Smartphone-based fall detection algorithm using feature extraction

The danger of falling among the elderly is a public concern and is becoming an important issue that needs further attention. Sensors embedded in smartphones provide information about user activity, such as the accelerometer which is widely used in fall detection. In this paper, we propose a fall detection algorithm which is formed by feature extraction processing and recognition processing. A total of six features were calculated in feature extraction processing. Four of them are related to the gravity vector which is extracted from accelerometer data by using low-pass filtering. As falling mostly occurs in a vertical direction, the gravity-related features are useful. In recognition processing, a set of six features was clustered by support vector machine. The main feature - acceleration in the gravity vector direction - contains the vertical directional information and provides a distinct pattern of fall-related activity. This feature acts as a trigger-key in recognition processing to avoid false alarms which lead to excessive computation. The results show that our algorithm could achieve a sensitivity of 96.67% and specificity of 95%.