Evaluation of the biomechanics of lower limbs in normal, overweight, and obese people using the SMART F-IMU system Authors: Udomporn Manupibul, Wimonrat Jarumethitanont, Ratikanlaya Tanthuwapathom, Weerawat Limroongreungrat, Warakorn Charoensuk Published: 2026 ABSTRACT This study provided a pilot evaluation and proof of concept for the proposed SMART F-IMU system, designed to investigate gait characteristics in normal-weight (NW) and overweight/obese (OW/OB) participants. This system combined modified inertial measurement units (IMUs) and flexible force sensors to provide estimates of ground reaction force (GRF), joint angles, and spatiotemporal gait parameters. The SMART F-IMU appeared to address the limitations of traditional gait analysis methods, enabling convenient and affordable assessments in real-world settings. The OW/OB group exhibited significantly longer stride, stance, swing, double support, and cycle times, alongside decreased cadence and swing phase percentages, all demonstrating large effect sizes (> 1.0). Statistical significance (p < 0.01) and large effect sizes (> 0.8) were observed in the hip angle at initial contact, knee maximum extension angle in the stance phase, knee angle at toe-off, knee maximum flexion angle in the swing phase, ankle maximum plantarflexion angle in the pre-swing phase, hip adduction angle in the terminal stance, knee maximum adduction angle in the stance phase, and knee maximum adduction angle in the swing phase. This study’s results demonstrated the feasibility of the SMART F-IMU for detecting weight-induced gait alterations in non-laboratory settings, highlighting its potential for longitudinal monitoring of joint biomechanics during daily activities. Introduction Obesity and overweight are major global health concerns characterised by excessive fat accumulation, resulting from an imbalance in calorie intake and expenditure. In 2022, the World Health Organization (WHO) documented that over 890 million adults were obese, and more than 2.5 billion were overweight1. The prevalence of OW/OB adults aged 18 and older increased from 25% in 1990 to 43% in 2022 (43% men and 44% women). OW/OB prevalence varied by region, ranging from 31% in the WHO’s African and South-East Asian regions to 67% in the Americas. Overweight is defined as a body mass index (BMI) of more than 25 kg/m², while obesity is defined as having a BMI above 30 kg/m². Obesity presents greater health hazards for women compared to men. OW/OB increase the risk of numerous health complications, including cardiovascular diseases, type 2 diabetes, cancer, stroke, musculoskeletal disorders, and joint OA1,2. Excessive body weight adversely affects bones, joints, and muscles, leading to pain, decreased mobility, altered gait patterns, and reduced quality of life. To prevent these risks, evaluating gait biomechanics is essential, as increased body mass inherently alters ground reaction forces (GRF) and heavily impacts forefoot pressure during daily ambulation3. Traditional laboratory methods often lack the capacity to capture these dynamic, real-world kinetic shifts4. Consequently, leveraging wearable force sensors and localised insole pressure technology has emerged as a critical methodology for continuously tracking GRF adaptations outside a laboratory environment5,6. Thus, evaluating gait biomechanics in OW/OB participants is crucial to preventing and managing these severe musculoskeletal complications. Gait analysis, the most efficient technique for evaluating abnormal gait, can be performed using non-wearable or wearable systems. Non-wearable systems involve image processing and floor sensors, whereas wearable systems use sensors attached to the body7, enabling analysis to be conducted outside the laboratory. Traditional methods of gait analysis involve using force platforms and 3D motion capture technology8,9. However, they involve challenges such as complex equipment, high operating costs, and the need for skilled operation10,11. Further, these systems are limited to laboratory settings, making them impractical for daily activity monitoring. Wearable sensors, particularly inertial measurement units (IMUs) comprising accelerometers, gyroscopes, and magnetometers, offer a portable, cost-effective, and user-friendly alternative for real-world gait analysis12,13. However, their performance is directionally dependent; while inertial sensors demonstrate acceptable accuracy when determining primary sagittal plane joint angles, they exhibit reduced precision when capturing secondary joint variations in the frontal and transverse planes14. To address these multi-planar tracking demands in daily healthcare applications, researchers are continuously developing more affordable wearable frameworks15,16. Rahman et al. recommended tracking localised joint angles and motion during exercise to enhance treatment efficacy17, while Abu Ilius Faisal et al. devised a low-cost gait analyser relying entirely on inertial tracking18. Magnetometers, accelerometers, and gyroscopes were employed by researchers to investigate gait11,19,20. Beyond kinematic data, measuring kinetic factors like ground reaction force (GRF) was critical, prompting the widespread integration of force-sensing resistors (FSR), piezoelectric, capacitance, or piezo-resistive technologies into instrumented insoles21. Some low-cost platforms successfully blended these methods; Ming Gui Tan et al. developed a multi-sensor system utilising a combined accelerometer, FSR insoles, and EMG electrodes to capture locomotion metrics22. Furthermore, Agostini et al. successfully validated a wearable magneto-inertial system (H-Gait) specifically to compare normal weight and overweight/obese cohorts, which revealed definitive group differences such as altered cadence and variations in hip joint range of motion (ROM)23. Despite these advancements, a significant research gap remains. Many existing low-cost systems capture either kinematics (joint angles via IMUs) or kinetics (GRF via insoles), but rarely simultaneously in a fully integrated, affordable package. While wearable systems have been tested on healthy populations, there is limited research deploying combined kinetic-kinematic low-cost wearables specifically to distinguish the unique biomechanical gait characteristics of OW/OB participants outside of clinical environments. Building upon prior validation of the system’s joint angle tracking accuracy (the SMART F-IMU24), this study aimed to evaluate the lower-limb biomechanical adaptations of OW/OB participants during steady-state walking compared to NW participants. This pilot study is proof-of-concept of the SMART F-IMU—an affordable, portable gait assessment system integrating modified inertial measurement units (IMUs) and FlexiForce sensors. Designed to overcome the constraints of traditional laboratory environments, the SMART F-IMU measured both kinematics and kinetics, capturing ground reaction forces (GRF), joint angles, and spatiotemporal characteristics during daily activities. This research establishes a practical framework for continuous, real-world joint-loading monitoring, ultimately supporting the longitudinal clinical management of musculoskeletal conditions. Method Building upon the technical framework established in our previous study24, the SMART F-IMU system integrates seven GY-85 (Shenzhen Gintech Electronic Trade Co., Limited, China) nine-degree-of-freedom inertial measurement units with customised FlexiForce sensors (models A401 and A502) to provide a comprehensive, affordable gait analysis tool. The hardware architecture centres on ESP32 microcontrollers that sample data at 50 Hz, utilising an internal Analogue-to-Digital Converter (ADC) to translate force-sensitive voltage signals into digital values via a pre-established calibration equation established by testing the sensors with a DS2-220 digital force gauge (IMADA, Inc., Japan). In this study, kinetic validation involved comparing phase-based vertical ground reaction force (vGRF) data from the SMART F-IMU against the MatScan system (Tekscan, Inc., United States) across three discrete windows: loading response, mid-stance, and terminal stance. While full continuous dynamic waveform validation was established previously. The kinematic accuracy of the joint angle measurements was previously validated against the Motive-OptiTrack camera system (NaturalPoint, Inc., United States). To ensure kinematic stability and mitigate integration drift during movement, an Extended Kalman Filter (EKF) was implemented via the TinyEKF library framework in Arduino (https://github.com/simondlevy/TinyEKF) to fuse the tri-axial raw streams from the GY-85 IMU’s accelerometer, gyroscope, and magnetometer. The filter executes a recursive cyclic loop running at 50 Hz (∆t = 0.02 s) to project the system state forward (time update) and correct it using an observation model (measurement update). In this specific gait measurement application, the state vector is explicitly defined by a four-element unit quaternion \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:x={[{q}_{0},\:{q}_{1},\:{q}_{2},\:{q}_{3}\:]}^{T}\:$$\end{document}to track segment orientation free of gimbal lock. The state dynamics are driven directly by the gyroscope’s tri-axial angular velocity (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\omega\:={[{\omega\:}_{x},\:{\omega\:}_{y},\:{\omega\:}_{z}]}^{T}$$\end{document}) converted to radians per second. In the time update phase, the filter computes a priori estimate by projecting the previous state (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k-1}$$\end{document}) and the error covariance matrix (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k-1}$$\end{document}) forward using a discrete-time state transition matrix (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\phi}_{k}$$\end{document}) derived from a first-order Taylor series expansion of the quaternion derivative:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}^{-}=\:{\phi}_{k}{x}_{k-1}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}^{-}={\phi}_{k}{P}_{k-1}{\phi}_{k}^{T}+Q$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\phi}_{k}={I}_{4\times\:4}+\:\frac{{\Delta\:}t}{2}\:\left[\begin{array}{cc}\begin{array}{cc}0&\:-{\omega\:}_{x}\\\:{\omega\:}_{x}&\:0\end{array}&\:\begin{array}{cc}-{\omega\:}_{y}&\:-{\omega\:}_{z}\\\:{\omega\:}_{z}&\:-{\omega\:}_{y}\end{array}\\\:\begin{array}{cc}{\omega\:}_{y}&\:-{\omega\:}_{z}\\\:{\omega\:}_{z}&\:{\omega\:}_{y}\end{array}&\:\begin{array}{cc}0&\:{\omega\:}_{x}\\\:-{\omega\:}_{x}&\:0\end{array}\end{array}\right]$$\end{document} The process noise covariance matrix (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:Q\:\epsilon\mathbb{\:}\mathbb{R}^{4\times4}$$\end{document}) is a diagonal matrix initialised to 10− 4 rad2/s2 across its components to manage gyroscope bias instability over time. Feedback is supplied during the measurement update phase by introducing the actual 6 × 1 measurement vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{z}_{k}={[{a}_{x},\:{a}_{y},\:{a}_{z},{m}_{x},\:{m}_{y},\:{m}_{z}\:]}^{T}$$\end{document}, containing tri-axial accelerometer data (scaled to g) and magnetometer fields. Because the observation function h(x) mapping the quaternion state to gravitational and magnetic references is highly non-linear, the filter derives a 6 × 4 linear Jacobian matrix (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{H}_{k}$$\end{document}) based on partial derivatives evaluated at the predicted state:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\left.{H}_{k}=\:\frac{\partial\:h\left(x\right)}{\partial\:x}\right|}_{{x}_{k}^{-}}$$\end{document} This phase executes three sequential steps to compute the final orientation: 1. Compute the Kalman gain (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{K}_{k}$$\end{document}):\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{{K}_{k}=\:{P}_{k}^{-}{H}_{k}^{T}({H}_{k}{P}_{k}^{-}{H}_{k}^{T}+{R}_{k})}^{-1}$$\end{document} 2. Update the state estimate with actual measurement (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{z}_{k}$$\end{document}) to generate a posteriori estimate (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k}$$\end{document}):\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{x}_{k\:}=\:{x}_{k}^{-}+\:{K}_{k}({z}_{k}-h({x}_{k}^{-}\left)\right)$$\end{document} 3. Update the error covariance matrix to generate a posteriori estimate (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}$$\end{document}):\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{P}_{k}=(I-{K}_{k}{H}_{k}){P}_{k}^{-}$$\end{document} The measurement noise covariance matrix (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:R\:\epsilon \mathbb{\:} \mathbb {R}^{6\times6}$$\end{document}) is structured as a block diagonal matrix, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{(R}_{k}\:=\:diag({R}_{accel,\:k}\:\cdot\:{\:I}_{3\:\times\:3},\:{R}_{mag}\cdot\:{\:I}_{3\:\times\:3}))$$\end{document}, and is dynamically scaled to handle movement artefacts. Under normal tracking conditions where the calculated acceleration vector norm remains within a strict 10% deviation of the earth’s gravity baseline. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(0.9g\le\:\sqrt{{a}_{x}^{2}+{a}_{y}^{2}+{a}_{z}^{2}}\le\:1.1g)$$\end{document}, base parameters are assigned as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{accel,\:k\:}=\:{10}^{-2}\:\:$$\end{document}and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{mag\:}=\:{10}^{-1}\:$$\end{document}to prioritise gravity-based orientation corrections while mitigating magnetic anomalies. Conversely, when aggressive movements or impact spikes disrupt the acceleration norm baseline \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:(\left|\sqrt{{a}_{x}^{2}+{a}_{y}^{2}+{a}_{z}^{2}}-1g\right|>0.1)$$\end{document}, the system automatically penalises the accelerometer readings by scaling \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{accel,\:k\:}$$\end{document} up four orders of magnitude to a flat value of 102 while keeping \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{R}_{mag\:}$$\end{document}fixed at 0.1. This expansion heavily attenuates the innovation step, driving the accelerometer’s Kalman gain toward zero and forcing the filter to rely entirely on gyroscope kinematic predictions during impacts. To protect against numerical rounding errors accumulating over floating-point loop iterations, a strict quaternion normalisation step is performed at the tail end of each recursive cycle to preserve a unit norm:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{norm}_{q}=\sqrt{{q}_{0}^{2}+{q}_{1}^{2}+{q}_{2}^{2}+{q}_{3}^{2}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{q}_{i}=\frac{{q}_{i}}{{norm}_{q}}$$\end{document} Finally, long-term yaw drift from environmental magnetic fields is resolved by applying a local magnetic declination angle correction directly to the final yaw angle during its extraction from the stable quaternion state. This comprehensive sequence effectively isolates noise contamination to output smooth, repeatable joint angles for downstream gait tracking. Before each trial, a calibration was performed: participants maintain a static, neutral standing posture for five seconds. This period allowed the system to establish a zero-drift reference and align the sensor axes with the anatomical axes of the lower limbs. To accurately capture lower limb kinematics, the seven IMUs were strategically orientated vertically within their respective circuit boxes and positioned on specific body segments: the posterior lumbar (pelvis), thighs, shanks, and dorsi-feet. Assumed alignment method was utilized—a simply and frequently employed protocol where the researcher visually matched the IMU sensing axes directly with the underlying anatomical axes of each segment during placement25. Relative joint angles for the hip, knee, and ankle were then derived from the absolute orientations of adjacent segments through a hierarchical kinematic chain. The knee angle was derived from the relative orientation of the thigh and shank segments. The hip angle was derived from the lumbar and thigh segments. The ankle angle was derived from the shank and instep/shoe segments. Anatomical joint angles were extracted via a standard Euler angle rotation sequence, mapping the relative orientations to adduction/abduction (X-axis), flexion/extension (Y-axis), and internal/external rotation (Z-axis). The participants were divided into two groups: 10 participants with NW and 10 who were OW/OB participants. The rationale for the sample size (N = 20) was determined based on the study’s primary objective as a pilot evaluation and proof-of-concept for the SMART F-IMU system. This cohort size was consistent with established preliminary wearable sensor testing protocols designed to assess concurrent validity against reference systems. Furthermore, the study served an exploratory purpose to identify potential biomechanical deviations between BMI groups. This pilot approach was intended to establish the system’s feasibility in a controlled setting and to provide the necessary mean and standard deviation data required for sample size calculations for clinical trials in the future. Selection criteria for the OW/OB group required a BMI > 25 kg/m², while the NW group required a BMI between 18.5 and 24.9 kg/m², following WHO guidelines. Participants were included if they were aged 18–45 and able to walk independently without assistive devices. This age range was selected to ensure a stable proof-of-concept by minimising the presence of age-related comorbidities that could confound initial biomechanical assessments. Exclusion criteria included any history of major lower-limb surgery, neurological conditions affecting balance, or musculoskeletal injuries within the past six months that could interfere with natural gait patterns. The study protocol was approved by the Ethics Committee of the Mahidol University Institutional Review Board (COA No. MU-CIRB 2021/161.0707). Informed consent, including permission to publish information and images in online publications, was obtained from all participants before the start of the study. Table 1 presents the demographic data of the participants, including age, gender, body mass, height, and BMI. Table 1Demographic data of NW and OW/OB participants.CharacteristicsNW (n = 10)OW/OB (n = 10)Age (years)29.1 ± 5.9732.1 ± 7.22Gender (F, M)7 F, 3 M5 F, 5 MBody Mass (kg)62.08 ± 10.6685.22 ± 12.34Height (cm)163.9 ± 9.02164.95 ± 8.63Body Mass Index (kg/m²)22.96 ± 2.0731.25 ± 3.42 Table 1Demographic data of NW and OW/OB participants.CharacteristicsNW (n = 10)OW/OB (n = 10)Age (years)29.1 ± 5.9732.1 ± 7.22Gender (F, M)7 F, 3 M5 F, 5 MBody Mass (kg)62.08 ± 10.6685.22 ± 12.34Height (cm)163.9 ± 9.02164.95 ± 8.63Body Mass Index (kg/m²)22.96 ± 2.0731.25 ± 3.42 Demographic data of NW and OW/OB participants. To perform the experiment, participants were instructed to wear the SMART F-IMU system, which consisted of seven motion sensors and shoes containing seven force sensors in each shoe. The SMART F-IMU system was designed for versatile applications, allowing sensors to be attached both directly to the skin and over tight-fitting clothing to balance data accuracy with user comfort. In this study, five motion sensors were positioned at the posterior trunk, right and left lateral thighs, and lateral shanks to set up the device. Two more motion sensors on the insteps were integrated with shoe force sensors. To minimise movement relative to the skin and mitigate soft-tissue artefacts, participants were required to wear tight-fitting clothing. To ensure stability and prevent the sensors from shifting during dynamic movement, they were secured by elastic rubber bands. This setup was specifically designed to minimise vibration noise and prevent sensor shifting during dynamic movement. Figure 1 shows the sensor locations placed on the lower limbs. This setup minimised zero drift and vibration noise by maintaining a rigid connection between the sensor and the body segment throughout the gait cycle. Fig. 1The sensor locations placed on the lower limbs. The sensor locations placed on the lower limbs. Statistical analysis To validate the force measurement accuracy of the SMART F-IMU, the agreement between the developed system and the MatScan system was assessed using the Intraclass Correlation Coefficient (ICC). Additionally, a Bland-Altman analysis was performed to determine the mean bias and the 95% limits of agreement (LoA), calculated as the mean difference ± 1.96 times the standard deviation of the differences. To further quantify device precision across different weight categories, absolute error (N) and percentage error (%) were calculated for standing ground reaction force (GRF). Collectively, these metrics were used to evaluate the stability and reliability of the modified FlexiForce sensors in comparison to established commercial technology. For each participant, 10 consecutive stable gait cycles were selected for analysis, excluding the initial acceleration and final deceleration strides. Data were analysed using SPSS. The normality of the data distribution was assessed using the Shapiro-Wilk test. To compare the spatiotemporal gait parameters, joint angles, and GRF between the NW and OW/OB groups, an independent t-test (for normally distributed data) or a Mann-Whitney U test (for non-normally distributed data) was performed. All results were reported as mean ± standard deviation (SD). Due to the exploratory nature of this pilot study and many gait parameters analysed, the interpretation of results prioritised effect size over absolute p-value thresholds to mitigate the risk of Type I error inflation. This approach ensured that the discussion focuses on the most biologically and scientifically meaningful differences between the cohorts. Statistical significance was set at p < 0.05 (indicated by a single asterisk *) and p < 0.01 (indicated by double asterisks **). Effect sizes were calculated for all joint angle and GRF parameters to quantify the magnitude of differences between the NW and OW/OB groups. Cohen’s d was used for parametric data, with values interpreted as small (0.2), medium (0.5), or large (0.8 and above). For non-normally distributed data, the effect size r was calculated. By emphasising these effect magnitudes, the analysis provided an identification of biomechanical trends regardless of the number of comparisons performed. Experiment and results The experiment commenced with the participant being instructed to maintain a standing position for five seconds on the MatScan. Subsequently, they walked at a normal velocity for two minutes. To ensure the validity and reliability of the gait parameters, participants performed the walking trial for two minutes. They were instructed to maintain a self-selected normal walking velocity to capture their natural steady-state gait. After two minutes, the participant ended their walking and stood still for another five seconds. For the dynamic gait evaluation, data from both the left and right limbs were recorded and analysed. Based on the assumption of bilateral gait symmetry in NW and OW/OB participants without known orthopaedic injuries or asymmetries, the parameters from both sides were averaged to provide a comprehensive representation of each participant’s gait. To ensure that the analysis captured only steady-state walking, a strict buffer zone was implemented: the first and last 10 s of each recording were excluded to eliminate artefacts related to gait initiation (acceleration) and termination (deceleration). Within the stabilised mid-section of the trials, 10 consecutive gait cycles per limb were identified using a force-threshold algorithm triggered by initial contact events. All participants repeated the experiment twice to ensure data consistency. The data obtained from the experiment were examined to determine the gait deviation parameters and compare them between the two groups of participants. The GRF results of standing still, compared between the weight scale, MatScan, and SMART F-IMU for ten NW participants, are presented in Table 2. Figure 2 illustrates a comparison of the GRF between the participants’ weight in newton units, the MatScan system, and the total GRF from the SMART F-IMU system in NW participants. Similarly, Table 3 shows the GRF results for ten OW/OB participants, compared between the weight scale, MatScan, and SMART F-IMU. Figure 3 provides a comparison of the GRF between the participants’ weight in newton units, total GRF from the MatScan system, and the total GRF from the SMART F-IMU system in OW/OB participants. The statistical reporting for Tables 2 and 3 involved comparing the SMART F-IMU and MatScan outputs against the participants’ actual weight in newtons. Force measurements were normalised by body weight to facilitate a direct comparison across participants with varying masses. Table 2GRF compared between MatScan and SMART F-IMU in ten NW participants.NW participantsWeight (N)MatScan (N)SMART F-IMU(N)MatScan normalised by weightSMART F-IMU normalised by weight1619.36552.55443.600.890.722509.60405.85421.660.800.833604.66574.66520.490.950.864452.76429.71343.370.950.765755.58735.22720.810.970.956664.44692.27522.351.040.797545.86474.77416.410.870.768637.00657.46482.461.030.769529.20454.64424.580.860.8010777.14698.61632.400.900.81 Table 2GRF compared between MatScan and SMART F-IMU in ten NW participants.NW participantsWeight (N)MatScan (N)SMART F-IMU(N)MatScan normalised by weightSMART F-IMU normalised by weight1619.36552.55443.600.890.722509.60405.85421.660.800.833604.66574.66520.490.950.864452.76429.71343.370.950.765755.58735.22720.810.970.956664.44692.27522.351.040.797545.86474.77416.410.870.768637.00657.46482.461.030.769529.20454.64424.580.860.8010777.14698.61632.400.900.81 GRF compared between MatScan and SMART F-IMU in ten NW participants. Fig. 2Comparison of the GRF between the participants’ weight in newton units, MatScan system, and the total GRF from the SMART F-IMU system in NW participants. Comparison of the GRF between the participants’ weight in newton units, MatScan system, and the total GRF from the SMART F-IMU system in NW participants. Table 3GRF compared between MatScan and SMART F-IMU in ten OW/OB participants.OW/OB participantsWeight(N)MatScan(N)SMART F-IMU(N)MatScan normalised by weightSMART F-IMU normalised by weight1914.341065.71938.841.171.032935.901026.421023.241.101.093830.06830.89791.271.000.954829.08884.98786.791.070.955921.201062.77960.041.151.046723.24714.97686.350.990.957686735.00575.791.070.848619.36552.55542.750.890.889976.081024.91941.341.050.9610916.30891.53814.710.970.89 Table 3GRF compared between MatScan and SMART F-IMU in ten OW/OB participants.OW/OB participantsWeight(N)MatScan(N)SMART F-IMU(N)MatScan normalised by weightSMART F-IMU normalised by weight1914.341065.71938.841.171.032935.901026.421023.241.101.093830.06830.89791.271.000.954829.08884.98786.791.070.955921.201062.77960.041.151.046723.24714.97686.350.990.957686735.00575.791.070.848619.36552.55542.750.890.889976.081024.91941.341.050.9610916.30891.53814.710.970.89 GRF compared between MatScan and SMART F-IMU in ten OW/OB participants. Fig. 3Comparison of GRF between the participants’ weight in newton units, total GRF from the MatScan system, and total GRF from the SMART F-IMU system in OW/OB participants. Comparison of GRF between the participants’ weight in newton units, total GRF from the MatScan system, and total GRF from the SMART F-IMU system in OW/OB participants. The validation analysis for standing ground reaction force (GRF) demonstrated agreement between the SMART F-IMU and MatScan systems across both participant groups. The comparison yielded an ICC of 0.97, indicating excellent reliability for this wearable system. Although the mean absolute error was 75.39 N (10.83 relative error), this variance was primarily attributed to the pilot nature of the sensor integration within the shoe insole. Furthermore, the Bland-Altman analysis confirmed a high level of agreement for both groups; for NW participants, the mean bias was negligible, with the majority of data points falling within the 95% limits of agreement. These results suggested that the modified FlexiForce sensors, when secured with the rigid fixation protocol, provided stable and accurate kinetic data suitable for proof-of-concept gait assessment. Collectively, these validation statistics confirm that the SMART F-IMU system provided reliable force measurements for biomechanical gait assessment across different weight categories. While full dynamic waveform validation was established in the previous work24, the kinetic evaluation in this study focused specifically on a phase-based analysis of vertical GRF. Normalised vertical GRF parameters were statistically compared between the normal weight and overweight/obese cohorts across three key phases of the gait cycle: load response, mid-stance, and terminal stance. The reported biomechanical parameters were derived by selecting 10 representative gait cycles from each two-minute steady-state walking trial to ensure data consistency. While participants walk around 70–110 cycles per trial, analysis of the selected cycles confirmed that the parameters remained highly consistent with low standard deviation throughout the trial. All participants repeated the experiment twice to verify the stability and consistency of the recorded kinematic and kinetic data. Regarding the distribution of these data, normality was assessed using the Shapiro-Wilk test. It was found that nearly all spatiotemporal, joint-angle, and kinetic parameters followed a normal distribution. The only exceptions requiring non-parametric analysis were the knee maximum flexion angle during load response, the hip maximum flexion angle during load response, and the vertical GRF during both the load response and mid-stance phases. Consequently, independent t-tests were employed for the normally distributed variables, while the Mann-Whitney U test was used for those parameters that deviated from normality. Table 4The spatiotemporal gait parameters compared between the two groups of participants.Gait ParametersNWOW/OBp-valueEffect SizeMeanSDMeanSDStride time (s)1.210.111.460.220.0001**1.44Stance time (s)0.780.080.970.150.0001**1.57Swing time (s)0.430.040.490.070.0006**1.05Double support time (s)0.470.080.690.150.0001**1.83Single support time (s)0.310.120.270.010.15970.47Cycle time (s)1.220.101.660.420.0001**1.44Cadence (steps/min)97.9010.2577.1014.200.0001**1.68Stance phase (%)64.771.5166.411.250.0038**1.18Swing phase (%)35.231.5133.591.250.0038**1.18 Table 4The spatiotemporal gait parameters compared between the two groups of participants.Gait ParametersNWOW/OBp-valueEffect SizeMeanSDMeanSDStride time (s)1.210.111.460.220.0001**1.44Stance time (s)0.780.080.970.150.0001**1.57Swing time (s)0.430.040.490.070.0006**1.05Double support time (s)0.470.080.690.150.0001**1.83Single support time (s)0.310.120.270.010.15970.47Cycle time (s)1.220.101.660.420.0001**1.44Cadence (steps/min)97.9010.2577.1014.200.0001**1.68Stance phase (%)64.771.5166.411.250.0038**1.18Swing phase (%)35.231.5133.591.250.0038**1.18 The spatiotemporal gait parameters compared between the two groups of participants. Table 5Hip, knee, and ankle angles and GRF in different phases of the gait of 10 NW participants compared to 10 OW/OB participants.Gait Parameters(Joint Angle and Ground Reaction Force)NW(n = 10)OW/OB(n = 10)p-valueEffect SizeMean ± SD (degrees)Mean ± SD (degrees)Hip flexion angle at initial contact16.69 ± 4.4021.29 ± 6.100.0046**0.86Hip maximum flexion angle in load response10.48 ± 4.7113.00 ± 7.760.28990.24Hip maximum extension angle in the stance phase−19.47 ± 6.98−24.66 ± 7.000.0216*0.74Hip extension angle at toe-off−21.91 ± 9.61−26.86 ± 12.370.06910.45Hip maximum flexion angle in the swing phase25.50 ± 8.6626.20 ± 10.500.40270.07Knee flexion angle at initial contact3.28 ± 2.774.84 ± 2.740.05410.57Knee maximum flexion angle during load response21.84 ± 6.5216.59 ± 5.090.0157*0.57Knee maximum extension angle in the stance phase−6.00 ± 3.46−2.80 ± 3.970.0085**0.86Knee flexion angle at toe-off37.92 ± 9.6322.33 ± 6.740.0004**1.88Knee maximum flexion angle in the swing phase59.30 ± 13.3440.67 ± 8.890.0009**1.65Ankle dorsiflexion angle at initial contact2.35 ± 3.003.67 ± 3.100.09900.43Ankle maximum plantarflexion angle in load response−7.97 ± 3.40−10.93 ± 4.630.0112*0.73Ankle maximum dorsiflexion angle in mid-stance8.67 ± 8.505.06 ± 6.080.10610.49Ankle angle at the terminal stance phase−1.83 ± 3.69−1.48 ± 3.830.38590.09Ankle maximum plantarflexion angle in the pre-swing phase−20.25 ± 5.69−26.36 ± 5.300.0040**1.11Ankle maximum dorsiflexion angle in the swing phase6.79 ± 5.878.80 ± 8.400.15350.28Hip adduction angle in the terminal stance4.26 ± 3.277.75 ± 3.610.0041**1.01Hip abduction angle in the terminal stance−3.03 ± 2.88−5.51 ± 4.000.0117*0.71Hip adduction angle in pre-swing2.83 ± 3.695.81 ± 4.440.0154*0.73Hip abduction angle in pre-swing−3.16 ± 3.05−5.37 ± 3.400.0238*0.68Knee maximum adduction angle in the stance phase3.23 ± 3.537.73 ± 5.710.0015**0.95Knee maximum abduction angle in the stance phase−2.25 ± 2.53−3.02 ± 3.560.17980.25Knee maximum adduction angle in the swing phase3.32 ± 4.4912.52 ± 8.560.0001**1.34Knee maximum abduction angle in the swing phase−3.45 ± 3.22−4.11 ± 4.090.26830.18Ankle maximum eversion angle2.67 ± 4.284.29 ± 3.430.13040.42Ankle maximum inversion angle−3.29 ± 3.89−4.92 ± 5.400.10910.35Vertical GRF in load response (normalised by weight in newton units)1.57 ± 0.441.25 ± 0.250.0082**0.59Vertical GRF in mid-stance (normalised by weight in newton units)0.93 ± 0.170.86 ± 0.100.0412*0.46Vertical GRF in terminal stance (normalised by weight in newton units)1.80 ± 0.261.17 ± 0.100.0173*1.74Note: For sagittal plane kinematics, positive values indicate flexion (or dorsiflexion at the ankle), while negative values indicate extension (or plantarflexion). For frontal plane kinematics, positive values denote adduction (or eversion at the ankle) and negative values denote abduction (or inversion at the ankle). Table 5Hip, knee, and ankle angles and GRF in different phases of the gait of 10 NW participants compared to 10 OW/OB participants.Gait Parameters(Joint Angle and Ground Reaction Force)NW(n = 10)OW/OB(n = 10)p-valueEffect SizeMean ± SD (degrees)Mean ± SD (degrees)Hip flexion angle at initial contact16.69 ± 4.4021.29 ± 6.100.0046**0.86Hip maximum flexion angle in load response10.48 ± 4.7113.00 ± 7.760.28990.24Hip maximum extension angle in the stance phase−19.47 ± 6.98−24.66 ± 7.000.0216*0.74Hip extension angle at toe-off−21.91 ± 9.61−26.86 ± 12.370.06910.45Hip maximum flexion angle in the swing phase25.50 ± 8.6626.20 ± 10.500.40270.07Knee flexion angle at initial contact3.28 ± 2.774.84 ± 2.740.05410.57Knee maximum flexion angle during load response21.84 ± 6.5216.59 ± 5.090.0157*0.57Knee maximum extension angle in the stance phase−6.00 ± 3.46−2.80 ± 3.970.0085**0.86Knee flexion angle at toe-off37.92 ± 9.6322.33 ± 6.740.0004**1.88Knee maximum flexion angle in the swing phase59.30 ± 13.3440.67 ± 8.890.0009**1.65Ankle dorsiflexion angle at initial contact2.35 ± 3.003.67 ± 3.100.09900.43Ankle maximum plantarflexion angle in load response−7.97 ± 3.40−10.93 ± 4.630.0112*0.73Ankle maximum dorsiflexion angle in mid-stance8.67 ± 8.505.06 ± 6.080.10610.49Ankle angle at the terminal stance phase−1.83 ± 3.69−1.48 ± 3.830.38590.09Ankle maximum plantarflexion angle in the pre-swing phase−20.25 ± 5.69−26.36 ± 5.300.0040**1.11Ankle maximum dorsiflexion angle in the swing phase6.79 ± 5.878.80 ± 8.400.15350.28Hip adduction angle in the terminal stance4.26 ± 3.277.75 ± 3.610.0041**1.01Hip abduction angle in the terminal stance−3.03 ± 2.88−5.51 ± 4.000.0117*0.71Hip adduction angle in pre-swing2.83 ± 3.695.81 ± 4.440.0154*0.73Hip abduction angle in pre-swing−3.16 ± 3.05−5.37 ± 3.400.0238*0.68Knee maximum adduction angle in the stance phase3.23 ± 3.537.73 ± 5.710.0015**0.95Knee maximum abduction angle in the stance phase−2.25 ± 2.53−3.02 ± 3.560.17980.25Knee maximum adduction angle in the swing phase3.32 ± 4.4912.52 ± 8.560.0001**1.34Knee maximum abduction angle in the swing phase−3.45 ± 3.22−4.11 ± 4.090.26830.18Ankle maximum eversion angle2.67 ± 4.284.29 ± 3.430.13040.42Ankle maximum inversion angle−3.29 ± 3.89−4.92 ± 5.400.10910.35Vertical GRF in load response (normalised by weight in newton units)1.57 ± 0.441.25 ± 0.250.0082**0.59Vertical GRF in mid-stance (normalised by weight in newton units)0.93 ± 0.170.86 ± 0.100.0412*0.46Vertical GRF in terminal stance (normalised by weight in newton units)1.80 ± 0.261.17 ± 0.100.0173*1.74Note: For sagittal plane kinematics, positive values indicate flexion (or dorsiflexion at the ankle), while negative values indicate extension (or plantarflexion). For frontal plane kinematics, positive values denote adduction (or eversion at the ankle) and negative values denote abduction (or inversion at the ankle). Hip, knee, and ankle angles and GRF in different phases of the gait of 10 NW participants compared to 10 OW/OB participants. Note: For sagittal plane kinematics, positive values indicate flexion (or dorsiflexion at the ankle), while negative values indicate extension (or plantarflexion). For frontal plane kinematics, positive values denote adduction (or eversion at the ankle) and negative values denote abduction (or inversion at the ankle). Fig. 4Hip, knee, and ankle angles and GRF of the gait of 10 NW participants compared to 10 OW/OB participants. Hip, knee, and ankle angles and GRF of the gait of 10 NW participants compared to 10 OW/OB participants. Discussion The rising prevalence of musculoskeletal disorders underscores the need to detect early biomechanical alterations before irreversible joint degeneration occurs. While compensatory gait patterns are well-documented in clinical populations, a critical gap remains in monitoring how healthy OW/OB individuals experience altered gait patterns during daily activities. Portable technologies are needed to capture these real-world kinetic and kinematic shifts. Therefore, the primary purpose of this study was to evaluate the utility of the wearable SMART F-IMU system in identifying gait adaptations of OW/OB participants during steady-state walking compared to NW participants. The dynamic testing of motion and force sensors involved 20 participants, divided into 10 NW and 10 OW/OB participants, to identify significant differences in gait parameters. To ensure data stability, the reported parameters were derived from 10 representative gait cycles selected from each two-minute trial. Based on the recorded cadence, NW participants completed an average of 98 volumes of recorded steps, analysis confirmed that the parameters remained highly consistent with low standard deviation throughout the steady-state period. The results (Table 4) revealed that OW/OB participants had significantly longer stride time, stance time, swing time, double support time, and cycle time than NW participants (p-value < 0.01). These differences were further supported by large effect sizes (ranging from 1.05 to 1.83), indicating a substantial impact of body weight on these metrics beyond simple p-value thresholds. The result showed that OW/OB participants walk slower and spend more time in the stance phase. Additionally, OW/OB participants had decreased cadence and percentage of swing phase. The only parameter without a significant difference was single support time (p = 0.1597, effect size = 0.47). Overall, the findings suggest that gait parameters are perceptibly affected by weight, with OW/OB participants exhibiting distinct patterns compared to NW participants. Due to the exploratory nature of this pilot study, the analysis of joint angles and kinetics (Table 5; Fig. 4) focused on the magnitude of differences (effect size) to mitigate the risk of Type I error inflation from multiple comparisons. The analysis of joint angle and kinetic data during each gait phase (Table 5; Fig. 4) revealed statistical significance in several parameters. Specifically, the hip maximum extension angle in the stance phase, knee maximum flexion angle during the load response, ankle maximum plantar flexion angle in the load response, hip abduction angle in both terminal stance and pre-swing, hip adduction angle in pre-swing, and vertical GRF in mid-stance and terminal stance exhibited significant differences (p < 0.05). These parameters generally displayed medium to large effect sizes, such as a large effect size (d = 0.90) for knee maximum flexion during load response, indicating substantial biomechanical shifts. Further, certain parameters exhibited statistical significance (p-value < 0.01), including the hip angle at initial contact, knee maximum extension angle in the stance phase, knee angle at toe-off, knee maximum flexion angle in the swing phase, ankle maximum plantar flexion angle in the pre-swing phase, hip adduction angle in terminal stance, knee maximum adduction angle in both the stance and swing phases, and vertical GRF in load response. The magnitude of these differences was underscored by large effect sizes, notably for knee angle at toe-off (d = 1.88) and vertical GRF in terminal stance (d = 1.74), which represented the most significant kinetic deviation between the two groups. However, other parameters, such as the hip maximum flexion angle in load response, hip angle at toe-off, hip maximum angle in the swing phase, knee angle at initial contact, ankle angle at initial contact, ankle maximum dorsiflexion angle in mid-stance, ankle angle at terminal stance, ankle maximum dorsiflexion angle in the swing phase, knee maximum abduction angle in the stance and swing phases, and ankle maximum eversion and inversion angles, showed no significant difference (p-value > 0.05). These parameters typically showed small or negligible effect sizes, confirming that weight had a minimal impact on these specific movements in this pilot evaluation. A comparison of kinetic data revealed that the OW/OB group exhibited significantly lower normalised vertical GRF peaks, particularly during terminal stance, compared to the NW group. While absolute force was higher in heavier participants, the lower normalised values (N/BW) observed here were indicative of a cautious gait strategy. This strategy was characterised by the significantly decreased cadence and slower walking speeds reported in Table 4. By walking more slowly and increasing double-support time, OW/OB participants effectively reduced the vertical acceleration of their centre of mass, leading to a reduction in the peak impact forces relative to their total body weight. This biomechanical adjustment served as a compensatory mechanism to enhance stability associated with increased body mass. Consequently, the SMART F-IMU could capture these subtle biomechanic shifts, demonstrating its utility in detecting the protective gait adaptations often seen in populations at risk for osteoarthritis. Existing literatures have indicated that some gait characteristics observed in OW/OB populations share cross-sectional biomechanical similarities with patterns reported in knee OA cohorts26–30. Studies have noted that participants with knee OA frequently exhibit a decreased range of motion or reduced knee flexion during the swing and stance phases31,32, alongside a generally slower walking speed33,34. While these overlapping trends are notable, it is essential to emphasise that the gait adaptations isolated in this healthy OW/OB cohort are not equivalent to prodromal pathological markers. Because this study did not include patients diagnosed with clinical osteoarthritis or evaluate clinical diagnostic outcomes, any utility of the standalone platform for clinical OA screening, risk identification, or active disease monitoring remains strictly hypothetical. Instead, these observed deviations should be viewed as preliminary indicators of altered mechanical joint loading. Longitudinal tracking of these relative kinetic and kinematic shifts using a portable framework may assist in monitoring the overall mechanical stress associated with weight-induced joint loading over time, highlighting areas for future targeted clinical investigations. Despite the significant findings, several limitations must be acknowledged. This research served as a pilot evaluation and proof-of-concept for the SMART F-IMU system. The sample size of 10 participants per group was targeted for initial validation, which limited the broad generalisability of the findings across the entire OW/OB population. The study did not control walking speed, a known confounder for both kinetics and kinematics; the slower speeds observed in the OW/OB group likely influenced the lower normalised GRF peaks. The analysis was restricted to two minutes of steady-state walking on level ground, which might not fully capture the complexity of gait during varied daily activities. Furthermore, the study did not include clinical populations, such as patients with symptomatic osteoarthritis, which limited the direct application of these findings to pathological gait management. Technical challenges such as soft-tissue artefact—where sensors move relative to the bone due to adipose tissue oscillation—may have influenced kinematic accuracy, particularly in the OW/OB cohort. Additionally, the merging of OW/OB categories into a single group and the potential for gender-based gait differences may have masked specific variations unique to these subgroups. Regarding statistical analysis, the large number of parameters tested was addressed by prioritising effect size interpretation rather than unadjusted p-values to mitigate Type I error inflation. A comparison of kinetic data revealed specific measurement discrepancies: the SMART F-IMU tended to underestimate force in some NW participants, while the MatScan system recorded values slightly exceeding body weight in the OW/OB group. The underestimation in the wearable system is likely due to load distribution across insole regions not covered by the discrete FlexiForce sensors. Conversely, the overestimation in the MatScan system may be attributed to high-pressure contact characteristics of heavier participants, which can influence the capacitive response of pressure mats. Despite these absolute variances, the strong ICC (0.97) and consistent Bland-Altman limits of agreement suggest that the SMART F-IMU is sufficiently reliable for identifying biomechanical trends and differences between study cohorts. A known limitation of wearable inertial sensors is sensor drift, where the integration of angular velocity and acceleration signals leads to a gradual divergence from the true position or angle over time. In this study, researchers minimised drift by implementing a Kalman filter and resetting the system to a zero-drift state before each trial. However, while these measures were effective for the 2-minute walking duration, the current system may still face precision challenges during continuous and long-term monitoring. To build upon the exploratory findings of this pilot evaluation, future studies will focus on critical technical optimisations, methodological expansions, and broader clinical translations to address current limitations. To isolate the confounding effects of the slower self-selected walking speeds observed in the overweight and obese cohort, upcoming iterations will implement speed-matching protocols or utilise advanced statistical covariates to separate the independent impacts of body mass and velocity. Technically, the integration of zero-velocity update (ZUPT) algorithms and sophisticated sensor fusion techniques will be explored to minimise cumulative integration errors during extended testing. Methodologically, the research will feature expanded cohort sizes determined by pilot effect sizes to ensure robust statistical power, alongside precise participant stratification by age, gender, and specific BMI subgroups (distinguishing between overweight, Class I, and Class II obesity). Furthermore, formal test-retest reliability protocols will be established to quantify the biomechanical impact of sensor removal and re-application. Finally, the research scope will extend to older populations to evaluate how the synergy of increased body mass and age-related muscle weakness influences gait stability and fall risks, while validating the portable platform within real-world, non-laboratory environments for pathological gait management in target populations, such as participants suffering from symptomatic knee osteoarthritis. Conclusion The SMART F-IMU system was successfully employed as a pilot evaluation and proof-of-concept to analyse gait parameters in NW and OW/OB participants. By integrating modified inertial measurement units and flexible force sensors, the system provided a comprehensive assessment of both kinematic and kinetic data in a portable format. The exploratory findings identified distinct between-group differences, including significantly longer stride, stance, and cycle times in the OW/OB cohort, alongside variations in knee and ankle joint angles. While these gait alterations share certain biomechanical characteristics with those observed in clinical populations, this study primarily demonstrated the feasibility of the SMART F-IMU for detecting subtle, weight-induced kinetic and kinematic shifts. By identifying these biomechanical trends early through an affordable, portable platform, the system serves as a potential functional tool for the longitudinal monitoring of mechanical joint loading in non-laboratory environments. Further research with larger cohorts is necessary to determine the long-term clinical implications of these observed gait patterns.