in accelerometry-based activity recognition.” Computer Communications and Networks (ICCCN), 2013 22nd International Conference on. IEEE, 2013.
  5. Rampp, Alexander, et al. “Inertial sensor-based stride parameter calculation from gait sequences in geriatric patients.” IEEE transactions on biomedical engineering 62.4 (2015): 1089-1097.
  6. Hegde, Nagaraj, Matthew Bries, and Edward Sazonov. “A comparative review of footwear-based wearable systems.” Electronics 5.3 (2016): 48.
  7. Rault, Tifenn, et al. “A survey of energy-efficient context recognition systems using wearable sensors for healthcare applications.” Pervasive and Mobile Computing 37 (2017): 23-44.