Design and Development of a Daily Activity Aware System for Assisted Charting and Caregiving

Abstract

Providing an appropriate care plan for the elderly to participate in everyday life activities is important as is developing the technological means to support physical assessments In clinical practice observation of physical activities and movement patterns is crucial though current protocol is generally episodic from subjective assessments and interviews With the maturing development of information and communication technologies application of said technologies for quantizing activities could provide continuous and objective data to adjust caregiving and support assisted living The purpose of this research is to design and develop an activity awareness system via ambient sensing technology to automatically identify and record daily in-house living activities for assisting healthcare charting and monitoring needs for the health related quality of life (HRQL) assessment This research was aims to: 1) develop an ambient awareness system with pressure and acoustic sensing mechanisms to collect measure and store lying posture and environmental sounds for monitoring bedside activities Lying pressure distribution data was gathered from a sensor pad developed for this purpose Acoustic streams were recorded for designed scenarios within a mock living space; 2) develop a probabilistic based activity recognition system to efficiently model life activities and in-bed posture for providing quantitative measurements towards healthcare charting and lifestyle monitoring A Gaussian mixture model (GMM) based classifier trained with a minimum classification error (MCE) criterion was adopted for robust posture classification; 3) develop an identification system for bedsore risk monitoring The fuzzy c-means (FCM) algorithm was used to transform the pressure contours and identify regions of interest (ROI) having high pressure for ulcer prevention Latent semantic analysis (LSA) extracted the significant features from the transformed ROI images in order to develop an artificial neural network model for posture recognition; 4) develop an acoustic activated recognition system to efficiently model daily bedside activities for providing quantitative evidence of physical health and social interactivity A hidden Markov model with a behavior grammar network developed for this research automatically recognized acoustic events A Gaussian mixture model combined with multidimensional scaling was proposed for fast speaker diarization; 5) evaluate the performance of the developed system towards resident safety and caregiving efficacy by using scenario-driven daily activity datasets Several objective evaluations and field trials were performed to investigate the performance of lying posture and activity detection Experimental results show the average posture recognition rate was 95 89% when the pressure distributions were divided into 4 clusters FCM with LSA transformation improved the recognition rate and could be used to locate the corresponding risky compressed bony prominences The high detection rates in both the recognition of acoustic events and speakers demonstrated the feasibility of efficiently modeling daily activities and providing quantitative evidence of health conditions and social interaction The case study and simulation tests show the potential and applicability for in-bed and bedside activity monitoring and also shows suitability for objectively evaluating the physical functions in the activity of daily life (ADL) instrument activity of daily life (IADL) and HRQL assessments Some ways to build upon the contributions of the current study include: improving the developed prototypes such as new pressure pads to provide more efficient in-bed monitoring or enhance the quality/resolution of the captured sound to increase the rate of acoustic event recognition performance Extended applications include real-time alerts for excess pressure duration on bony prominences and specific HRQL assessment indexes

Date of Award	2014 Aug 30
Original language	English
Supervisor	Kuo-Sheng Cheng (Supervisor)