Automatic Fall Risk Detection Based on Imbalanced Data

In recent years, the declining birthrate and aging population have gradually brought countries into an ageing society. Regarding accidents that occur amongst elderly, falls are essential problem quickly causes indirect physical loss. this paper, we propose a pose estimation-based fall detection algorithm to detect risks. We use body ratio, acceleration deflection as key features instead of using keypoints coordinates. Since data is rare in real-world situations, train evaluate our approach highly imbalanced setting. assess not only different handling methods but also machine learning algorithms. After oversampling on training data, K-Nearest Neighbors (KNN) achieves best performance. The F1 scores for three classes, Normal, Fall, Lying, 1.00, 0.85 0.96, which comparable previous research. experiment shows more interpretable with feature from skeleton information. Moreover, it can apply multi-people scenarios has robustness medium occlusion.