Table 2.

Key points pertaining to insole-based systems used for fall risk assessment and fall detection.

Reference	Sensor system and focus of work	Sensor type(s) and number	Sensor placement	Population	Parameters investigated	Data acquisition and analysis information	Other characteristics
Kraus et al. (67)	• Insole-based • Physical frailty prediction	Pressure sensors and 6-axis IMU		• Orthogeriatric patients • n = 57 • 93% women • Mean age: 77 years (SD = 6)	• Number of steps • Stride length • Gait speed • Acceleration over gait cycle • Gait cycle time • Cadence • COP variability • Double support time	• Physical frailty classified using:   – SPPB (Short Physical Performance battery)   – kNN (k-Nearest Neighbor)   – RF (Random Forest)	• Moticon Science3 insole used
Ayena et al. (63)	• Insole-based with UWB (Ultra-wideband) Radar • Fall risk assessment	Piezoresistive pressure sensors (FSRs) + 3D Accelerometer + Radar system 5 sensors: • 4 FSRs (2 at hindfoot (medial heel and lateral heel), 1 at first metatarsal, and 1 at fifth metatarsal) • 1 3D Accelerometer (outside of sole)		• One healthy young adult participant	• Instrumented insole provides:   – Acceleration-related information   – Other gait information (Temporal features)   – Cadence   – Stride time   – Stride length   – Stride speed • Radar UWB provides information regarding:   – Position-based activities (Spatial features)   – Stride length   – Stride speed	• Risk of Falling Score informed by stride data • Fall risk based on analyzed variability • Fall detection (based on static and dynamic acceleration) • Kalman Filter for Gait Velocity estimation • Algorithm to segment TUG (Timed Up and Go) radar signal: used for stride length, stride time, cadence, stride speed	• FSR diameter: 13 mm • High resolution acceleration measurement: 13-bit and up to ± 16 g • Radar range: 10 m • Radar accuracy: ±10 cm • Low-power radar system
Bucinskas et al. (64)	• Insole-based • Fall Risk Assessment	3 piezoelectric pressure sensors • Pressure sensors developed by researchers using PVC, Velostat and aluminum foil		• One participant (three trials)	• Pressure distribution • Duration of stance phases for both feet • Variation of stepping abruptness • Stepping unevenness parameters • Stepping rhythm • Size of step • Gait phase timing	• Analysis of sensor signals in time domain • Correlation-regression analysis for absolute measurement error • Single amplitude values extracted from raw data for load distributions	• Wireless (2.4 GHz WiFi) • Battery-powered 1,300 mAh lithium polymer battery • Activities investigated: “Turnaround,” “Scrolling,” “Upstairs,” “Downstairs,” “Upstairs one by one,” “Walk with left straight leg”
Chen et al. (65)	• Insole-based • Fall Hazard Identification	Pressure sensor array layer (96 pressure sensors)		• Healthy individuals • n = 10	• Ground reaction force differences • Swing phase acceleration magnitude signal threshold crossing points • Pitch angle at initial foot contact • Pitch angle during midstance • Double support %	• Five features used to train SVM model for fall hazard identification and safe floor activities • One-feature accuracy: 39.54% • Five-feature accuracy: 95.78%	• Device for fall hazard identification • Activities investigated: Walking, Running, Stair ascent, Stair descent
Ji et al. (66)	• Insole-based • Fall Detection	4 FSR pressure sensors (2 at forefoot, 1 at midfoot, 1 at hindfoot)		–	• Plantar pressure • Plantar pressure variation • Walking state	–	• Bluetooth data transmission
Antwi-Afari et al. (60)	• Insole-based • Fall Risk Assessment	• 13 Capacitive Pressure sensors: 2 at Toes; 3 at Metatarsal Head; 4 at Arch; 4 at Heel • 1 3D Acceleration sensor at middle of Arch		• Construction workers • n = 10 • Mean age: 26.50 (SD = 3.35) years	Biomechanical gait stability parameters: • Mean pressure • Peak pressure • Pressure-time integral • Anterior/Posterior center of pressure • Medial/Lateral center of pressure (Investigated through simulation of loss-of-balance events and normal gait)	• 50 Hz pressure sampling rate • Equations used for biomechanical gait stability parameters: • Mean Pressure = $\frac{1}{N}\sum _{i=1}^{N}Pi$ • Peak Pressure = Maximum (Pi, …, PN) • Pressure-Time Integral = $\sum _{t=1}^{N}Pi\times t$ • Anterior/Posterior Center of Pressure = $\frac{\sum _{i=1}^{N}XiPi}{\sum _{i=1}^{N}Pi}$ • Medial/Lateral Center of Pressure = $\frac{\sum _{i=1}^{N}YiPi}{\sum _{i=1}^{N}Pi}$ N = number of pressure sensors i = pressure sensor value (ith sensor) Xi and Yi = pressure sensor value coordinates	• Wireless Data Transmission • Insole thickness: 2.5 mm • 16 MB flash memory integrated in sole • Pressure range: 0 to 40 N/cm2 • Simulated loss-of-balance events = Slip, Trip, Unexpected step-down, Twisted ankle
Cates et al. (61)	• Insole-based • Fall Classification	• 4 Pressure sensors (FSRs): 2 at forefoot and 2 at hindfoot • 1 IMU Sensor at midfoot		• Healthy males • n = 20 • Age: 28 ± 5 years	Low-acceleration Activities of Daily Life (ADL): • Standing • Lying • Sitting • Walking • Running High-acceleration ADLs: • Stair ascent • Stair descent • Jump falls	• Threshold and machine learning methods • Signal of sum vector magnitude filtered using 1st order low-pass butterworth filter (1Hz cut-off) • Support vector machine (SVM) fall detection algorithm • 45 features used for fall classification model • Feature selection using genetic algorithm process • 18 features associated with highest performance of fall detection	• Device for fall classification • 20 Hz sampling rate
Hu et al. (62)	• Insole-based • COP Trajectory	12 FSRs (at toes, metatarsophalangeal joints, foot arch, heel)		• n = 20 • Younger participants:   – n = 10   – Age: 22.6 ± 1.5 years • Older participants:   – n = 10 • Age: 65.7 ± 3.4 years	Center of Pressure (COP) Trajectories (indicative of postural control) • Anterior-posterior direction trajectory • Medial-lateral direction trajectory	Non-linear model used to estimate COP more accurately than typical weighted models	• 50 Hz sampling frequency • FSR diameter: 12.7 mm • Bluetooth data transmission
di Rosa et al. (59)	• Insole-based • Fall Risk Score	Pressure sensors + 6D Accelerometer and Gyroscope (Sensors embedded in two layers: Pressure array layer on upper pressure-sensing layer; Other components (including inertial sensors) in second/matrix layer)		• Older adults (over 65 years) • n = 29 • Diverse sample used (sex, health status, mobility, etc.)	• Double support right (fall risk index weighting: 52%) • Single support left (weighting: 31%) • Mediolateral average acceleration amplitude (weighting: 12%) • Heel strike force slope left (weighting: 5%)	• Cluster analysis   – Selected indicators:   – POMA (Performance Oriented Mobility Assessment Tool)   – DGI (Dynamic Gait index)   – TUG (Timed Up and Go test)	• Short range communication (Bluetooth) to mobile device • Long range communication from mobile device to computer • Worn during daily activities for 2 weeks • Comprehensive daily activity monitoring not possible (device designed for steady-state gait parameters only)
Das and Kumar (58)	• Insole-based • Postural Stability and Gait Parameters	7 Piezoresistive pressure sensors • Hallux • Metatarsal 1 • Metatarsal 2 • Metatarsal 4 • Metatarsal 5 • Medial Heel • Lateral Heel		• Healthy males • n = 3 • Age range: 22–28 years	Postural stability and spatiotemporal gait parameters: • Plantar Pressure • Force variation during standing and from accidental falls • Gait cycle duration • Stance duration • Swing phase time • Single support	Data filtered using 3rd order Butterworth low-pass filter (cut-ff frequency = 50 Hz)	• Parameters calculated using Heel strike, Heel off, Toe off, Toe strike, Timer • Timer is triggered upon detection of specific gait events • Force range: 0–100 N • Accuracy: ± 2 N
Lincoln and Bamberg (57)	• Insole-based system + camera-based system • Slip Detection	• 6 pressure sensors (FSRs): 4 at forefoot and 2 at heel • 1 3-axis accelerometer (not in sole)		• n = 2 • 1 male (age = 23 years) • 1 female (age = 35 years)	Plantar force during slip gait • Normal force • Lateral shear force • Progressional shear force • Body weight acceleration during slip gait	Pressure and acceleration data filtered through low-pass butterworth filter (cut-off frequency = 60 Hz)	• Real-time slip detection • Slips following heel strike were investigated without falls recorded • 114 Hz sampling rate • 90% accuracy