Toward Dignity-Aware AI: Next-Generation Elderly Monitoring from Fall Detection to ADL

TL;DR

This position paper envisions a next-generation elderly monitoring system that moves beyond fall detection toward the broader goal of Activities of Daily Living (ADL) recognition, and outlines open challenges such as domain shift, data scarcity, and privacy risks.

Abstract

This position paper envisions a next-generation elderly monitoring system that moves beyond fall detection toward the broader goal of Activities of Daily Living (ADL) recognition. Our ultimate aim is to design privacy-preserving, edge-deployed, and federated AI systems that can robustly detect and understand daily routines, supporting independence and dignity in aging societies. At present, ADL-specific datasets are still under collection. As a preliminary step, we demonstrate feasibility through experiments using the SISFall dataset and its GAN-augmented variants, treating fall detection as a proxy task. We report initial results on federated learning with non-IID conditions, and embedded deployment on Jetson Orin Nano devices. We then outline open challenges such as domain shift, data scarcity, and privacy risks, and propose directions toward full ADL monitoring in smart-room environments. This work highlights the transition from single-task detection to comprehensive daily activity recognition, providing both early evidence and a roadmap for sustainable and human-centered elderly care AI.

Figures (10)

• Figure 1: Conceptual system-level overview of the proposed federated monitoring framework. Current experiments use SISFall IMU data only. Other modalities (e.g., LiDAR, pressure mats) are part of our planned smart-room deployments.
• Figure 2: Simulated Elderly Room Testbed with Anonymous Sensors at Toyohashi University of Technology (TUT). Data collected from non-identifiable sensors such as floor pressure, LiDAR, and motion sensors are processed on Edge AI devices installed in each home. Anomaly detection models are locally trained on these devices. Through federated learning, only the parameters $(\theta)$ are aggregated and updated in the cloud, without transmitting personal data, thereby enabling the construction of a high-accuracy model while preserving privacy.
• Figure 3: Conceptual spatiotemporal pipeline with Conv-GRU and planned GAT fusion. In this position paper, GAT is a planned extension for multi-sensor ADL datasets; current experiments use Conv-GRU on SISFall IMU data only.
• Figure 4: cGAN architecture for generating synthetic SISFall sequences (IMU only).
• Figure 5: t-SNE visualization comparing real and GAN-synthesized SISFall data distributions. The close overlap indicates that synthetic sequences achieve high fidelity relative to real data.