Multimodal wearable fall risk assessment for independent people living with dementia

Older adults with dementia face a twice higher risk of falls compared to cognitively healthy peers, largely due to impaired cognition, reduced self-monitoring, and poorer balance and movement control. Traditional fall prevention strategies effective in cognitively healthy older adults have generally not reduced fall risk in People with Dementia (PwD) and individual with Mild Cognitive Impairment (MCI). While interventions such as home modifications may improve safety and independence, adherence can be challenging. Technology-based approaches, particularly Augmented Reality (AR) assistive tools, show promising potential; however, most existing AR solutions fail to adequately adapt to users’ fluctuating or declining cognitive states. This thesis therefore, aims to develop an adaptive multimodal AR–based Assistive Technologies (AT) that leverages Deep learning (DL) and Hidden Markov Model (HMM) to assess fall risk for PwD and individual with MCI. The thesis began with an extensive systematic review of 1,449 studies on AR–based AT for the safety of PwD and MCI, of which 31 met the final inclusion criteria. The review demonstrated the potential of AR technologies in enhancing safety, independence, and Quality of Life (QoL) for the target population. It also identified persistent challenges such as the need for more adaptive and intuitive interfaces, improved ergonomic design for prolonged use, and robust privacy protections, particularly in location and movement monitoring. This was followed by a cross-sectional survey involving 121 caregivers to examine their perspectives on PwD’s attitudes toward home hazards and safety interventions, with the aim of identifying unmet needs. The findings indicate that PwD exhibited limited hazard avoidance and low adherence to existing home safety interventions, largely due to cognitive and behavioural impairments associated with the condition. Based on insights obtained from the systematic review and cross-sectional survey, an adaptive multimodal AR–based AT for fall risk assessment system aimed at supporting safe mobility through personalised, context-aware notifications that align with the user’s cognitive attention state was proposed. The proposed system employs wearable smart glasses equipped with an integrated camera to capture environmental images and an Inertial Measurement Unit (IMU) to collect motion data. DL techniques were employed to analyse visual data, enabling detection of potential environmental hazards and the identification of existing safety interventions within the environment, while HMM was applied to infer the user’s cognitive attention state from IMU sensor data. This approach enables the delivery of personalised safety alerts and guidance only when cognitive attention is reduced, thereby minimising unnecessary notifications that can lead to alert fatigue while preserving user autonomy. The system was evaluated based on Unified Theory of Acceptance and Use of Technology (UTAUT) framework. To better capture the needs of individuals with cognitive impairment, the UTAUT framework was extended to include Social Sensitivity (SS), which reflects concerns related to stigma and social judgement, and Technology Anxiety (TA), both of which significantly influence technology acceptance. Findings from 68 participants emphasise that emotional and stigma-related factors strongly influence adoption, underscoring the need for socially sensitive, human-centred AT that provide functional benefits while minimising stigma and discomfort. This thesis contributes to the field of AT by introducing a scalable, adaptive, and human-centred framework for fall risk assessment. It demonstrates that intelligent wearable multimodal systems can detect environmental hazards and integrate existing home safety interventions to provide real-time personalised support, enhancing safety, autonomy, and QoL for individuals with cognitive impairment.