Vision-Based Toilet Assistant for People with Dementia in Real-Life Situations

Abstract

Dementia, impacting over 55 million people and projected to reach 139 million by 2050, stands as a major cause of disability and dependence in older people globally. The challenges associated with dementia include the need for assistance in toileting, a distressing routine for both individuals with dementia and their caregivers. To address this issue, we present ToiletHelp, a system that utilizes a 3D sensor to detect difficulties in toilet use and provides customized step-by-step guidance for independent use while maintaining user privacy. ToiletHelp processes privacy-preserving depth maps on the device using a rule-based engine and convolutional neural networks to determine the need for assistance. It then guides the user through verbal and visual prompts. Our evaluation involves real-life scenarios with 20 individuals with dementia in two settings, achieving an accuracy of 80% and an F1-score of 0.83 in the recognition of user actions. This study underscores ToiletHelp's robustness and contributes to advancing assistive technology, fostering independence for people with dementia.