Sensor-based Human Motion Analysis

基於感測器的人體運動分析

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date25 Jul 2022

Abstract

The Healthcare system in East Asia is facing significant challenges. The primary difficulties come from the aging population, increasing chronic disease burden, and the expectation of using advanced technologies. 17.9% of Hong Kong's population was above 65 years old, and by average, around 4 adults needed to support an older adult in 2018. It is projected that these figures will drastically worsen in the following decades. The aging society means that the healthcare workforce will face a severe shortage. More attention is addressed to caring for older adults in the upcoming decades.

Gerontechnology is developed to assist older adults daily activity with sensors, mobile devices, and information technologies. Recent advancement in technology offers the chance to alleviate the burden of the healthcare workforce and provide better care quality for older adults. Sensor technology is one of the keys to enabling gerontechnology, as it collects the data from the surroundings to provide humans and computers to make in-depth decisions.

One of the potential applications in gerontechnology is sensor-based human activity analysis, but there are still some challenges that are not yet solved. With the assistance of a camera or inertial sensors, quantitative analysis of human motions is attainable. Such analysis can benefit older adults to assess their fall risk and daily exercise movement. Besides, the workload for medical professionals is reduced by utilizing the automation of sensors. Nonetheless, a comprehensive data analysis framework and interpretability of sensor-data analysis are not yet well-established.

This thesis aims to analyze human motions through sensor devices. This goal can be divided into two sub-problems: (1) How to conduct health surveillance through sensor, and (2) how to analyze the time-series sensor data to reach desirable interpretability and accuracy.

This thesis first presents a novel method to count the periodical repetition of human motion through an ordinary mobile camera or depth camera. The proposed method leverages the human skeleton detection method to facilitate viewpoint invariant, and multiple people exercise counting. The first session demonstrates a novel approach conduct multivariate time series to estimate the repetition of the exercise. This technique possesses strong potential in group training and work design to quantify the performance of human movements.

Following the health surveillance, the second part introduces a method to predict the fall risk for post-stroke patients using inertial sensor data from complex exercise movements. Stroke is common in older adults, and it will weaken the balance capability for the patients. We used the data-drive approach to extract and select the interpretable features and model. The proposed method can benefit both healthcare professionals and patients. It facilitates clinical interpretation and opens a new window in quantitative motion analysis for medical professionals and in-house self-assessment for patients to assess their fall risk.

The last part of this thesis explicitly delineates a comprehensive data pipeline for analyzing sensor movement data from health surveillance and data analysis. Human activity recognition techniques are adopted to identify the motion of community older adults. After the model identifies the motion, the sensor-based motion analysis is conducted. The result shows that the sensor-based motion analysis pipeline agrees with expert opinions.

The thesis contributes to depicting a comprehensive framework of sensor-based human motion analysis with detailed implementation of each stage. This work explicitly bridges the sensor-based health surveillance and the time-series classification problem to offer a systematical solution to human motion analysis. The problems solved in this thesis offer a prototype to evaluate the fall risk and exercise quality through sensor-based motion analysis techniques. It can significantly benefit older adults and athletes to improve their movements.