Gesture-aware fall detection system: Design and implementation

Falling is the most frequent accident for the elderly in daily life. Falls can cause seriously injuries and even death. Although many fall detection methods have been developed to detect falls in real-time, most are inaccurate and inconvenient to use. Previous studies failed to correctly identify the relationship between normal activities and falls. Another problem is that many proposed methods considered only one device placement and did not investigate how different positions influence recognition accuracy. This study proposes the Gesture-Aware Fall Detection (GAFD) system, which uses a smartphone worn on an arm, the chest, waist or thigh to provide caregivers and instantaneously notify caregivers. While the GAFD system enhances willingness to wear a fall detection device, the difficulty of detection accuracy is increased. To improve fall detection accuracy, the GAFD system uses the back-propagation neural network to establish an awareness model that analyzes historical data of a subject´s gestures, and then establishes an empirical model of the subject´s gestures to improve identification accuracy. Experimental results show that the proposed system accurately identifies falls. The correction rate for gestures and fall detection is influenced by the position at which a subject wears the smartphone. Finally, strengths and weaknesses of different wear positions are discussed.