South Dakota State University
Human Activity Recognition Based On Wearable Flex Sensor and Pulse Sensor
Abstract
dc:description.abstractIn order to fulfill the needs of everyday monitoring for healthcare and emergency advice, many HAR systems have been designed [1]. Based on the healthcare purpose, these systems can be implanted into an astronaut’s spacesuit to provide necessary life movement monitoring and healthcare suggestions. Most of these systems use acceleration data-based data record as human activity representation [2,3]. But this data attribute approach has a limitation that makes it impossible to be used as an activity monitoring system for astronavigation. Because an accelerometer senses acceleration by distinguishing acceleration data based on the earth’s gravity offset [4], the accelerometer cannot read any type of acceleration when it is in the actual free fall environment. Since astronauts will experience microgravity and/or zero environments in outer space, all existing acceleration data-based HAR systems cannot fulfill this requirement. Therefore, it is necessary to design a new data attribute for HAR systems to specifically work under microgravity and zero gravity environments. The angular change of body joints during activity can be a good solution. By attaching sensors onto body joints, the system can recognize an activity by analyzing the change pattern of bend angles similarly to how people recognize others’ activity by looking at their posture during movement. Considering the possibility of overlapping data from multiple different activities that may have similar angular changes, a life activity related data called Beats Per Minute (BPM) is thus used to differentiate overlapping activities. With the new compilation and format of activity data, the HAR system should be able to work under both microgravity and non-gravity environments with similar or better accuracy than existing HAR system implementations. This paper demonstrates the implementation of new data attributes based on existing HAR systems by using angular data and BPM data, then makes comparison between acceleration data-based HAR and angular data-based HAR systems to verify the performance similarities, and comparison among different neural network structures to analyze and provide the most suitable machine learning technique to train the system.
Degree
thesis:*- Name thesis:degree_name
- Master of Science (MS)
- Level thesis:degree_level
- Thesis - Open Access
- Discipline thesis:degree_discipline
- Electrical Engineering and Computer Science
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jin, Xiaozhu
- Contributors dc:contributor
-
- Kwanghee Won
Subjects
dc:subject × 4Rights
dc:rightsIdentifiers
dc:identifier.*- Repository record dc:identifier
- https://openprairie.sdstate.edu/etd/5273