DETECTION OFPATIENT’S SIGN OFFALLSMaki Takeda, Nagisa Sasaki, Kayo YoshimotoTakeshi Ando, Sachiko Shimizu, Kenji Yamadaand Yuko OhnoDepartment of Health Sciences,Graduate School of Medicine, Osaka University CAUTION
BACKGROUND Nurse No problem. Nurse stationInpatient It is difficult to predict falls.
PROPOSED SYSTEM Motion detection Motion prediction Video sensor Catch Roll overthe sign or Patient’s room Wake up Bayesian approach
METHOD & EXPERIMENT Center of gravity of head Bayesian approach from original image a x, y Lx, y a a ������ ������ : prior distribution ������ ������ ������, ������ : posterior distribution ������ ������, ������ ������ ：likelihood Culculating velocity of ������： x-coordinate of center of center of gravity of head gravity of head ������： y-coordinate of center ofVelocity [pixel/frame] 10 25frame ? gravity of head 5 ������： regression coefficient 0 Estimate gradient by 0 20 40 60 80 Frame maximum likelihood estimate Motion of posterior distribution detection
EXPERIMENTAL RESULTS Subjects performed 3 situations on bed ( wake-up, right or left roll-over) 150 X-axial displacement [pixel] 120 起き上がり Wake-up 90 右 Right roll-over 60 左 Left roll-over 30 0 0 0.1 0.2 0.3 0.4 Estimated gradient It is possible to classify wake-up or roll-over by estimated gradient and x-axial displacement of center of gravity.