Effective detection method for falls according to the distance between two tri-axial accelerometers
Jae Hyung Kim, Geun-Chul Park, Soo-Hong Kim, Soo-sung Kim, Hae-Rim Lee, Gye-Rok Jeon · Journal of the Korean Physical Society · 2016
Falls and fall-related injuries are a significant problem in the elderly population. A number of different approaches for detecting falls and activities of daily living (ADLs) have been conducted in recent years. However, distinguishing between real falls and certain fall-like ADL is often difficult. The aim of this study is to discriminate falls from fall-like ADLs such as jogging, jumping, and jumping down. The distance between two tri-axial accelerometers attached to the abdomen and the sternum was increased from 10 to 30 cm in 10-cm intervals. Experiments for falls and ADLs were performed to investigate the feasibility of the detection system for falls developed in this study. When the distances between the two tri-axial electrometers were 20 and 30 cm, fall-like ADLs were effectively distinguished from falls. The thresholds for three parameters — SVM, Diff Z, and Sum_diff_Z — were set; falls could be distinguished from ADL action sequences when the SVM value was larger than 4 g (TH1), the Diff_Z parameter was larger than 1.25 g (TH2), and the Sum_diff_Z parameter was larger than 15 m/s (TH3). In particular, when the SVM, Diff_Z, and Sum_diff_Z parameter were sequentially applied to thresholds (TH1, TH2, and TH3), fall-like ADL action sequences were accurately discriminated from falls.