A Data-Driven Monitoring Technique for Enhanced Fall Events Detection

Fall detection is a crucial issue in the health care of seniors. In this work, we propose an innovative method for detecting falls via a simple human body descriptors. The extracted features are discriminative enough to describe human postures and not too computationally complex to allow a fast processing. The fall detection is addressed as a statistical anomaly detection problem. The proposed approach combines modeling using principal component analysis modeling with the exponentially weighted moving average (EWMA) monitoring chart. The EWMA scheme is applied on the ignored principal components to detect the presence of falls. Using two different fall detection datasets, URFD and FDD, we have demonstrated the greater sensitivity and effectiveness of the developed method over the conventional PCA-based methods.