TABLE 2.
Inertial sensor-based gait and balance analysis: from the prospect of event detection, spatiotemporal parameter estimation, joint ROM, and balance analysis.
| Reference | Parameter(s) | Technique(s) used | Subject(s) | Calibration/validation technique | Remarks |
| Event detection/temporal parameter measurement | |||||
| (Jasiewicz et al., 2006) | HS and TO | One gyroscope, two linear accelerometers, peak detection, zero crossing, heuristic | 26 HC + 14 SCI + Charcot-Marie-Tooth (CMT) | Foot switches | TO latency 50 ms, 100 ms for HS detection; obtrusive due to semi-wired connectivity |
| (Raveendranathan et al., 2011) | HS and HO | Single 3-axis accelerometer at alternate/multiple positions; peak detection + HMM | 1 HC | None | Adaptive to Sensor placement |
| (Yang et al., 2012) | HS and asymmetry feature | Single accelerometer placed at the lower back, peak detection | 15 HC | N/A | No specific accelerometer; The developed iGAIT tool requires manual intervention to set input pre-sets |
| (Mariani et al., 2013) | HS, TO, HO, and TS | Foot mounted 3D accelerometer+ gyroscope, pitch velocity, negative peak, zero crossing | 10 HC, 12 AO, 11 TAR, and 9 AA | Pedar-X Pressure Insole | −33 ± 14 for angular velocity, 81 ± 15 for acceleration |
| (Hundza et al., 2013) | Temporal, stride length | 4 IMU (gyroscope), Y-angular rate reversal | 6 (PD) + 7 (HC) | GAITRite, OMC | 100% event detection, SD of 6.6 ms and 11.8 ms in HC and PD |
| (Joshi et al., 2016) | TO | Three-axis accelerometer, wavelet decomposition | 6 HC | Foot switch | The transition between level ground and ramps |
| (Das et al., 2019) | HS and TO | Six-axis IMU, Foot angle variation, peak detection | 34 HC | FSR | Improved detection latency of 16 ms |
| Joint kinematics and ROM | |||||
| (Dejnabadi et al., 2005) | Knee angle | One IMU (Two accelerometer + one gyro) placed at shank and thigh; virtual projection of physical sensor into rotation joint | 8 HC | OMC | Absolute angle calculation with no drift error; subject-specific modeling requires prior anatomical information |
| (Dorschky et al., 2019) | Hip, knee, and ankle angle | 07 six-axis IMU, musculoskeletal model, trajectory optimization | 10 HC (M) | OMC | P ≥ 0.93 |
| (Gholami et al., 2020) | Hip, knee, and ankle flexion/extension | 01 accelerometer placed at foot + CNN | 10 HC (M) | OMC | RMSE <3.4% for intra-subject and <6.5% for inter subject |
| Spatiotemporal parameters | |||||
| (Salarian et al., 2012) | HS, TO, SL, and gait velocity | Two gyroscopes placed at the shank, a Double pendulum model with two gyroscopes + Fourier series, and most minor square optimization | 10 PD, 18 HC, 36 hip-replacement, and seven orthosis | OMC | Validated on a sizeable patient population with multiple disorders |
| (Takeda et al., 2014) | Gait phases, SL, and LStep | One IS on each ankle, shank, and thigh; one on the pelvis. Peak detection for events, drift reduction protocol for spatial parameters | 5 HC, 10 m walk-test | OMC | Linear drift modeling does not hold for extended walking |
| (Rampp et al., 2015) | SL, GCT, Tswing, and Tstance | Inertial sensors, Template Search for events | 101 NW, 84 WW | GAITRite® | 0.93 and 0.95 in NW and 0.80 and 0.95 in WW for SL and GCT, respectively |
| (Wang and Ji, 2015) | Tstance, Tswing, and SL | Foot mounted IS, peak-peak detection+ adaptive thresholding for event detection; CF+ ZUPT+ double integration for SL | 15 HC | Non-standard | 1.64 ± 0.839 for SL |
| (Liu et al., 2016) | SL | 3D acceleration and angular rate, Dual-ZUPT | 14 steps | Videography | |
| (Ferrari et al., 2015) | GCT, SL, and stride velocity | Foot mounted IMU, Medial-lateral foot angle peak detection for events; KF+ZUPT for stride length | 12 HC, 16 PD | GAITRite® | Real-time computation on a smartphone, RMSE SL = 4% |
| (Hao et al., 2019) | SL | 3D Euler angle, acceleration, discrete KF, smoother | 9 HC (male adults) | OMC | −0.24 ± 1.1 cm for SL |
| (O’brien et al., 2019) | Tstance, TSw, SL, step velocity, and step count | One IMU at hip; Local minima/maxima + Butterworth filter for events; IPM + Double integration for SL | 51 HC | GAITRite® | Need for additional optimization constant that is derived from GAITRite® for SL estimation |
| (Das and Kumar, 2021) | SL | Six-axis IMU at foot dorsum; foot angle for gravity compensation and double integration of foot acceleration | 10 HC | Zebris walkway, outdoor marking | Acceleration integrated only for swing duration; compensated with foot length |
| Spatiotemporal + Joint kinematics | |||||
| (Teufl et al., 2019a) | 12 STP including Lstp, step width, 6 DoF kinematics | 7 Xsens IS | 24 HC (12 M + 12 F) | OMC | Detection means error ∼1.6%, Step width, and swing width RMSE > 30% |
| (Yeo and Park, 2020) | Stt, SL, cadence, step length; knee and hip ROM | Five triaxial accelerometers, gyroscope, and magnetometer (LEGSys+ wearable device) placed at shank, thigh, and pelvis; self-selected walking at the 7-m walkway | 30 HC | OMC | The significant difference in hip ROM; measurement within 95% limit of agreement |
| Balance | |||||
| (Hsu et al., 2014) | CoM, postural sway rate | Three-axis accelerometer in waist | 21 PD + 50 HC | N/A | Validated on a large group; Only static balance |
| (Wang W.-H. et al., 2015) | TUGT | Three IMUs (1 at hip + 1 at each foot); Signature matching of lateral angular rate + thresholding | 21 AD + 25 HC | N/A | Test specific |
| (O’brien et al., 2019) | 10 MWT, BBS, and TUGT | One IMU at hip; FFT+ integration for static balance; Daubechies wavelet approximation for dynamic balance | 51 HC | GAITRite® | 178 features extracted for three balance assessment tests |
| (Noamani et al., 2020) | Two minutes standing test, inter-segmental moments, and CoP | Accelerometer + gyroscope placed at foot, leg, pelvis, and head-arms-trunk; Musculoskeletal inverse dynamics model | 10 HC | OMC+ force plates | Accelerometers alone provide reliable data for standing balance analysis |
| (Dugan et al., 2021) | Two minutes barefoot standing in EO, EC | 17 IMU placed at whole body; jerk index and complexity index from postural sway from pelvis accelerometer | 38 concussed patients | N/A | Single accelerometer yields information about postural sway |
HS, heel strike, TO, toe off, HC, healthy control, SCI, spinal cord injury, CMT, Charcot-Marie-Tooth, HMM, Hidden Markov Model, HO, heel off, TS, toe strike, AO, ankle orthosis, TAR, total ankle replacement, AA, ankle arthrodesis, OMC, optical motion camera, CNN, convolution neural network, CF, complementary filter, ZUPT, zero update, KF, Kalman Filter, GCT, Gait cycle time, IPM, inverted pendulum model, Stt, stance time, TSw, swing time, Lstp, step length, CoM, center of Mass, TUGT, time-up and go test, 10 MWT, 10 meter walk test, BBS, Berg Balance Scale, AD, Alzheimer’s disease, FFT, Fast Fourier transform.