How Minimal Skeleton Features Can Improve Real-Time Human Action Recognition

on

Human Action Recognition (HAR) has become one of the most active research areas in computer vision, intelligent surveillance, and human-centered artificial intelligence. Modern applications—including smart security systems, healthcare monitoring, autonomous robotics, and intelligent transportation—depend on accurate recognition of human activities in dynamic environments. However, many existing recognition methods rely on large volumes of video frames and complete body skeleton information, leading to high computational costs and limiting their deployment in real-time or resource-constrained systems.

The study reviewed in this article proposes an alternative perspective by demonstrating that reliable recognition of walking and running actions can be achieved using only temporal foot-lift information extracted from a small number of frames. Rather than increasing model complexity, the researchers focus on identifying the most informative motion features while reducing computational requirements. This work provides valuable insights into efficient human motion analysis and highlights opportunities for developing lightweight action recognition systems suitable for next-generation intelligent engineering applications.

Bibliographic Information

Item Information
Article Title Recognition Human Walking and Running Actions Using Temporal Foot-Lift Features
Authors Khin Cho Tun, Hla Myo Tun, Lei Lei Yin Win, and Khin Kyu Kyu Win
Journal Innovation in Engineering
Volume & Issue Volume 1, Issue 1
Publication Year 2024
Pages 1–20
DOI https://doi.org/10.58712/ie.v1i1.1
Publisher Researcher and Lecturer Society
License Creative Commons Attribution 4.0 International (CC BY 4.0)

1. Research Background

  • Human action recognition has become an essential enabling technology for intelligent systems. Modern engineering applications increasingly rely on automatic recognition of human movements to improve safety, automation, and decision-making. Fields such as smart surveillance, healthcare monitoring, sports analytics, assistive robotics, and human–robot interaction require reliable identification of fundamental activities including walking and running.
  • Walking and running are among the most frequently analyzed human actions. These movements serve as basic indicators of human behavior and often represent the first stage of more complex activity recognition systems. Accurate differentiation between walking and running is important for security monitoring, emergency detection, rehabilitation assessment, and intelligent mobility analysis.
  • Earlier vision-based approaches exhibit several practical limitations. Traditional methods frequently relied on RGB images, silhouette extraction, motion vectors, gait frequency, or binary image sequences. While these techniques achieved reasonable performance, they were often sensitive to lighting conditions, self-occlusion, viewpoint variation, and environmental complexity, limiting their robustness in real-world applications.
  • The adoption of RGB-D cameras and skeleton joint data significantly improved action recognition. Three-dimensional skeleton representations provide richer information about body posture while reducing sensitivity to illumination changes. Consequently, skeleton-based approaches have become one of the dominant research directions for human motion analysis.
  • Existing skeleton-based methods typically require extensive spatial information. Many published studies utilize between 15 and 30 body joints for every frame, producing large volumes of spatial data. As the number of joints and recorded frames increases, computational complexity, memory requirements, and processing time also increase substantially, making real-time implementation more challenging.
  • Large temporal windows remain another important challenge. Numerous previous studies require dozens or even hundreds of video frames to recognize a single human action reliably. Longer observation periods increase computational cost and delay system responses, which is undesirable for applications requiring immediate decision-making.
  • The literature reveals an opportunity for feature reduction. Although many recognition systems continue to exploit complete-body information, relatively few studies investigate whether a small subset of highly informative motion features can provide comparable recognition performance. This raises an important research question regarding the balance between computational efficiency and classification accuracy.
  • The present study introduces a minimalist feature extraction strategy. Instead of processing the entire human skeleton, the researchers focus exclusively on temporal foot-lift characteristics. The proposed approach extracts discriminative motion information from only the feet over five to seven consecutive frames, aiming to reduce computational requirements while preserving recognition accuracy.
  • The novelty lies in combining simplicity with practical performance. By reducing both the number of required skeleton joints and the temporal observation window, the proposed framework seeks to establish an efficient recognition method suitable for real-time intelligent systems without depending on computationally intensive deep learning architectures.

2. Research Objectives

  • To develop a lightweight human action recognition approach capable of distinguishing walking and running using temporal foot-lift features derived from skeleton joint data.
  • To investigate whether reliable action recognition can be achieved using only foot joint information instead of complete human body skeleton representations.
  • To minimize the number of video frames required for action recognition by extracting informative temporal features from short motion sequences consisting of only five to seven consecutive frames.
  • To design a set of normalized temporal foot-lift descriptors that remain robust against differences in body size, movement direction, and individual physical characteristics.
  • To evaluate the effectiveness of the proposed temporal foot-lift features using a weighted K-Nearest Neighbor (KNN) classifier.
  • To validate the proposed method across four publicly available human action datasets representing different recording environments, skeleton configurations, and participant characteristics.
  • To compare the recognition performance of the proposed framework with existing approaches reported in previous studies while emphasizing computational efficiency.

3. Why This Research Matters

  • Enables efficient real-time intelligent surveillance. Reducing computational complexity allows action recognition systems to respond more quickly, making them suitable for smart surveillance environments where immediate detection of abnormal human behavior is essential.
  • Supports lightweight artificial intelligence applications. By extracting only the most informative motion features, the proposed approach demonstrates that effective recognition does not always require large-scale deep neural networks or complete body representations, opening opportunities for lightweight AI deployment.
  • Improves engineering efficiency. Lower memory requirements and fewer processed frames reduce computational load, enabling implementation on embedded systems, edge devices, mobile robots, and other hardware with limited processing capability.
  • Enhances healthcare and assisted living technologies. Reliable recognition of walking and running can contribute to patient monitoring, elderly care, rehabilitation assessment, and fall-risk analysis while supporting continuous, automated observation without requiring wearable sensors.
  • Contributes to robotics and autonomous systems. Human motion understanding is fundamental for safe human–robot collaboration. Efficient recognition algorithms improve robot awareness of surrounding human activities, facilitating safer interaction in industrial, healthcare, and public environments.
  • Promotes scalable smart city infrastructure. Lightweight recognition methods can be integrated into distributed surveillance networks, intelligent transportation systems, and public safety applications without demanding expensive computational resources.
  • Demonstrates the value of feature engineering. At a time when many studies focus primarily on increasingly complex deep learning models, this research highlights that carefully designed domain-specific features remain highly valuable for solving engineering problems efficiently.
  • Provides a practical foundation for future human activity recognition research. The proposed methodology offers a flexible framework that can potentially be extended to recognize more complex activities while maintaining computational efficiency, making it attractive for future research in computer vision, biomechanics, and intelligent engineering systems.

4. Research Methodology

  • Research Design

    This study employed a quantitative experimental research design to develop and evaluate a lightweight human action recognition method based on skeleton joint information. Rather than utilizing complete body posture or deep learning architectures, the researchers proposed a feature engineering approach that extracts temporal characteristics from foot movements. The methodology emphasizes computational efficiency while maintaining competitive recognition performance for distinguishing walking and running actions.

  • Input Data

    The proposed method uses three-dimensional skeleton joint data rather than conventional RGB images. Skeleton information was obtained from four publicly available benchmark datasets that had previously been developed using RGB-D cameras and motion capture technologies. Although the datasets contain different numbers of skeleton joints, the proposed algorithm only requires the coordinates of both feet, allowing it to operate independently of the complete body structure.

  • Datasets

    Experimental validation was performed using four well-established human action datasets representing various recording environments, camera configurations, and human subjects:

    • KARD Dataset
    • UTKinect Dataset
    • G3D Dataset
    • CMU Motion Capture (CMU MoCap) Dataset

    These datasets include walking and running sequences performed by multiple male and female participants under different movement conditions. The diversity of datasets allows the proposed method to be evaluated under varying skeleton structures, recording perspectives, and environmental settings.

  • Feature Engineering

    The core contribution of the methodology is the extraction of temporal foot-lift features. Instead of analyzing all body joints, the researchers calculate the vertical displacement of both feet relative to the walking path over a short sequence of consecutive frames. To improve robustness against differences in body size, all measurements are normalized using individual body height.

    Five discriminative features are subsequently derived from the normalized foot trajectories:

    • Difference between normalized mean foot lifts (ΔRfm)
    • Maximum normalized foot lift (Rf,max)
    • Root Mean Square Error (Erms) between both feet
    • Vertical Pixel Motion Index (PMIy)
    • Three-dimensional Pixel Motion Index (PMIx,y,z)

    Together, these descriptors capture both spatial and temporal characteristics of foot motion while requiring only a small amount of skeletal information.

  • Temporal Window

    Unlike many existing action recognition approaches that require long video sequences, the proposed method extracts features from only five or seven consecutive frames. At a recording rate of 30 frames per second, this corresponds to observation windows of approximately 0.167 seconds and 0.233 seconds, respectively. Such short temporal windows enable rapid recognition while substantially reducing computational requirements.

  • Mathematical Modeling

    The methodology estimates the local walking path using linear regression based on the coordinates of the lower foot. The vertical distance between each foot and the estimated motion path is then calculated to determine normalized foot-lift values. Additional statistical and motion-based descriptors are computed to characterize differences between walking and running gait patterns across consecutive frames.

  • Classification Algorithm

    Human actions were classified using a weighted K-Nearest Neighbor (Weighted KNN) classifier. Euclidean distance was employed to measure similarity between training and testing samples, while neighboring observations contributed to the final prediction according to inverse-distance weighting. The classifier parameter was fixed at K = 10 based on previous studies demonstrating the effectiveness of KNN for skeleton-based human action recognition.

  • Training and Testing Strategy

    Each dataset was divided equally into training and testing subsets. Fifty percent of the observations were used to construct the classification model, while the remaining fifty percent were reserved for independent performance evaluation. This balanced strategy helps reduce the risk of both overfitting and underfitting while providing an objective assessment of the proposed recognition framework.

  • Software Environment

    All feature extraction, model development, and experimental evaluations were implemented using MATLAB. The computational framework enables reproducible experiments and provides a flexible environment for processing skeleton joint data and evaluating recognition performance across multiple benchmark datasets.

  • Performance Evaluation

    The effectiveness of the proposed method was assessed by measuring classification accuracy across the four benchmark datasets. Additional comparisons were conducted with previously published human action recognition methods to evaluate whether the proposed lightweight feature extraction strategy could achieve competitive performance despite using substantially fewer skeleton joints and shorter observation periods.

5. Key Findings

Temporal Foot-Lift Features Successfully Distinguish Walking and Running

The study demonstrates that temporal foot-lift characteristics provide sufficient information to discriminate between walking and running without requiring complete body skeleton data. Across all evaluated datasets, running actions consistently produced greater foot elevation, larger motion amplitudes, and stronger temporal variations than walking actions. These distinctive motion patterns enabled reliable classification using only five engineered features extracted from both feet.

This finding is particularly significant because it challenges the common assumption that accurate human action recognition necessarily requires information from the entire human body. Instead, the results indicate that carefully selected local motion descriptors can capture essential biomechanical differences while substantially reducing computational complexity.

High Recognition Accuracy Was Achieved Using Very Short Motion Sequences

One of the most notable outcomes of the research is the ability to recognize human actions using remarkably short temporal windows. The proposed method achieved an overall recognition accuracy of approximately 88.6% when only five consecutive frames were analyzed. When the observation window was increased to seven frames, the average recognition accuracy improved further to approximately 90.7%.

Considering that many conventional human action recognition methods require dozens or even hundreds of frames, these results demonstrate that reliable recognition can be achieved with substantially less temporal information. This characteristic makes the proposed approach highly attractive for real-time engineering applications where rapid decision-making is essential.

Reducing Skeleton Information Does Not Necessarily Reduce Performance

The experimental evaluation included datasets containing between fifteen and forty-one skeleton joints. Despite these differences, the proposed framework relied exclusively on the coordinates of two foot joints throughout the entire recognition process. Because all datasets were processed using the same simplified feature extraction strategy, classification performance remained relatively stable regardless of the original skeleton complexity.

This observation suggests that increasing the number of input features does not automatically improve recognition accuracy. Instead, selecting highly informative biomechanical features may provide a more efficient solution while minimizing storage requirements and computational cost.

Normalization Improved Robustness Across Different Participants

Human participants naturally differ in body height, stride length, gait style, and movement speed. To reduce the influence of these physical differences, the researchers normalized all foot-lift measurements using each participant's body height before feature extraction. As a result, the proposed descriptors became less sensitive to individual anatomical variations.

Experimental results indicate that this normalization strategy contributed to stable recognition performance across male and female participants as well as across multiple datasets collected under different experimental conditions. Such robustness is essential for practical deployment in heterogeneous real-world populations.

Weighted KNN Provides an Effective Lightweight Classifier

Although recent human action recognition studies frequently employ deep neural networks, this research demonstrates that a conventional machine learning classifier remains highly effective when combined with informative handcrafted features. The weighted K-Nearest Neighbor classifier successfully separated walking and running actions using only five temporal descriptors extracted from short motion sequences.

These findings highlight that recognition performance depends not only on classifier complexity but also on feature quality. The study therefore illustrates the continuing importance of feature engineering as an efficient alternative for applications where computational resources are limited.

The Proposed Framework Shows Strong Potential for Real-Time Engineering Applications

By requiring only two foot joints and a maximum of seven video frames, the proposed recognition framework substantially reduces processing time, memory usage, and computational workload compared with many existing skeleton-based approaches. The simplified workflow is well suited to embedded computing platforms and intelligent edge devices where processing resources are constrained.

The overall findings suggest that temporal foot-lift analysis offers a practical foundation for future real-time human activity recognition systems deployed in surveillance, healthcare, robotics, sports engineering, and other intelligent environments requiring rapid and reliable movement recognition.

6. Scientific Contribution

  • Introduces a novel temporal foot-lift feature representation. Rather than relying on complete human skeleton information, the study proposes a compact set of temporal foot-lift descriptors that effectively capture the biomechanical differences between walking and running. This feature engineering strategy demonstrates that discriminative motion information can be extracted from only two foot joints.
  • Reduces the spatial complexity of skeleton-based action recognition. Most previous approaches process 15–40 body joints simultaneously, increasing computational cost and memory requirements. By restricting analysis to foot joint coordinates, the proposed framework significantly simplifies feature extraction while maintaining competitive recognition performance.
  • Minimizes temporal requirements for human action recognition. The proposed methodology recognizes actions using only five to seven consecutive frames, substantially reducing the observation period compared with many conventional approaches that require much longer video sequences.
  • Develops normalized temporal motion descriptors. The researchers normalize foot-lift measurements using body height, making the extracted features more robust to variations in participant stature and gait characteristics. This contributes to improved generalization across different individuals and datasets.
  • Demonstrates the continuing importance of feature engineering. While recent studies increasingly depend on computationally intensive deep learning models, this work illustrates that carefully designed handcrafted features remain capable of delivering reliable classification performance with considerably lower computational demands.
  • Provides a practical methodology for lightweight intelligent systems. The combination of compact feature extraction and weighted KNN classification offers a computationally efficient framework suitable for real-time implementation on embedded and edge computing platforms.
  • Validates the proposed framework across multiple benchmark datasets. Performance evaluation using four publicly available datasets strengthens the generalizability of the proposed methodology by demonstrating consistent behavior under different recording environments and skeleton configurations.
  • Expands the methodological foundation of skeleton-based human action recognition. The study contributes an alternative design philosophy that prioritizes informative feature selection rather than increasing model complexity, providing useful guidance for future research in computer vision and intelligent engineering.

7. Industrial Implications

  • Smart Surveillance Systems. Lightweight action recognition enables surveillance cameras to distinguish walking and running behaviors with minimal computational resources, supporting faster detection of abnormal or emergency situations in public spaces such as airports, transportation hubs, hospitals, shopping centers, and university campuses.
  • Healthcare and Assisted Living. The proposed approach can support continuous monitoring of patients and elderly individuals by automatically recognizing locomotion patterns. Such capability may contribute to rehabilitation monitoring, mobility assessment, and early detection of unusual movement behaviors without requiring wearable sensors.
  • Human–Robot Collaboration. Robots operating in shared environments require rapid interpretation of surrounding human activities. Efficient recognition of walking and running allows collaborative robots, service robots, and autonomous assistants to anticipate human movement and improve interaction safety.
  • Embedded Artificial Intelligence. Because the proposed framework requires only limited computational resources, it is suitable for deployment on embedded processors, edge AI devices, microcontrollers, and Internet of Things (IoT) platforms where processing power and memory are constrained.
  • Sports Engineering. Motion analysis based on temporal foot-lift features may support athlete performance evaluation, gait analysis, movement classification, and sports training systems by providing efficient recognition of locomotion activities without requiring full-body motion capture.
  • Digital Engineering Applications. The proposed methodology aligns with current trends toward lightweight computer vision solutions that reduce computational overhead while maintaining reliable system performance. Such characteristics are valuable for distributed sensing networks and real-time intelligent monitoring systems.
  • Industry 4.0. Intelligent factories increasingly integrate computer vision for workplace safety and human activity monitoring. Efficient recognition of worker movements may improve occupational safety, collaborative manufacturing, and automated production supervision while reducing hardware requirements.
  • Engineering Practice. The study demonstrates that engineering solutions can benefit from carefully selected domain-specific features rather than relying exclusively on increasingly complex machine learning architectures. This perspective encourages the development of computationally efficient intelligent systems across multiple engineering disciplines.

8. Research Limitations

  • The proposed framework was developed specifically to distinguish between two fundamental locomotion activities—walking and running. Its applicability to more complex human behaviors remains to be investigated.
  • Experimental validation relied on previously published benchmark datasets rather than continuous real-world surveillance scenarios. Additional evaluation under operational conditions would further strengthen practical applicability.
  • Only foot joint information was utilized throughout feature extraction. Although this substantially reduces computational complexity, certain complex human activities may require additional body joints to achieve reliable discrimination.
  • The classification model employed a weighted K-Nearest Neighbor algorithm. Performance comparisons with more recent machine learning and deep learning approaches were outside the primary scope of the study.
  • The proposed methodology assumes that reliable skeleton joint data are available prior to feature extraction. Consequently, recognition performance may depend on the accuracy of the preceding skeleton estimation process.
  • Only short temporal observation windows consisting of five and seven frames were investigated. Alternative temporal window sizes may provide additional insights into the balance between computational efficiency and recognition accuracy.
  • The research focused on offline evaluation using publicly available datasets. Real-time deployment aspects such as latency, hardware implementation, and energy consumption were not experimentally assessed.
  • Although multiple datasets were included, further validation using larger and more diverse populations would strengthen evidence regarding the generalizability of the proposed temporal foot-lift descriptors.

9. Future Research Opportunities

  1. Extend the proposed temporal foot-lift framework to recognize a broader range of human activities, including jumping, climbing, sitting, falling, and transitional movements.
  2. Investigate the integration of temporal foot-lift features with deep learning architectures to combine lightweight handcrafted descriptors with automated feature learning.
  3. Evaluate the proposed methodology under real-time surveillance conditions involving occlusion, crowd density, varying illumination, and outdoor environments.
  4. Study the robustness of the proposed features across different age groups, physical conditions, gait disorders, and rehabilitation scenarios.
  5. Develop adaptive feature extraction strategies capable of automatically selecting the optimal temporal observation window according to movement characteristics.
  6. Integrate temporal foot-lift analysis with wearable sensors, inertial measurement units (IMUs), or multimodal sensing systems to improve recognition reliability.
  7. Investigate deployment on edge computing platforms and embedded AI hardware to quantify computational efficiency, energy consumption, and real-time processing performance.
  8. Compare the proposed feature engineering approach with contemporary transformer-based and graph neural network methods for skeleton-based action recognition.
  9. Explore applications of temporal foot-lift descriptors in sports biomechanics, clinical gait assessment, rehabilitation engineering, and occupational safety monitoring.
  10. Develop generalized lightweight human action recognition frameworks that combine multiple localized biomechanical features while maintaining low computational complexity for large-scale intelligent monitoring systems.

10. Potential for Public Policy Citation (Overton)

Although this study is primarily an engineering and computer vision investigation, it demonstrates meaningful potential for informing future public policies related to intelligent surveillance, digital infrastructure, healthcare technology, and artificial intelligence deployment. Rather than proposing regulatory frameworks or policy recommendations directly, the research provides technical evidence supporting the development of lightweight human action recognition systems that could be incorporated into broader smart city initiatives and public digital transformation strategies.

The proposed temporal foot-lift approach emphasizes computational efficiency without sacrificing recognition performance, making it particularly relevant for public-sector applications where scalability, affordability, and real-time processing are essential. Because the methodology reduces hardware requirements and computational costs, it may support wider adoption of intelligent monitoring technologies in both developed and resource-constrained environments.

  • Government Reports. The study may be cited in reports discussing artificial intelligence, intelligent surveillance systems, digital public services, computer vision, and smart infrastructure development.
  • Smart City Roadmaps. The lightweight recognition framework aligns well with smart city initiatives that require efficient computer vision algorithms for public safety, intelligent transportation, and urban monitoring.
  • Healthcare Digitalization Strategies. Automated recognition of locomotion activities has potential relevance for digital healthcare policies involving elderly care, rehabilitation monitoring, remote patient observation, and assisted living technologies.
  • Artificial Intelligence Strategies. National AI roadmaps increasingly encourage efficient, trustworthy, and deployable machine intelligence. This research contributes practical evidence supporting lightweight AI implementations suitable for real-world deployment.
  • Technical Standards. Although the study does not propose new standards, its feature extraction methodology may contribute to future technical guidelines for skeleton-based human activity recognition and intelligent vision systems.
  • Industry 4.0 Policies. Human activity recognition plays an important role in intelligent manufacturing, collaborative robotics, workplace safety, and industrial automation, making the proposed approach relevant to digital manufacturing initiatives.
  • Innovation Policies. The research illustrates how computational efficiency can accelerate technology adoption, supporting innovation strategies that encourage practical and scalable artificial intelligence solutions.
  • Current Policy Impact. Since the article focuses primarily on methodological innovation rather than policy evaluation, its immediate likelihood of citation in public policy documents is considered moderate. Nevertheless, as lightweight computer vision technologies become increasingly integrated into smart infrastructure and digital public services, the study's long-term policy relevance is expected to grow.

11. Who Should Read This Paper?

  • Researchers working in computer vision and human action recognition.
  • Artificial intelligence and machine learning researchers.
  • Computer engineers developing intelligent vision systems.
  • Robotics engineers working on human–robot interaction.
  • Graduate students studying computer vision, artificial intelligence, biomechanics, or intelligent systems.
  • Researchers in healthcare engineering and rehabilitation technology.
  • Sports scientists and biomechanical engineers.
  • Embedded systems and edge AI developers.
  • Engineers designing smart surveillance platforms.
  • Industry practitioners implementing Industry 4.0 technologies.
  • Smart city planners and digital infrastructure developers.
  • Government agencies exploring artificial intelligence for public services.
  • Educators teaching computer vision, machine learning, and intelligent systems engineering.

12. Final Thoughts

This study presents a thoughtful and well-executed contribution to the field of skeleton-based human action recognition by demonstrating that efficient feature engineering can significantly reduce computational requirements without substantially compromising classification performance. Rather than pursuing increasingly complex recognition models, the authors focus on identifying a compact set of informative temporal foot-lift descriptors capable of distinguishing walking and running using only two skeletal joints and a very short observation window. This design philosophy offers an attractive alternative for applications where computational efficiency is as important as recognition accuracy.

A notable strength of the research lies in its emphasis on practical engineering implementation. The methodology is mathematically transparent, experimentally validated using multiple benchmark datasets, and sufficiently lightweight for deployment on embedded platforms, intelligent sensors, and edge computing devices. Although the investigation is limited to two locomotion activities and employs a conventional machine learning classifier, the proposed framework establishes a strong foundation for future lightweight human activity recognition systems. As intelligent surveillance, healthcare monitoring, collaborative robotics, and smart city technologies continue to expand, this research demonstrates that carefully designed domain-specific features remain an important complement to modern artificial intelligence. Overall, the article represents a valuable contribution for researchers and practitioners seeking computationally efficient approaches to real-time human motion recognition.

Suggested Citation

UNP–Teknomekanik Style

Tun KC, Tun HM, Win LLY, Win KKK. Recognition Human Walking and Running Actions Using Temporal Foot-Lift Features. Innovation in Engineering. 2024;1(1):1–20. Available from: https://doi.org/10.58712/ie.v1i1.1.

APA (7th Edition)

Tun, K. C., Tun, H. M., Win, L. L. Y., & Win, K. K. K. (2024). Recognition human walking and running actions using temporal foot-lift features. Innovation in Engineering, 1(1), 1–20. https://doi.org/10.58712/ie.v1i1.1

IEEE Style

K. C. Tun, H. M. Tun, L. L. Y. Win, and K. K. K. Win, "Recognition Human Walking and Running Actions Using Temporal Foot-Lift Features," Innovation in Engineering, vol. 1, no. 1, pp. 1–20, 2024. doi: 10.58712/ie.v1i1.1.

Harvard Style

Tun, K.C., Tun, H.M., Win, L.L.Y. & Win, K.K.K., 2024. Recognition human walking and running actions using temporal foot-lift features. Innovation in Engineering, 1(1), pp.1–20. Available at: https://doi.org/10.58712/ie.v1i1.1.

Vancouver Style

Tun KC, Tun HM, Win LLY, Win KKK. Recognition human walking and running actions using temporal foot-lift features. Innovation in Engineering. 2024;1(1):1–20. Available from: https://doi.org/10.58712/ie.v1i1.1

Chicago (Author–Date)

Tun, Khin Cho, Hla Myo Tun, Lei Lei Yin Win, and Khin Kyu Kyu Win. 2024. "Recognition Human Walking and Running Actions Using Temporal Foot-Lift Features." Innovation in Engineering 1 (1): 1–20. https://doi.org/10.58712/ie.v1i1.1.

MLA (9th Edition)

Tun, Khin Cho, et al. "Recognition Human Walking and Running Actions Using Temporal Foot-Lift Features." Innovation in Engineering, vol. 1, no. 1, 2024, pp. 1–20. https://doi.org/10.58712/ie.v1i1.1.

Editorial Note

Editorial Note: This blog post is an independent scholarly review intended for educational and scientific communication purposes. It summarizes and discusses the published article in the author's own words while providing full attribution to the original publication, consistent with the principles of the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Comments

Post a Comment