Skip to main content
eScholarship
Open Access Publications from the University of California

Robust Human Activity Classification and Motion Monitoring Systems Using Inertial Sensors

  • Author(s): Wu, Xiaoxu
  • Advisor(s): Pottie, Gregory J,
  • et al.
Abstract

The proliferation of powerful microcomputers and the development of modern machine learning tools have enabled human daily activity monitoring systems using wearable inertial sensor like accelerometers and gyroscopes. These systems fulfilled the urgent need in health and wellness industries in helping doctors and clinicians during diagnosis, treatments and rehabilitation processes for neurological diseases like strokes and Parkinson’s.

For most current activity monitoring systems, there exists an assumption that the sensors are always securely and correctly mounted by the users. Unfortunately, such assumptions do not hold as the scale of studies increase. And it is especially challenging for subjects with neurological diseases to follow instructions about how to mount the sensors everyday, because some of the elderlies tend to be technophobic and neurological diseases are often accompanied with cognitive difficulties. Errors in sensor mounting pose can cause large amount of data loss and distortion and will affect the robustness of the systems severely.

In observance of these issues, a series of solutions for sensor orientation and position errors in human motion monitoring and activity classification will be presented. Opportunistic calibration methods to find the true sensor orientation and position will be discussed. In addition, systems that provide robust monitoring regardless of the exact sensor pose will be proposed.

Main Content
Current View