Evaluation of a low-complexity fall detection algorithm on wearable sensor towards falls and fall-alike activities

Fall accidents cause severe damage to health, sometimes even mortality in older adults. With the increasing number of the elderly suffering from fall events, wearable products are in great demand. However, most of them for fall detection have difficulty in reducing false positive caused by fall-alike activities. In this study, we focus on evaluating the accuracy of fall detection among a set of fall-alike activities using a low-complexity fall detection algorithm and a 3-axis accelerometer. Quantitative evaluation in controlled study tunes the algorithm's parameters and provides us a 90% fall detection accuracy. The experiment result shows that jumping onto bed followed by a rest has the highest false positive rate of 45% and running followed by a sudden stop reaches 32%, while running upstairs or downstairs and standing quickly from sofa is less confusing with the false positive rates of 20% and 5%, respectively. The false positive rate is decided by the sensitivity of the threshold and the intensity of motions in the experiment. We also perform a 10-hour longitudinal study on real-life activities of one subject. In the longitudinal real-life pilot study, the low-complexity algorithm demonstrates the high accuracy, which indicates its effectiveness in real life.