A classifier based approach to real-time fall detection using low-cost wearable sensors

In this paper, we present a novel fall detection method using wearable sensors that are inexpensive and easy to deploy. A new, simple, yet effective feature extraction scheme is proposed, in which features are extracted from slices or quanta of sliding windows on the sensor's continuously acceleration data stream. Extracted features are used with a support vector machine model, which is trained to classify frames of data streams into containing falls or not. The proposed method is rigorously evaluated on a dataset containing 144 falls and other activities of daily living (which produces significant noise for fall detection). Results shows that falls could be detected with 91.9% precision and 94.4% recall. The experiments also demonstrate the superior performance of the proposed methods over three other fall detection methods.