Author | Year | Technology | Fall indicators | Classification method | Fall types | Subjects | Accuracy | Lead time |
---|---|---|---|---|---|---|---|---|
Wu | [47] | MCS | Horizontal and vertical profile of trunk velocity | Single THD based on the trunk velocity, THD = −1 m/s | 3 SMF: tripping, forward fall, backward fall | 3 Young adults | Not specified | 300–400 ms |
Lindemann et al. | [21] | A 3D ACCM, attached to head | Sum-vector of head acceleration | Multiple THD: (1) the sum-vector of 2D (i.e. planar) acceleration from head >2 g; (2) the sum-vector of 3D velocity >0.7 m/s | 7 SMF: forward fall, backward fall, sideway fall, fall to the back with hip flexion, fall backwards against a wall, imitation of a collapse, fall while picking up an object | 1 Young adult and 1 elderly | SEN = 100 %, SPE not specified | Not specified |
Nyan et al. | [33] | Three 3D GYRO, attached to sternum, waist and underarm | Angular rate of sternum, waist and underarm | Multiple THD method: (1) determine the THD (for angular rate) based on a series of ADLs: (2) the THDs were ranged from 100–170 degree/s for different fall indicators | 2 SMF: sideway fall; backward fall | 10 Young adults | SEN = 100 %, SPE = 92.5–97.5 % (different fall types) | 98–220 ms |
Bourke et al. | [8] | An integrated 3D ACCM and 3D GYRO, attached to chest | Negative downward vertical velocity of trunk | Single THD based on the trunk velocity, THD = 1.3 m/s | 4 SMF: forward fall with legs straight; forward fall with knee flexion; backward fall with knee flexion; side fall to right with knee-flexion | 5 Young male adults | SEN = 100 %, SPE = 100 % | 150–750 ms; mean = 323 ms |
Bourke et al. | [8] | A 3D ACCM, attached to chest | Negative downward vertical velocity of trunk | Single THD based on the trunk velocity, THD = 1.3 m/s | 4 SMF: forward fall with legs straight; forward fall with knee flexion; backward fall with knee flexion; side fall to right with knee-flexion | 5 Young male adults | SEN = 100 %, SPE = 100 % | 150–750 ms, mean = 323 ms |
Bourke et al. | [8] | A 2D GYRO, attached to chest | Negative downward vertical velocity of trunk | Single THD based on the trunk velocity, THD = 1.3 m/s | 4 SMF: forward fall with legs straight; forward fall with knee flexion; backward fall with knee flexion; side fall to right with knee-flexion | 20 Young male adults | SEN = 100 %, SPE = 100 % | Not specified |
Nyan et al. | [31] | (1) An integrated 3D ACCM and 3D GYRO, attached to chest; (2) an integrated 3D ACCM and 2D GYRO, attached to right thigh | (1) Sagittal and lateral angles of thigh; (2) correlation of angles between thigh and torso; (3) correlation of angular velocity between thigh and a pre-defined segment | Multiple THD: (1) the THD for thigh angles = ±10 degree; (2) the THD for correlation = 0.99 | 3 SMF: forward fall; backward fall; side fall | 10 Young adults | SEN = 95.2 %, SPE = 100 % | 727 ± 190 ms |
Nyan et al. | [32] | (1) An integrated 3D ACCM and 3D GYRO, attached to chest; (2) an integrated 3D ACCM and 2D GYRO, attached to right thigh | (1) Torso segment orientation; (2)thigh segment orientation | Multiple THD: (1) 2D thigh orientation THD = ±10 degree; (2) the correlation coefficient between the thigh segment and torso segment orientation >0.8; (3) the correlation coefficient between the body segment orientation and template >0.8 | 3 SMF: forward fall; backward fall; side fall | 13 Young male and 8 young female adults | SEN = 95.2 %, SPE = 100 % | Mean = 700 ms |
Wu and Xue | [48] | An integrated 3D ACCM and 3D GYRO, attached to anterior waist | Velocity of inertial sensor frame (i.e. the waist) | Single THD: the THD was first set to −1 m/s and then set to the maximum value of vertical velocity during the non-fall activities | 4 SMF: forward fall; backward fall; sideways fall; downward fall | 10 Young adults and 8 old adults (ADLs) | SEN = 100 %, SPE = 100 % | 70–375 ms |
Shi et al. | [41] | An integrated 3D ACCM and 3D GYRO, attached to waist | Waist acceleration and angular velocity | MLM: (1) setting up a database of “falling down” and “non-falling down”; (2) feature selection based on Principle Component Analysis; (3) derive Singular Vector Machine (SVM) classifier by data training; (4) Fall detection by SVM classifier and inflation of airbag for fall detection | 2 SMF: lateral fall with fast and slow fall motion | 2 Young adults | SEN = 100 %, SPE = 100 % | Not specified |
Tamura et al. | [43] | An integrated 3D ACCM and 3D GYRO, attached to waist | Waist acceleration and angular velocity | Multiple THD: (1) waist acceleration <3 m/s2 (assumed to be free fall); (2)angular velocity >0.52 rad/s; | 3 SMF: forward fall; backward fall; side fall | 16 Young adults | SEN = 93.1 %, SPE not specified | Lead time not specified; the detection time = 93–197 ms |
Shan and Yuan | [40] | A 3D ACCM, attached to posterior waist | Waist acceleration | MLM: (1) feature selection based on the acceleration data by a discriminant analysis; (2) support vector machine classifier was used | 3 SMF: forward fall; backward fall; side fall | 5 Young male adults | SEN = 100 %, SPE = 100 % | 182–228 ms, mean = 204 ms |
Zhao et al. | [49] | A 9-axial IMU (Xsens Tech. B.V.), attached to upper trunk | (1) Acceleration from upper trunk; (2) angular velocity of trunk | Multiple THD: (1) trunk acceleration >7 m/s2; (2) trunk angular velocity >3 degree/s | 3 SMF: forward fall; backward fall; side fall | 8 Young adults | Not specified | 257–329 ms |
Liu and Lockhart | (1) MCS; (2)an IMU attached to trunk; a 3D ACCM attached to thigh | Trunk sagittal angular angle and trunk angular velocity | Individualized THD extracted from a fall discriminant function | Slip induced falls | 10 Older adults | SEN = 100 %, SPE = 96.5 % | Mean detection time = 255 ms | |
Tong et al. | [44] | A 3D accelerometer, attached to chest | Acceleration time series (ATS) from upper trunk | MLM: (1) the ATS extracted from the accelerometer attached to the upper trunk was used to train a Hidden Markov Model (HMM); (2) The output of the HMM, known as the marching degrees (P), was compared with a pre-defined THD, i.e. P1 = 0.334 %; (3) Falls were detected if P < P1 | 2 SMF: forward fall; side fall | 8 Young adults | SEN = 100 %, SPE = 88.75 % | 200–400 ms |
Martelli et al. | [26] | MCS | Linear acceleration from all body segment | MLM: (1) The linear acceleration of all the body segments was parsed by independent component analysis; (2) a Neural Network was used to classify walking from unexpected perturbations | Slip induced falls | 15 Young adults | SEN = 92.7 %, SPE = 98 % | Mean detection, time = 351 ms |
Aziz et al. | [1] | A 3D ACCM and a 3-axis GYRO, attached to waist | Waist acceleration, velocity and angular velocity | MLM: (1) use the means and variances of X-, Y- and Z-axis accelerations, velocities and angular velocities to form the 18 features; (2) These features were used in Support Vector Machine to for activities classification | 7 Types of falls (involuntary and simulated): slip-induced fall; trip-induced fall; fall from hit or bump by an object or another person; collapse or loss of consciousness; misstep or cross step while walking and; incorrect shift of bodyweight while sitting down on or rising from a chair | 10 Young adults | SEN = 93.5−100 %, SPE = 85.6−99.7 % | 63–188 ms |
Hu and Qu | [13] | MCS | Five fall indicators, tested separately: head vertical acceleration, upper arm vertical velocity, trunk vertical velocity, shank frontal velocity, and head frontal angular velocity | Statistical: (1) an ARIMA model based statistical process control chart was constructed based on historical movement data; (2) The individual-specific control limit based on each fall indicator was used for fall detection | Slip induced falls | 60 Young adults | SEN = 88.5−94.7 %, SPE = 92.9−99.2 % | mean detection time = 620–710 ms |
Sabatini et al. | [38] | An integrated device with a 3D ACCM, a 3-axis GYRO and a barometric altimeter, attached to right anterior iliac spine | Downward vertical velocity at waist | Single THD based on the trunk velocity, THD = 1.38 m/s | 7 Types of falls (involuntary and simulated): slip-induced fall; trip-induced fall; fall from hit or bump by an object or another person; collapse or loss of consciousness; misstep or cross step while walking and; incorrect shift of bodyweight while sitting down on or rising from a chair | 25 Young adults | SEN = 80 %, SPE = 100 % | 40–300 ms mean = 157 ms |
Hu and Qu | [12] | MCS | Linear combination of body kinematic measures | Statistical: (1) define the fall indicator by a linear combination of two kinematic measures; (2) determine the weighting factor of the linear combination by optimization procedure; (3) The individual-specific control limit based on each fall indicator was used for fall detection | Slip induced falls | 60 Young adults | SEN = 97.3 %, SPE = 99.2 % | Not specified |
Kianoush et al. | [17] | Radio frequency | Difference of RF signal fluctuation between ADLs and falls | MLM: (1) RF power fluctuation was received by the wireless link in the cover area; (2) Hidden Markov Model (HMM) was trained by the RF signal perturbations; (3) The output from the HMM model, was used to distinguish falls from other activities | 2 SMF: fall from chairs; fall from stand | 2 Young adults | 100 % Accuracy | Not specified |
Lee et al. | [18] | A 9-axial IMU, attached to waist | Negative downward vertical velocity of trunk | Single THD based on the trunk velocity: THD = 1.2 m/s for fall vs. ADL, THD = 1.4 m/s for fall vs. Non-fall | 5 Types of falls (involuntary and simulated): slip-induced fall; trip-induced fall; fall from hit or bump; misstep fall; sit-to-stand fall; falls due to fainting; stand-to-sit fall | 11 Young male adults were asked to mimic the falling behavior of elderly | SEN = 97.4 %, SPE = 99.4 %, (falls vs. ADL), SEN = 95.2 % SPE = 97.6 % (falls vs. near-fall) | 184–231 ms |