WATCHDOG: Autonomous Elderly Assistance Via Attention-Based Fall Detection and Trajectory Prediction
Antonello Longo, Annaclaudia Bono, Giovanna Guaragnella, Pietro Boccadoro, Antonio Petitti, Arianna Rana, Tiziana D'Orazio
AI summary
Problem
Vision-based fall detection and people-following systems struggle with temporary occlusion, static viewpoints, and lack integrated solutions for continuous monitoring, while wearable alternatives face poor user compliance.
Approach
The system uses a dual-branch transformer architecture on a mobile legged robot to jointly classify fall behaviors from 2D body landmarks and predict future 3D hip trajectories, linked by cross-attention to maintain tracking during occlusions.
Key results
- 99.71% fall detection accuracy on the FALL-UP dataset
- 0.14 m average displacement error for trajectory prediction on Human3.6M
- 50% reduction in inference time and computational load compared to baselines
- Real-world validation of robust people-following and fall detection in occluded environments
Why it matters
Provides a scalable, non-invasive solution for continuous elderly safety monitoring, advancing autonomous care robotics and addressing critical needs for aging populations.
Abstract
Service robots designed to assist elderly people are receiving significant attention since they can improve their quality of life, promote their independence, and provide daily support. These mobile platforms can observe people moving around their homes, recognize dangerous events, and detect them promptly. This paper introduces a novel framework to perform fall detection and people following on board an autonomous legged robotic platform. The system operates on the Unitree Go2 robot and comprises two main building blocks. The first component consists of a Body Landmarks extractor and a Transformer-based network that performs binary classification, distinguishing between Fall behaviours and Activities of Daily Living (ADL). The second component is a target-driven path planner that enables the robot to follow and maintain a full-body view of the target in complex environ- ments. Experiments on public datasets and comparison with state-of-the-art works have been conducted to demonstrate the reliability of both blocks. Real experiments in a cluttered environment have been performed to illustrate how the mobile platform is able to follow people moving around obstacles, detect falls in occluded areas, and predict people’s trajectories to maintain a full-body view. Code and additional material are available at the following link: https://github.com/Antus8/WatchDog.