Accelerometry-based motion pattern analysis for reliable fall detection

Falls are the primary cause of accidents for elderly people and often result in fatal and non-fatal injuries that are associated with a large amount of medical cost. The reliable and effective fall detection system can reduce the fear of falling and give the urgent medical support. Nowada ys, the advances in MicroElectroMechanical Systems (MEMS) and Information and Communication T echnologies (ICT) have made the wearable fall-detecting sensors more possible. However, the assessment of falls is difficult due to the subtle and complex body movement which requires precise and reliable fall-detecting mechanism. Considering the wearing comfortability and convenience, we use a tri-axial accelerometer mounted on the wrist to capture movement data of the human body, and propose a three-stage fall detection algorithm to detect the free fall, the impact and the vibration after impact from the fall patterns respectively. The volunteers are 4 young people that do both simulated falls and Activities of Daily Living (ADLs) and 3 healthy elders that do only the ADLs. The collected motion data are analyzed by MATLAB to establish the thresholds of three-stage fall detection algorithm that includes: the Sum Vector Magnitude of Accelerations (SVM _a) is smaller than 0.75 G continued for 25 ms; the vertical acceleration is greater than 4.5 G; and the difference between the signal magnitude area of SVM _a and the sum of tri-axial signal magnitude areas is greater than 1.5 m/s. The experimental results show that the three-stage fall detection algorithm achieves 92.71% sensitivity, 100% specificity, 96.11% accuracy and 100% precision. The mechanism is not only cost effective but also portable that fulfills the requirements of fall detection.