Low Bandwidth Inertial Sensors For Upper Extremity Monitoring During Rehabilitation

Open Access
Author:
Hornung, Taylor Christian
Graduate Program:
Mechanical Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 30, 2015
Committee Members:
  • Dr Jason Moore, Thesis Advisor
  • Stephen Jacob Piazza, Thesis Advisor
Keywords:
  • Wearable Sensors
  • Inertial Sensors
  • Motion Capture
  • Stroke
  • Rehabilitation
  • Therapy
Abstract:
This thesis compares the quantifiable data from a low bandwidth inertial sensor to that of a high bandwidth, high accuracy video based motion capture system for assessing upper extremity motion. The goal is to develop a system which can collect and characterize motion data that is comparable to current state of the art research equipment using only commercially viable products. First a low cost, low bandwidth inertial measurement unit (IMU) sensor that can be strapped to a person’s wrist is interfaced via I2C connection to an Arduino board. Custom firmware on the Arduino is used to collect relevant motion data pertaining to the person’s upper extremity. This data is sent to a LabVIEW interface on a computer where it can be monitored in real time by a physician. For comparison, data from the exercises is also collected from the Eagle motion capture system from Movement Analysis Co. A side by side comparison is given between the IMU sensor and the motion capture system. The same IMU sensor system, firmware and software, is then implemented on a standard weighted bar, a common rehabilitation device. Using a motor driven weight and closed loop feedback control, the system can adjust the weight distribution of the bar. The sensitivity and balance of the bar can all be fine-tuned for specific patient needs. The goal is to allow clinicians to provide personalized training and exercise adjustments, while simultaneously monitoring a patient’s performance in real time for a more targeted rehabilitation. When analyzing wearable IMU sensors, the correspondence of individual features extracted varies depending on several aspects of the sensors and motion being performed. In particular, mean acceleration does not correspond well to the camera system. Linear negative mean jerk metric had high correspondence, particularly for cases with less smooth motion. RMS acceleration magnitude showed high correspondence when not corrupted by measurement error. Dominant frequencies only show high correspondence when noticeable shaking is present, and may not be useful otherwise. The frequencies also tend to appear on specific axes, which is dependent on the motion being performed.