Machine learning and autonomous system for human gait analysis based on walk speed

Elkurdi, A

Abstract

The quality of life and cost of care for elderly people varies dramatically between those living independently and those receiving acute or long-term care, which takes place at home, in residential care or in hospital. The common aim of national health service providers is to keep elderly people safe at their own homes for as long as possible to promote independent living, increase their quality of life and reduce hospital costs. Hence, the application of autonomous sensing systems to enhance everyday life of such population will be valuable and has been considered here.
Recently, Microsoft Kinect v2 has been used for gait analysis systems, to perform data classification of gait pattern changes based on walking speeds. This system enables the tracking without the need of any markers. Moreover, the Kinect camera is considered a low-cost device, and is quick to install, even in an unprepared environment. However, the primary challenge of such a device is that it provides a low data rate which leads to a decrease in the quality of extracted features, compared to other Motion Capture Systems (MoCap). Furthermore, in the data classification stage, the performance of classification is greatly affected by the boundary between different classes which is called decision boundary. This raises other questions such as: how to weight the features from the class labels, and which kind of similarity metric can be used.
To improve the quality of features, the Amplitude Modulation (AM) and Convolutional Encoder (CE) can play a major role in detection and in ranking the gait pattern changes based on walking speed. For this purpose, the collected data is mapped into a higher frequency spectrum using the AM domain. Consequently, the “AM-modified gait signal” is produced to improve the quality of extracted gait features, by increasing the level of the frequency sampling rate.
In this research, the main novelty is the combination of Amplitude Modulation (AM) and Convolutional Encoder (CE) techniques in one system (AM/CE) in order to understand and identify the walking speed effects on gait parameters. The former is proposed to extract new gait features without the need to determine the gait cycle phases, while the latter is developed to classify gait data based on walking speeds. Therefore, the performance of the CE technique is improved efficiently in gait data classification by weighting the bit positions in
xv

Hamming Distance (HD) length, which leads to an increase in the accuracy of measurement of the similarity metric.

Citation

Elkurdi, A. Machine learning and autonomous system for human gait analysis based on walk speed. (Thesis). University of Salford

• Thesis Type: Thesis
• Deposit Date: Apr 8, 2020
• Publicly Available Date: Apr 8, 2020
• Award Date: Apr 1, 2019