Analyzing gait fatigue with Kinect during prolonged treadmill walking at constant pace

Abstract

Fatigue can lead to costly and dangerous accidents in various settings. By analyzing human physiology in different situations, such as walking, we can objectively detect fatigue. This can help us prevent many accidents by identifying fatigue. Many studies have explored gait analysis for fatigue diagnosis. In this study, we aim to validate novel gait parameters for fatigue detection, employing non-invasive analytical methods on gait data. Gait kinematics data from 10 healthy young adults were collected as they walked at a constant speed of 1 m/s on a treadmill until fatigued. Their walking biomechanics were monitored using a Kinect sensor, with data analysis performed through a custom application developed by the research team. Our analysis found significant alterations in 6 key gait kinematics following the onset and progression of fatigue. Notable changes included a 52% increase in maximum spine angle per step, an 8.2% increase in maximum thigh sagittal angle per step, an 11% increase in maximum knee angle of the non-dominant leg per step, a 16.8% increase in maximum stride height per step, 23% increase in maximum toe out angle per step and 45% increase in maximum angular velocity of knee angle of the non-dominant leg per step. We have discovered six new biomechanical markers that enable us to track fatigue accurately and objectively during walking with minimal interference. These markers offer a dependable approach for monitoring levels of fatigue.

Introduction

Fatigue is a prevalent phenomenon that encompasses both subjective aspects, such as perceived tiredness, exhaustion, and declining vigor, and objective elements, including impairments in cellular, tissue, and organ functions. These impairments are induced by repeated or excessive stressors, stimulation, or physical exertion. ¹ It has been examined in various domains as an everyday construct ² that can also entail economic consequences by impairing efficiency and production.^3,4 Fatigue is a significant risk factor for injuries. A study found that high fatigue levels were associated with a 2.28 times higher risk of occupational accidents, ⁵ Another study indicated that male workers with moderate to high fatigue levels had adjusted odds ratios of 1.76 and 2.61 for hospital treatment due to injuries, ⁶ as result, various methods have been developed to detect and quantify fatigue, among which surface electromyography (sEMG) is one of the most widely used techniques.^7,8 Fatigue can arise in various scenarios even routine activities, such as walking, may be associated with the experience of fatigue, walking is a common daily activity ⁹ and is beneficial to health. ¹⁰ It is now possible to identify many medical conditions based on gait analysis while walking, including Parkinson’s disease (PD)^11,12 multiple sclerosis (MS), ¹³ and joint arthrosis, ¹⁴ specifically in the knee joint.^15,16 Previous studies have shown how fatigue affects human physical condition ¹⁷ with balance impairment being the most common consequence, ¹⁸ Fatigue may induce alterations in gait patterns.^19–22 Furthermore, these prior studies demonstrate that fatigue-induced gait alterations are most pronounced in spatiotemporal parameters such as stride length, gait velocity, and cadence,^23,24 especially when gait dynamics change due to muscle fatigue.^25,26 which impairs muscle coordination and responsiveness.^27–29

Using force plates is one of the most common ways of extracting gait features.^30,31 Barbieri et al. utilized gait analysis to compare the effects of knee muscle fatigue between young and elderly adults. The results showed that while stride length did not change significantly in the young, it increased considerably in the elderly. Stride time decreased in both young and old participants, but it was more pronounced in the older group. Stride speed increased in all age groups, but the increase was more pronounced among older individuals. The crossing step length was found to decrease for all participants. Besides these studies, it has been shown that knee muscle fatigue varies with age. ³² In the other study, Zhang et al. investigated fatigue by analyzing gait features using wearable sensors ³³ which is another prevalent way to measure gait parameters.^34–37 The results indicated that fatigue caused significant changes in the mean acceleration, angular velocity, and ankle rotation range. Moreover, this study corroborated the parameters obtained in previous studies^38,39 and revealed that fatigue increased the horizontal angle of ankle rotation, which impaired balance and caused heel pronation. ³³ Finally, Image processing in the identification of gate features is very convenient^40,41 especially the use of Kinect sensors, which are inexpensive and widely accessible devices make them perfect for the medical applications. ⁴² Aoki et al. implemented a Kinect-based gait analysis system leveraging deep learning techniques to evaluate fatigue levels. Their methodology utilized the supporting limb as input to the neural network model. Given the supporting foot remains stationary during single support phases of the gait cycle, the ankle joint served as the reference landmark. The neural network processed features of this joint to classify fatigue status and progression. ⁴³

Traditional methods for assessing walking fatigue rely on the use of wearable sensors to analyze gait parameters. Although these sensors provide valuable data, they can be intrusive for participants and primarily capture localized changes in movement, which may restrict their ability to accurately reflect whole-body fatigue levels. In this context, image processing, specifically utilizing the Kinect sensor, emerges as a promising alternative. It has the potential to provide a more comprehensive and holistic assessment of walking fatigue by capturing a wider range of relevant features. While previous research has identified various gait alterations following fatigue, most studies rely on discrete pre- and post-test comparisons. ⁴⁴ This “snapshot” approach captures the end state of fatigue but often misses the progressive transition and the specific markers that consistently track its onset. This study differentiates itself by employing continuous, real-time monitoring of gait kinematics throughout the entire fatiguing process. By focusing on parameters that exhibit a steady, monotonic increase—rather than just a final difference—we identify a set of biomechanical markers that are uniquely suited for the real-time detection and tracking of fatigue progression.

This study employs a mixed-methods approach, combining quantitative analysis using a Kinect sensor to measure gait parameters with qualitative assessment of fatigue levels using Borg scales. As a pilot study, it aims to replicate real-life walking conditions until participants experience fatigue, with a focus on minimizing disruptions from the experimental setup. While our controlled conditions do not fully replicate real-life scenarios, they provide a consistent testing environment that helps ensure participants safety, accurate and unbiased results. To achieve this, we are using the Kinect sensor, a non-intrusive technology that scans the entire body. This technology allows us to conduct comprehensive body scans without interfering with the subject’s natural movements. Our goal is to utilize the Kinect’s capabilities to discover new gait characteristics associated with fatigue during prolonged walking sessions. It is important to note that the findings from this initial investigation may not be generalizable to older adults, clinical populations, or varied real-world walking conditions; rather, this work serves as a foundational step for future, larger-scale studies. Additionally, our approach offers a distinct advantage over wearable sensors, which are typically limited to monitoring individuals one at a time. By leveraging the Kinect sensor, capable of detecting multiple people simultaneously in real-time, we can collectively monitor fatigue levels, especially in busy public spaces, and prevent accidents caused by fatigue during everyday activities.

Methods

This study is observed on individual participants walking on a treadmill without any distractions. Before taking part in the study, participants gave their written permission. The Semnan University of Medical Sciences ethics committee approved the study under the ethical principles of the 1964 Declaration of Helsinki (IR.SEMUMS.REC.1402.150).

Participants

In this study, we recruited 10 healthy young adults, with an equal distribution of five males and five females. Given the exploratory nature of this pilot study and the demanding, lengthy protocol required to induce fatigue, a small, well-defined sample was deemed appropriate for the initial validation of our novel measurement methodology. Participants were selected based on the following criteria:

• Age range: 20 to 30 years

• Engaged in at least 2 h of physical activity per week

• BMI below 35 kg/m²

• No history or current injury of the lower limbs or back

• Familiarity with treadmill walking

Those who did not meet these criteria were excluded from the study. These criteria ensure a group with similar physical conditions and activity levels.

In this study, the mean (SD) age, height, weight, and BMI of the female participants were 23.2 (2.9) years, 1.62 (0.03) m, 58.0 (9.3) kg, and 35.8 (5.6) kg/m², respectively. The corresponding values for the male participants were 26.4 (1.5) years, 1.79 (0.06) m, 76.3 (12.1) kg, and 23.8 (3.0) kg/m². Additionally, the mean (SD) walking duration for female participants was 16.83 (1.13) minutes and for males was 35.5 (12.24) minutes.

Experimental protocol

Prior to beginning the treadmill test, participants were given a comprehensive explanation of the Borg scale to ensure accurate reporting during the trial. All participants were required to remove their footwear for a consistent and unbiased testing environment before mounting the treadmill. In preliminary trials, we observed significant changes in gait parameters when the same individual wore different shoes, which reinforced our decision to have participants remove their footwear to minimize variability and ensure equal conditions for every participant. The test began with a warm-up phase where the treadmill speed was gradually increased over a 3-min period until reaching the target speed of 1 m/s, which was verified using the shaft rotation feedback sensor.

During the main testing phase, participants walked continuously at the fixed speed of 1 m/s. Throughout the trial, researchers collected Borg ratings at one-minute intervals to monitor participant exertion levels. The test continued until one of three termination criteria was met:

• The participant reported a Borg rating of 17 (defined as the operational threshold for significant physical fatigue)

• Voluntarily withdrew due to fatigue or discomfort

• The researchers observed signs of severe fatigue such as stumbling or instability that could increase fall risk.

This protocol, with its multiple termination criteria and continuous monitoring by researchers, was designed to ensure participant safety while still allowing collection of valuable data across both moderate and high fatigue states. In this study, all 10 participants completed the protocol by reaching a Borg rating of 17; the safety criteria were not triggered in any case. The test was structured to conclude before participants reached dangerous levels of exertion, with researchers maintaining vigilant observation throughout the trial period. For this trial, a minimum test duration of 10 min was required for a participant’s data to be included in the analysis. This threshold was established based on the clinical judgment of an orthopedic surgery specialist within the research team (co-author M. Zahraei), who indicated that a healthy young adult without musculoskeletal or cardiovascular impairment should be capable of sustaining treadmill walking at 1 m/s for at least 10 min prior to experiencing fatigue or discomfort. Consequently, failure to sustain walking for this minimum duration was interpreted as a potential indicator of an underlying health condition, and such participants’ data were excluded to preserve the integrity of the healthy-cohort sample.

Surface condition

An electric treadmill (Model DX3-A5, manufactured by DK City) with dimensions of 1.25 m length and 0.6 m width was utilized for the walking trials. To optimize motion capture, the treadmill handles, and control panel were removed (Figure 1). We created a custom Arduino program to regulate treadmill speed via interfaced with a laptop computer. A fall protection circuit was integrated into the Arduino board to immediately stop the treadmill if any loss of balance occurred. This system allowed precise control of the walking pace and safety monitoring throughout the experiment.

Figure 1.

The experimental set-up consists of a treadmill, a Kinect sensor, and an operator console. The participants are positioned on the treadmill facing the Kinect sensor, which is mounted on a tripod at a suitable height and distance. To ensure the safety of the participants, the operator monitors the speed of the treadmill and the risk of falling closely and stops the test if necessary. Additionally, a rope is connected to the participants’ clothes from the control board, which triggers the treadmill to stop instantly if the participants lose their balance.

Data collection

We used a Microsoft Kinect for Xbox 360 for data collection. This sensor simultaneously acquires RGB and depth imagery and automatically extracts the 3D coordinates of 20 anatomical skeletal landmarks in real time using its built-in depth-sensing and computer vision algorithms. ⁴⁵ Importantly, no physical markers are attached to participants at any stage; all landmark positions are detected virtually by the Kinect SDK skeletal tracking system. To ensure accurate data collection, we mount the sensor on a tripod positioned 1.1 m high and 1.6 m in front of the treadmill, facing the user (see Figure 1). This positioning allows for the full capture of the treadmill walkway. The Kinect uses IR sensors to detect joint depth, which enables it to capture sagittal angles and parameters even when positioned in front of the person. The sensor is connected to a laptop computer for data communication via a USB cable.

Gait analysis

Kinect sensor data was captured at a rate of 16 frames per second, and the 3D spatial information for each frame was stored in text files. We created a custom application to process and analyze this data. Gait cycles were identified by detecting instances of ankle crossover. With the precise 3D locations of 20 body landmarks available for each step, we could derive various gait parameters. To estimate any missing coordinate data due to limitations of the Kinect, we used a FUZZY algorithm that utilized valid neighboring frames as references. To implement the FUZZY algorithm, we first identified the missing data. We then utilized the 3D coordinates from 20 preceding and subsequent frames as inputs into the fuzzy mathematical formula (referred to as Formula (1)). In this formula, $C_n$ denotes the set of established fuzzy rules, which in this case are the positions of points in preceding and subsequent frames, while $Y_n$ represents the output calculated based on these inputs. Additionally, $X$ signifies the input data (namely, the missing points), T indicates the predictions generated by the fuzzy system, which are contingent on the input values $X$ , and $S$ refers to standard deviation, which we assigned a value of one.

$\mathrm{T}=\frac{{\displaystyle \sum }_{n=0}^{20}{Y}_n\times \left(e^{-{\left(\frac{X-C_n}{S}\right)}^2 }\right)}{{\displaystyle \sum }_{n=0}^{20}{e}^{{-\left(\frac{X-C_n}{S}\right)}^2 }}$

(1)

For example, we calculated the knee angle by examining the vectors between the hip, knee, and ankle joints. Additionally, our video-based motion capture approach allowed us to calculate additional kinematic parameters beyond basic spatiotemporal gait metrics. Specifically, by using the time-stamped position data across frames, we could derive angular velocities for the hip, knee, and ankle joints during the gait cycle. In total, we examined fluctuations in 14 distinct gait features throughout the fatigue progression. Afterwards, for each gait kinematic, we determined the maximum, minimum, and mean values within a single-step cycle. For the purpose of statistical comparison, the “initial value” for each parameter was calculated as the mean value during the first 2 minutes of stable walking (baseline), and the “final value” was calculated as the mean of the final minute before the trial was terminated due to fatigue. This allowed us to statistically quantify the total magnitude of change while maintaining our focus on the continuous progression of these features throughout the trial.

Statistical analysis

Statistical analysis was performed to evaluate the impact of fatigue on gait kinematics. The normality of the data was assessed, and a Paired Sample t-test was employed to compare the mean initial values (baseline) with the mean final values (fatigued state) for each of the six gait parameters. A p-value of less than 0.05 was considered statistically significant. To account for potential baseline differences between male and female participants, an Independent Samples t-test was performed on the magnitude of change (Final minus Initial values) for each parameter. This allowed us to evaluate whether the biomechanical response to fatigue was consistent across genders regardless of their starting gait characteristics. All statistical computations were implemented within the custom-developed gait analysis software using Delphi (Version 11).

Results

We examined how walking characteristics change as individuals experience fatigue, both in terms of time and perceived fatigue levels. To illustrate these changes over time, we calculated and presented these parameters for each step. Additionally, we presented these parameters based on the participants’ fatigue levels, as determined by the Borg criterion.

Assessing gait parameter changes associated with fatigue onset

The graphs below illustrate the significant changes in gait parameters observed as fatigue progresses in this study. We examined the temporal variations of these parameters and their correlation with participants’ fatigue levels, as measured by the Borg scale. The transition from non-fatigue to fatigue was not treated as a binary event but as a progressive shift. By correlating the Kinect-derived kinematics with the minute-by-minute Borg ratings, we identified the trajectory of fatigue. However, to confirm the robustness of these landmarks, we performed a statistical comparison between the starting period and the end period of walking, ensuring that the cumulative effect of fatigue resulted in significant biomechanical alterations (see Table 1).

Table 1.

Mean and standard deviations of the initial and final value of six gate characteristics of 10 participants and changes percentage.

Gait specification	MeanInitial value (SD)	MeanFinal value (SD)	Change percentage (%)	t-value	p-value
Maximum spine angle (˚)	2.2 (0.81)	3.4 (1.22)	52	−7.075	< 0.001
Maximum thigh sagittal angle (˚)	30.2 (2.42)	32.7 (2.01)	8.2	−3.954	0.003
Maximum knee angle of the non-dominant leg (˚)	21.25 (4.00)	26.25 (3.90)	23	−9.127	< 0.001
Maximum stride height per step (cm)	7.7 (2.33)	9 (2.81)	16.8	3.661	0.005
Maximum toe-out angle (˚)	45.3 (14.25)	50.4 (15.07)	11	−2.619	0.028
Maximum angular velocity of knee angle of the non-dominant leg (˚/s)	50 (9.67)	72 (26.22)	45	−3.026	0.014

Our findings revealed that the onset of fatigue resulted in alterations in six key gait parameters: maximum spine angle per step (Figure 2(a) θ ₁ ), maximum thigh sagittal angle per step (Figure 2(a) θ ₂ ), maximum knee angle of the non-dominant leg per step (Figure 2(a) θ ₃ ), maximum stride height (Figure 2(b) H), maximum toe out angle (external tibial rotation) between both feet (Figure 2(c) θ ₄ ), and maximum angular velocity of knee angle of the non-dominant leg per step (Figure 2(a) θ ₃ ).

Figure 2.

This Figure illustrates the anatomical locations of the body point landmarks captured by the Kinect sensor, including the shoulder center (1), spine (2), hip center (3), right and left knees (4), right and left ankles (5) and right and left foot (6). Additionally, this figure highlights the gait parameters analyzed to assess fatigue, in (a) namely spine angle (θ ₁ ), thigh sagittal angle (θ ₂ ), knee angle (θ ₃ ), in (b) stride height (H) and in (c) toe out angle (θ ₄ ).

Tracking these biomechanical landmarks could offer insights into the development of neuromuscular and postural fatigue during walking tasks. To enhance clarity and interpretability of the results, a regression analysis was performed to identify and exclude data points that deviated significantly from the expected trends. This reduced noise and improved the accuracy of the model. Additionally, a smoothing technique was employed by averaging adjacent data points to reveal the underlying patterns of parameter changes over time, while minimizing random fluctuations.

Study findings

Data analysis of the charts revealed clear trends in certain gait parameters as participants experienced fatigue. Specifically, the maximum spine angles increased (Figure 3 Chart A), and the maximum thigh sagittal angle increased on a per-step basis (Figure 3 Chart B), suggesting biomechanical compensations. The maximum knee angle of the non-dominant leg in each step also became larger (Figure 3 Chart C), indicating reduced flexion. Moreover, the maximum stride height per step increased (Figure 3 Chart D). In contrast, the toe-out angle also increased (Figure 3 Chart E), reflecting a more unstable gait pattern. Additionally, the angular velocity of the knee angle increased per step (Figure 3 Chart F), implying greater variability and motor control challenges. The strictly increasing patterns exhibited by these six walking characteristics demonstrate their potential as reliable indicators of fatigue progression.

Figure 3.

The six graphs presented visualize data collected from a study of ten participants undergoing a fatigue protocol. Each graph contains ten overlaid line charts, one for each participant. The horizontal axis conveys the number of steps taken during the protocol. The vertical axis indicates the measurement values for the following gait parameters: maximum spine angle per step (Chart A), maximum thigh sagittal angle per step (Chart B), maximum knee angle of the non-dominant leg per step (Chart C), maximum stride height per step (Chart D), maximum toe out angle between both feet per step (Chart E) and maximum angular velocity of knee angle per step (Chart F).

As visualized in Figures 3 and 4, the transition from baseline walking to a state of significant fatigue is characterized by a clear upward trajectory in each landmark. These plots provide a granular view of how each participant’s gait biomechanics adapted over time, with the final fatigued state (Borg 17) consistently representing the peak values for all parameters. This temporal analysis, combined with the statistical comparisons in Table 1, confirms that the identified landmarks are robust and sensitive to the progressive nature of walking fatigue. Our study observed that female participants exhibited quicker onset of fatigue compared to male participants. This is quantitatively supported by the significantly shorter walking duration observed in females (Mean (SD = Standard deviation) = 16.83 (1.13) minutes) compared to males (Mean (SD) = 35.5 (12.24) minutes). However, the resulting changes in gait parameters due to fatigue were similar across both genders, indicating that while the rate of fatigue onset varies, the biomechanical impact of fatigue on gait remains consistent.

Figure 4.

Similar to the previous figure, this figure presents data on six gait parameters collected from the ten study participants during the fatigue protocol. Specifically, the six graphs illustrate the change in participants’ fatigue level, as measured by Borg’s rating, for the following biomechanics: maximum spine angle per step (Chart A), maximum thigh sagittal angle per step (Chart B), maximum knee angle of the non-dominant leg per step (Chart C), maximum stride height per step (Chart D), maximum toe out angle between both feet per step (Chart E), and maximum angular velocity of knee angle per step (Chart F). Just like the previous visualizations, each graph contains ten line charts overlaid on top of each other, representing the trajectories for the ten individual participants.

To better understand how fatigue parameters change over time, we have plotted these parameters against the Borg scale, which measures perceived exertion. The Borg scale reflects the level of fatigue experienced by the participants in this study. The results are presented in Figure 4.

We calculated the mean initial and final values of six gait features for our 10 participants and analysed the differences in means and their standard deviations (Table 1). The results indicated significant changes in several parameters. The most substantial change was observed in the maximum spine angle per step, with mean (SD) values increasing from 2.2 (0.81) to 3.4 (1.22), representing a 52% increase. The second most significant change was the maximum angular velocity of the knee angle of the non-dominant leg per step, with mean (SD) values increasing from 50 (9.67) to 72 (26.22), reflecting a 45% increase. Other changes included the maximum knee angle of the non-dominant leg per step, with mean (SD) values increasing from 21.25 (4.00) to 26.25 (3.90), marking a 23% increase. The maximum stride height per step had mean (SD) values of 7.7 (2.33) initially and 9 (2.81) finally, representing a 16.8% increase. The maximum toe-out angle between both feet per step showed mean (SD) values of 45.3 (14.25) initially and 50.4 (15.07) finally, indicating an 11% increase. Finally, the maximum thigh sagittal angle per step had mean (SD) values of 30.2 (2.42) initially and 32.7 (2.01) finally, showing an 8.2% increase. This analysis is illustrated in Figure 5.

Figure 5.

Chart of comparison of the value of changes in founded gait features in terms of percentage, sorted from most changes to the least changes.

The increase in standard deviation suggests greater variability in participants’ responses to fatigue, indicating differences in individual fitness levels, fatigue resistance, or personal gait mechanics.

As shown in Table 1, the Paired Sample t-test confirmed that the increases in all six biomechanical landmarks were statistically significant. Specifically, the maximum spine angle (p < 0.001) and maximum knee angle (p < 0.001) showed the highest levels of statistical significance.

To further investigate the consistency of the fatigue impact across genders, we compared the magnitude of change (Final - Initial) for each gait parameter between male and female participants (Table 2). An Independent Samples t-test revealed no significant differences in the magnitude of biomechanical alterations between the two groups (p > 0.05 for all parameters). This statistical evidence confirms that despite the differences in fatigue onset time and potential variations in baseline gait patterns between individuals, the biomechanical response to exhaustion remains consistent regardless of gender (Table 2). The use of delta values (magnitude of change) ensured that our analysis focused on the physiological impact of fatigue, effectively normalizing for individual and gender-based baseline differences.

Table 2.

Comparison of the magnitude of change in gait parameters between genders.

Gait feature	Mean male change (SD)	Mean female change (SD)	p-value (gender Diff)
Max spine angle (˚)	1.33 (0.41)	1.07 (0.66)	0.462
Max thigh sagittal angle (˚)	2.83 (2.53)	2.16 (1.52)	0.625
Max knee angle (˚)	5.03 (1.74)	4.97 (1.92)	0.964
Max stride height (cm)	1.34 (1.61)	1.26 (0.48)	0.918
Max toe-out angle (˚)	2.66 (1.74)	7.54 (8.21)	0.229
Max angular velocity (˚/s)	25.02 (24.95)	18.98 (23.33)	0.703

Discussion

The purpose of this research was to investigate how fatigue affects gait kinematics during prolonged walking at a steady pace. The results showed that fatigue caused significant changes in certain gait parameters, such as spine bending angle, knee angle, step height, thigh sagittal angle, toe out angle, and angular velocity of the knee angle. While some of these features have been individually noted in previous studies—for example, knee kinematics ²⁹ and toe-out angle as a stabilizer ⁴⁶ —their collective identification as a robust suite for continuous fatigue tracking is a key contribution of this work. Most prior investigations focus on localized changes or spatiotemporal metrics like step length. ⁹ In contrast, our use of whole-body imaging allowed for the discovery that spinal flexion (maximum spine angle) and maximum stride height also increase consistently with fatigue, likely as compensatory mechanisms for maintaining balance and clearing the treadmill surface as muscle efficiency declines.

Previous studies have reported different effects of fatigue on gait parameters. ²¹ Step length ⁴⁷ and knee kinematics ⁴⁸ have been identified as indicators of fatigue during walking. In addition, this study used whole-body imaging to measure other parameters such as spinal flexion and step height. Video-based motion capture allowed us to calculate additional kinematic parameters, such as angular velocity of the knee joint, which were also found to be influenced by fatigue. Unlike previous studies that relied on wearable sensors to monitor specific body parts of participants, this study used the Kinect sensor to capture the movements of multiple subjects without any interference.

Study limitations

While this study demonstrates the potential of the Kinect sensor for continuous fatigue monitoring, several technical and methodological limitations must be acknowledged. First, the Kinect sensor (v1) has a lower sampling rate (approximately 16–30 fps) and lower spatial resolution compared to high-end optoelectronic motion capture systems. ⁴⁹ This can lead to less precise joint coordinate estimation, particularly during rapid movements. Second, the sensor’s reliance on infrared depth sensing makes it sensitive to ambient lighting conditions and reflective surfaces, which may introduce noise into the skeletal data.

Furthermore, “self-occlusion”—where one body part obstructs the sensor’s view of another—is a known challenge in single-sensor setups. Although we employed a custom FUZZY algorithm to estimate missing coordinates and applied smoothing techniques to mitigate sensor-induced noise (deviations), such estimations are approximations of the true movement. These preprocessing steps were essential to ensure that the identified gait trends accurately reflected physiological fatigue rather than random sensor fluctuations. Finally, the Kinect’s skeletal tracking algorithm is optimized for upright, front-facing postures; therefore, subtle rotations or extreme sagittal movements may be subject to higher error margins than traditional marker-based systems. ⁵⁰ Other limitations is use the use of the Borg criterion rating as the only criteria for detecting fatigue, without incorporating other objective metrics such as Heart Rate (HR) or Heart Rate Variability (HRV), limits the conclusions that can be drawn. As noted in literature, physiological markers like HRV are more sensitive indicators of cardiovascular and autonomic fatigue compared to SpO2, which often remains stable in healthy individuals even under significant exertion.

Despite these constraints, the Kinect’s portability and non-intrusive nature provide significant advantages for real-time applications in non-laboratory settings, provided that appropriate data filtering and smoothing techniques are applied as demonstrated in this study.

Conclusion

Fatigue detection is an important aspect of human health, and it presents opportunities for non-invasive monitoring through various physiological changes, such as walking parameters. In this study, we have utilized the Kinect sensor, a widely available and affordable device, to analyze walking mechanics with minimal interference and extract innovative features. Our findings consistently show that certain gait kinematics, including the maximum spine angle, maximum thigh sagittal angle, maximum knee angle of the non-dominant leg, maximum stride height per step, maximum toe-out angle, and maximum angular velocity of the knee angle of the non-dominant leg per step, increase with the onset and progression of fatigue. Tracking and analyzing these changes have the potential to identify fatigue states in individuals and provide feedback or intervention to prevent adverse effects on health and performance. The future direction of this study will focus on mimicking even more realistic walking conditions in real life. To increase accuracy, it will involve using more advanced sensors and deploying artificial intelligence (AI) and machine learning to identify gait features.