• healthcare
• wireless sensor networks
• fall detection systems

Falls are a mayor cause of injury deaths and injury-related hospitalization among people older than 65 years. Even non-injurious falls can be really dangerous, as it has been shown that many elderly people lack the ability to stand up and remain on the ground for even longer than an hour. This is the so called "long-lie", and it has been proved to have devastating effects on health. At the same time, fall related admission of older adults are a significant burden to the health services world wide.

Therefore, in the latest years there has been a great interest on fall detection systems by the healthcare industry. Many attempts to solve this problem using wearable devices has been already made: prevalent methods use a fixed accelerometer threshold to isolate falls from the "activities of daily living" (ADL). This approach does not succeed in distinguishing actual falls from certain fall-like activities such as sitting down or lying quickly, and causes frequent false alarms. In order to improve the detection accuracy, some researchers propose to combine linear acceleration obtained from accelerometers with gyroscope measurement of angular velocity. Unfortunately gyroscopes have a really negative impact on device battery lifetime. Body orientation has also been used to improve detection, but it is not very useful as there is no clear connection between posture and fall.

In this work we present a novel method for false alarm recognition and filtering, which leads to a significantly improved level of detection accuracy. Our approach features low computational costs and real time response. Moreover, it requires only an accelerometer placed at user's waist, achieving a high degree of usability.