A high accuracy and low latency patient-specific wearable fall detection system

Falls are a critical public health issue among elderly people that requires continuous monitoring. This paper presents a patient-specific single sensor fall detection system that utilizes a tri-axial accelerometer data measured from the patient's trouser pocket to distinguish between activities of daily living (ADL) and falls. The proposed system which is implemented on FPGA provides the following novel features: 1) patient-specific single-threshold detection algorithm, 2) minimum latency for trigging a fall event of 1-sec, 3) reduced computation complexity in the detection algorithm, and 4) local memory for storing a fall event scenario. The proposed system achieves a sensitivity of 98.1% and a specificity is 99.2% for a total set of 500 measured ADL and fall events from 57 subjects.