Electronic Thesis and Dissertation Repository

Thesis Format

Monograph

Degree

Master of Engineering Science

Program

Electrical and Computer Engineering

Supervisor

Katarina Grolinger

Abstract

Human Activity Recognition (HAR) has been attracting significant research attention because of a wide range of applications from healthcare to security. Recently, deep learning approaches have demonstrated great success in the HAR area. However, these models are often evaluated on the same subjects as those used to train the model; thus, the provided accuracy estimates do not pertain to new subjects. Consequently, this thesis examines the generalization capability of different machine learning architectures using Leave-One-Subject-Out Cross-Validation (LOSOCV) and then proposes a personalized model. The accuracy is improved by considering two feature selection directions, time- and frequency-domain, and by dynamically selecting the best model for that subject. In the time-domain, the best results are achieved by CNN with two convolutions and a one-dimensional filter. In the frequency domain, the results are improved through Empirical Mode Decomposition (EMD) with cubic spline and the personalization technique.

Summary for Lay Audience

With the rapid development of sensors, our daily lives have been changing. Human activity recognition (HAR) aims to identify human activities employing signals received from the environment or from wearable sensors. HAR can improve the quality of life through applications such as health monitoring, assisted and active living. Machine learning and especially deep learning have shown potential in HAR; however, there is space for improvement in terms of accuracy, in particular for new users. Consequently, this thesis examines different preprocessing methods, deep learning architectures, and personalization models to increase HAR accuracy for new users. In the preprocessing, two directions, time-based and frequency-based features, are considered and their effect is investigated. In the deep learning part, different deep neural network architectures are examined to achieve better structures. Finally, in the personalization part, a new personalization approach is proposed to customize generic models for new users. The evaluation shows that deep learning models with signal preprocessing techniques and the personalization approach can successfully detect 91:2% of activities for a new person.

Download

Share

COinS