Application of single camera-based gait analysis for accessible fall risk assessment among community-dwelling older people

초록

Falls represent a significant public health challenge for older adults, given their high incidence and potential for severe outcomes. Proactive identification of individuals at high fall risk through effective assessment is crucial for implementing targeted prevention strategies. Despite numerous fall risk assessment solutions available, a notable gap persists in objective and readily accessible assessment tools. Addressing this need, we proposed a novel application of video-based gait analysis for fall risk assessment utilizing a single red-green-blue (RGB) camera. Key spatiotemporal gait parameters were derived for risk modeling using two-dimensional human pose estimation and a predefined distance measure. This approach advances beyond existing frameworks that mainly rely on wearable sensors or depth cameras by providing a more practical and accessible alternative. To assess the efficacy of this approach, an experimental study was conducted with 99 community-dwelling older adults. The findings revealed significant differences in gait parameters between individuals categorized into high and low fall risk groups. Among commonly used classification algorithms, support vector machine (SVM) classifier yielded optimal performance, achieving an accuracy of 86.35 % alongside balanced sensitivity (87.55 %) and specificity (85.16 %). Model interpretability was further substantiated by consistent feature importance patterns observed across cross-validation folds, elucidated by SHapley Additive exPlanations (SHAP). Compared to current state-of-the-art fall risk assessment studies, our single RGB camera-based methodology demonstrates a more favorable balance of accuracy and practicality. This solution exhibits substantial promise for wider adoption in facilitating routine and more accessible fall risk screening for the older people in their daily environments.

키워드