Classification of Altered Gai

This dissertation discusses the creation of a wearable gait analysis system based on six degree of freedom inertial measurement units and illustrates its use as a research platform by collecting data mimicking the altered gait that results from multiple conditions and classifying those conditions. By collecting data from five activities (walking on a flat surface, ascending stairs, descending stairs, ascending a slope, and descending a slope) for each of the four gait patterns, the interaction between gait pattern and activity classification can be observed in the results of phase one. This potential interaction has not been previously explored, but is important to quantify so that gait pattern classification can be performed in a larger number of environments because there is no significant interaction between activity and gait pattern. This could allow data to be collected from individuals while they are in the community instead of under direct supervision of a medical professional, because it is not essential that they perform a specific activity to provide data for gait classification (for example, a sloped sidewalk instead of a flat hallway would not change the gait pattern classification results). The combinations of sensors, that would provide results that were not significantly different from the complete five sensor system, were also explored.;In phase two gait pattern classification was performed to determine differences in classification accuracy for different classifiers and different sensor placement combinations. The classification accuracy for this section was highest for the ensemble method, in contrast with activity classification where support vector machine classification was most accurate. This highlights the differences between the two types of classification and the need for research into gait pattern classification in addition to activity classification. Similarly, there were differences in the sensor combination results, with only two sensor combinations (waist, right thigh, and left ankle; and waist, right thigh, and right ankle) without statistically significant differences with the five sensor combination. The waist and both ankles sensor combination was not statistically different for phase one, but was different for phase two.;This research validated the system used to collect IMU data, as well as sensor placement for gait pattern classification. This research's use of altered gait patterns in healthy control subjects allowed data to be collected during activities that would have caused difficulties for patients, so that the interaction effect between activity and gait pattern could be determined.