Unobtrusive, in-home gait measurement, fall risk assessment, and fall detection using depth imagery
Metadata[+] Show full item record
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Research has shown the importance of measuring a person's gait. However, current methods for measuring gait, such as observation by a clinician with a stop watch or evaluation in a physical performance lab, often lead to sparse, infrequent assessments that may not be representative of a person's true ability. This dissertation describes the development, evaluation, and testing of a system for measuring and tracking habitual, in-home gait parameters; generating automated fall risk assessments interpretable by clinicians from the in-home gait parameters; and, finally, detecting falls in the home, all using an inexpensive depth imaging sensor. The main advantages of this approach include the low cost, unobtrusive, environmentally mounted sensing platform, and the continuous measurement of mobility. An initial study, conducted in a motion capture laboratory, focused on evaluating the suitability and accuracy of the Microsoft Kinect depth imaging sensor for gait measurement in real-world, unstructured environments. This study, which had thirteen participants of varying ages who completed over 100 walking sequences, found good agreement between measurements computed using the Kinect and those made using a Vicon marker-based motion capture system. Issues with certain types of clothing, the range limitations of the device, and advantages over traditional camera systems were also studied. Additional algorithm development, including an adaptive background model and simple, yet robust, tracking of identified three-dimensional objects was completed to facilitate the use of the system in real-world environments. A Kinect sensor and computer were then deployed in apartments of older adults. At present, more than 35 systems have been deployed for time periods of up to two years, with current plans for 70 additional systems. Using the data collected in these apartments, along with traditional fall risk assessment data collected monthly by a clinician, a probabilistic methodology for generating automated gait parameter estimates
Access is limited to the campuses of the University of Missouri.