Dataset improves detection of falls and daily activities including wheelchair use
SAFER-Activities: A Dataset for Smart Assessment of Fall Events and Routine Activities
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Detecting falls and everyday physical activities accurately can help keep people safe, especially those with mobility challenges. The authors created a large video dataset called SAFER-Activities that shows 30 different actions with detailed frame-by-frame labeling. This dataset also includes videos specifically showing wheelchair users. They tested different computer models to recognize these actions and found some work better when seeing new environments. Their dataset and code are now available to help develop better health monitoring systems.
What this means in practice
- •For healthcare device developers: Create more accurate fall detection and activity monitoring features by training models on SAFER-Activities dataset including wheelchair-specific data.$Commercial implications: This dataset enables development of health-monitoring devices with robust fall detection marketed to elderly and mobility-challenged users.
- •For security monitoring teams: Deploy video analytics that recognize falls and routine activities more reliably in environments with varied camera setups and user conditions.
Authors
Diwas Lamsal, Pramod Wickramatilake, Jednipat Moonrinta, Mongkol Ekpanyapong, Matthew N. Dailey
Abstract
Smart healthcare monitoring systems require precise action recognition to ensure well-being and timely intervention in critical situations such as falls, particularly for mobility-challenged individuals. Existing datasets are often clip-based, lacking the frame-level detail needed to recognize actions online, as they unfold. To address this, we introduce SAFER-Activities, a dataset for fall detection and physical activity monitoring, with a dedicated subset for wheelchair use scenarios. It comprises over 66 hours of video data captured by multiple cameras, with 85,310 action instances and frame-level annotations for 30 action classes. We benchmark action recognition on SAFER-Activities with 2D and 3D skeleton models, RGB models with frozen backbones, and multimodal fusion strategies, and evaluate on in-lab, out-of-distribution, and cross-dataset test sets. Skeleton-based models generalize best under domain shift; fusing frozen RGB features with the skeleton stream improves in-domain recognition over the baseline CNN1D, most clearly on the wheelchair subset, but degrades out of distribution. Cross-dataset and qualitative evaluations confirm that models trained on SAFER-Activities transfer well to unseen environments and external fall data. To support research on robust fall detection and activity monitoring, we release the dataset and code at https://safer-activities.github.io/.