Human Motion Analysis Research Articles

A Review of 3D Human Pose Estimation

Abstract. Visual perception and human body recognition are fundamental capabilities required for effective and safe interactions between artificial intelligence (AI), computer vision, and humans in real-world scenarios. Recent groundbreaking developments in AI and computer vision have resulted in major advancements in human body recognition technology. However, research in human body recognition is still in the early stages of the product lifecycle. Identifying the three-dimensional locations of the joints in the human body from pictures or videos is known as 3D posture estimation. Although it is widely used in areas like human motion analysis and robotics, it continues to be a difficult task due to challenges such as depth ambiguity and the scarcity of robust datasets. Over the past decade, numerous methods have been developed, many of which are based on deep learning, significantly improving the performance of existing benchmarks. A comprehensive literature review of this field is crucial for future development. However, in nowadays,more and more such research has mainly concentrated on traditional techniques, requirement for a comprehensive examination of tools based on deep learning. This paper delivers a thorough overview of current deep learning-based 3D pose estimation algorithms, outlining their advantages and limitations while providing a detailed understanding of the field. It also explores commonly used benchmark datasets and methods for analyzing human poses in unlabeled field images, providing a thorough comparative analysis. Finally, insights are provided to aid in the design of future models and algorithms.

Biomechanical and machine learning approaches to automating the identification of musical styles and emotions through human motion analysis

This study explores the intricate relationship between biomechanical movements and musical expression, focusing on the identification of musical styles and emotions. Violin performance is characterized by complex interactions between physical actions—such as bowing techniques, finger placements, and posture—and the resulting acoustic output. Recent advances in motion capture technology and sound analysis have enabled a more objective examination of these processes. However, the current literature frequently addresses biomechanics and acoustic features in isolation, lacking an integrated understanding of how physical movements translate into specific musical expressions. Machine Learning (ML), particularly Long Short-Term Memory (LSTM) networks, provides a promising avenue for bridging this gap. LSTM models are adept at capturing temporal dependencies in sequential data, making them suitable for analyzing the dynamic nature of violin performance. In this work, they have proposed a comprehensive model that combines biomechanical analysis with Mel-spectrogram-based LSTM modeling to automate the identification of musical styles and emotions in violin performances. Using motion capture systems, Inertial Measurement Units (IMUs), and high-fidelity audio recordings, we collected synchronized biomechanical and acoustic data from violinists performing various musical excerpts. The LSTM model was trained on this dataset to learn the intricate connections between physical movements and the acoustic features of each performance. Key findings from the study demonstrate the effectiveness of this integrated approach. The LSTM model achieved a validation accuracy of 92.5% in classifying musical styles and emotions, with precision, recall, and F1-score reaching 94.3%, 92.6%, and 93.4%, respectively, by the 100th epoch. The analysis also revealed strong correlations between specific biomechanical parameters, such as shoulder joint angle and bowing velocity, and acoustic features, like sound intensity and vibrato amplitude.

THÖR-MAGNI: A large-scale indoor motion capture recording of human movement and robot interaction

We present a new large dataset of indoor human and robot navigation and interaction, called THÖR-MAGNI, that is designed to facilitate research on social human navigation: for example, modeling and predicting human motion, analyzing goal-oriented interactions between humans and robots, and investigating visual attention in a social interaction context. THÖR-MAGNI was created to fill a gap in available datasets for human motion analysis and HRI. This gap is characterized by a lack of comprehensive inclusion of exogenous factors and essential target agent cues, which hinders the development of robust models capable of capturing the relationship between contextual cues and human behavior in different scenarios. Unlike existing datasets, THÖR-MAGNI includes a broader set of contextual features and offers multiple scenario variations to facilitate factor isolation. The dataset includes many social human–human and human–robot interaction scenarios, rich context annotations, and multi-modal data, such as walking trajectories, gaze-tracking data, and lidar and camera streams recorded from a mobile robot. We also provide a set of tools for visualization and processing of the recorded data. THÖR-MAGNI is, to the best of our knowledge, unique in the amount and diversity of sensor data collected in a contextualized and socially dynamic environment, capturing natural human–robot interactions.

UWB distance estimation errors in (non-)line of sight situations within the context of 3D analysis of human movement

Abstract Integrated Ultrawideband (UWB) and Magnetic Inertial Measurement Unit (MIMU) sensors are becoming popular for indoor localization applications, as a higher accuracy can be achieved than with just MIMU sensors. These integrated sensors could extend stability and accuracy in the field of 3D analysis of human movement (3D AHM) if they can deliver position estimates with an accuracy close to 1 cm. Achieving this high accuracy of 1 cm remains challenging, with most studies reporting position estimation errors around 5 cm, often due to Non-Line of Sight (NLOS) conditions and systematic UWB sensor distance estimation errors. Studying the distance error characteristic of UWB in situations of 3D AHM is essential to deal with these errors. While research on UWB distance errors in Line of Sight (LOS) and NLOS situations exists, few studies focus on the NLOS errors caused by the human body and were not in the relevant scenarios of 3D AHM. Therefore, this article examines UWB sensor performance and distance error characteristics in LOS and NLOS situations typical for the 3D AHM. Both the LOS and NLOS situations were studied in the typical 3D AHM distance range of 0.2 m to 2 m. The NLOS situations were studied first with a human subject as NLOS causing object and then with simulated human body segments (PVC pipes filled with water) of varying diameters. In LOS situations, consistent systematic bias errors were observed, along with incidental errors at specific positions in the room. In NLOS scenarios caused by the human and simulated body segments, a consistent and reproducible overestimation of distances was found. The reproducibility of these errors based on relative node and object positions suggests that systematic mitigation methods could significantly reduce errors, enabling more accurate and reproducible 3D human movement analysis.

Modified Star Excursion Balance test: Prototype development, correction, and reliability of measurements

ContextBiomechanical analyses of human movement require precise methods for quantifying measurements. The Modified Star Excursion Balance Test (mSEBT) is utilized by healthcare professionals and researchers to assess dynamic postural control. Despite its reliability, the manual application of mSEBT can introduce errors. Objectives(1) Develop a prototype using Arduino and a laser-sensor; (2) Evaluate and correct measurement errors using linear regression models; (3) Verify the reliability of the prototype under various configurations (height in relation to the ground, target distance, and test direction) of the mSEBT. Design: Observational Cross-sectional Case study. MethodsData were collected from a healthy participant using the mSEBT. The prototype was tested in various scenarios, encompassing different range directions, target distances, laser-sensor configurations, and heights from the ground. Linear regressions were applied to correct the values estimated by the prototype, and Intraclass Correlation coefficients (ICC) were calculated to validate the correction. ResultsThe prototype presented systematic errors in certain scenarios, especially at distances of 10–40 cm and heights of 3–5 cm in relation to the ground. After applying linear regressions and reliability tests, the prototype demonstrated excellent reliability in all evaluation scenarios (r > 0.90; p < 0.001). ConclusionIn summary, correction equations in the prototype algorithm derived from linear regressions of various test scenarios, which allows for the automation of measurements in the mSEBT clinical test. This advancement can enhance accuracy, efficiency, and clinical applicability, benefiting physiotherapists, physical educators, doctors, and biomechanics in monitoring dynamic postural control across diverse populations.

Novel calibration method for improved UWB sensor distance measurement in the context of application for 3D analysis of human movement

Integrated UWB and MIMU sensor systems have become popular for pedestrian tracking and indoor localization, since this facilitates data fusion that improves position estimation accuracy by exploiting the complementary nature of their error sources. Integrated UWB/MIMU sensors also have great potential in only on-body use for 3D analysis of human movement, as with MIMU sensors alone accurate direct estimation of (relative) body segment position is not possible. For this, a position estimation accuracy with errors smaller than 1 cm is deemed required. The lowest position estimating error with integrated UWB/MIMU systems, reported so far, is around 5 cm. The main accuracy limiting factors were found to be the systematic errors in the distance estimates from the UWB sensor. Multiple reported attempts to calibrate for these systematic errors failed to achieve the desired accuracy. This article presents a novel distance-bias calibration method that minimizes the residual systematic distance estimate errors using multiple sensors in a swarm configuration. Validation was performed against synthetic reference data and against reference data measured with an optical motion tracking system. Significantly reduced systematic distance estimate errors (≤0.5 cm) were found. These results promise to facilitate significantly better position estimates in future UWB/MIMU data fusion.

Generisch-Net: A Generic Deep Model for Analyzing Human Motion with Wearable Sensors in the Internet of Health Things.

The Internet of Health Things (IoHT) is a broader version of the Internet of Things. The main goal is to intervene autonomously from geographically diverse regions and provide low-cost preventative or active healthcare treatments. Smart wearable IMUs for human motion analysis have proven to provide valuable insights into a person's psychological state, activities of daily living, identification/re-identification through gait signatures, etc. The existing literature, however, focuses on specificity i.e., problem-specific deep models. This work presents a generic BiGRU-CNN deep model that can predict the emotional state of a person, classify the activities of daily living, and re-identify a person in a closed-loop scenario. For training and validation, we have employed publicly available and closed-access datasets. The data were collected with wearable inertial measurement units mounted non-invasively on the bodies of the subjects. Our findings demonstrate that the generic model achieves an impressive accuracy of 96.97% in classifying activities of daily living. Additionally, it re-identifies individuals in closed-loop scenarios with an accuracy of 93.71% and estimates emotional states with an accuracy of 78.20%. This study represents a significant effort towards developing a versatile deep-learning model for human motion analysis using wearable IMUs, demonstrating promising results across multiple applications.

Validity and Reliability of OpenPose-Based Motion Analysis in Measuring Knee Valgus during Drop Vertical Jump Test.

OpenPose-based motion analysis (OpenPose-MA), utilizing deep learning methods, has emerged as a compelling technique for estimating human motion. It addresses the drawbacks associated with conventional three-dimensional motion analysis (3D-MA) and human visual detection-based motion analysis (Human-MA), including costly equipment, time-consuming analysis, and restricted experimental settings. This study aims to assess the precision of OpenPose-MA in comparison to Human-MA, using 3D-MA as the reference standard. The study involved a cohort of 21 young and healthy adults. OpenPose-MA employed the OpenPose algorithm, a deep learning-based open-source two-dimensional (2D) pose estimation method. Human-MA was conducted by a skilled physiotherapist. The knee valgus angle during a drop vertical jump task was computed by OpenPose-MA and Human-MA using the same frontal-plane video image, with 3D-MA serving as the reference standard. Various metrics were utilized to assess the reproducibility, accuracy and similarity of the knee valgus angle between the different methods, including the intraclass correlation coefficient (ICC) (1, 3), mean absolute error (MAE), coefficient of multiple correlation (CMC) for waveform pattern similarity, and Pearson's correlation coefficients (OpenPose-MA vs. 3D-MA, Human-MA vs. 3D-MA). Unpaired t-tests were conducted to compare MAEs and CMCs between OpenPose-MA and Human-MA. The ICCs (1,3) for OpenPose-MA, Human-MA, and 3D-MA demonstrated excellent reproducibility in the DVJ trial. No significant difference between OpenPose-MA and Human-MA was observed in terms of the MAEs (OpenPose: 2.4° [95%CI: 1.9-3.0°], Human: 3.2° [95%CI: 2.1-4.4°]) or CMCs (OpenPose: 0.83 [range: 0.99-0.53], Human: 0.87 [range: 0.24-0.98]) of knee valgus angles. The Pearson's correlation coefficients of OpenPose-MA and Human-MA relative to that of 3D-MA were 0.97 and 0.98, respectively. This study demonstrated that OpenPose-MA achieved satisfactory reproducibility, accuracy and exhibited waveform similarity comparable to 3D-MA, similar to Human-MA. Both OpenPose-MA and Human-MA showed a strong correlation with 3D-MA in terms of knee valgus angle excursion.

Human movement monitoring system for classification of strength exercises and verification of their execution technique

Human movement analysis is critical to optimizing sports training and influencing exercise intensity and effectiveness. In the age of modern technology, more and more advanced systems are emerging to support coaches and expand the range of analysis performed. This article aims to verify that artificial intelligence, together with machine learning algorithms, can accurately classify exercises in a dynamic gym environment and effectively assess the correctness of their performance. For the initial analysis of movement, the Google MediaPipe Pose model was used, which was responsible for detecting the human silhouette and determining the coordinates of the position of critical joints. Based on these coordinates, the angles between each joint were calculated, and then their sequences were further analyzed. The sequences were analyzed using the following three algorithms: support vector machine (SVM), dense neural network, and LSTM recurrent network. As a result, the system based on recurrent LSTM networks achieved the best prediction efficiency of approximately 98%, enabling accurate exercise classification. Subsequently, verification of the activities' correctness was also carried out, and the system, based on recursive LSTM networks, again achieved the best efficiency, this time equal to 96% on average for all exercises. On this basis, it was concluded that the discussed approach enables practical analysis of human movement, which can significantly improve training methods and facilitate coaching work.

Conformal Self‐Powered Inertial Displacement Sensor with Geometric Optimization for In Situ Noninvasive Data Acquisition

AbstractThe growing focus on health management and smart technology advancements have propelled the use of wearable sensors in healthcare and human body motion analysis, particularly in preventing work‐related upper limb musculoskeletal disorders (MSDs) through guided exercises. However, most available wearable medical sensors are rigid, bulky, and incapable of in situ recognition of the comprehensive motion of human body. Here, a conformal self‐powered inertial displacement sensor (CSIDS) with geometric optimization for in situ noninvasive inertial data acquisition is proposed. Leveraging template‐assisted processing and COMSOL simulation, the geometric configurations of tribo‐layer materials, specifically focusing on the curvature of semicylindrical protrusions is systematically altered. This enhancement significantly improves the detection accuracy of joint range of motion. The features of shoulder joint bending angles and linear accelerations of the humerus are accurately captured using a deep learning model based on multilayer perceptron (MLP) networks, resulting in an exceptional recognition accuracy of 99.38% and 99.58%. Compared to traditional TENG wearable sensors that can only identify single metrics, CSIDS achieves a more comprehensive health assessment through inertial data detection. This system provides early warning for shoulder joint diseases, prevents MSDs, and extends to smart wearables for comprehensive joint health and ergonomic monitoring.

Machine Learning Approaches for 3D Motion Synthesis and Musculoskeletal Dynamics Estimation: A Survey.

The inference of 3D motion and dynamics of the human musculoskeletal system has traditionally been solved using physics-based methods that exploit physical parameters to provide realistic simulations. Yet, such methods suffer from computational complexity and reduced stability, hindering their use in computer graphics applications that require real-time performance. With the recent explosion of data capture (mocap, video) machine learning (ML) has started to become popular as it is able to create surrogate models harnessing the huge amount of data stemming from various sources, minimizing computational time (instead of resource usage), and most importantly, approximate real-time solutions. The main purpose of this paper is to provide a review and classification of the most recent works regarding motion prediction, motion synthesis as well as musculoskeletal dynamics estimation problems using ML techniques, in order to offer sufficient insight into the state-of-the-art and draw new research directions. While the study of motion may appear distinct to musculoskeletal dynamics, these application domains provide jointly the link for more natural computer graphics character animation, since ML-based musculoskeletal dynamics estimation enables modeling of more long-term, temporally evolving, ergonomic effects, while offering automated and fast solutions. Overall, our review offers an in-depth presentation and classification of ML applications in human motion analysis, unlike previous survey articles focusing on specific aspects of motion prediction.

Measuring shoulder range of motion to diagnose shoulder injury among weightlifters: a study in athletes with and without shoulder injury

Background: Weightlifting is one of the sports that contribute to Olympic medals for Indonesia. With the demand to contribute medals at the Olympics, a thorough screening is needed to evaluate athletes for possible risks of injury. Moreover, the lifting movements involve the joints, muscles, and skeleton. Therefore, the sport poses a very high risk of injury, especially to the shoulders. Knowing the athlete's Range of Motion (ROM) is important to see a good range of motion. This study aimed to observe ROM differences between injured and non-injured weightlifters. Method: An observational study was conducted on 16 athletes (10 male and 6 female) with the following characteristics for male (24.1 ± 6.7 years, 168.2 ± 2.5 cm, 83.0 ± 11.9 kg, 29.2 ± 4.0 kg/m2) and female (18.8 ± 1.6 years, 158.6 ± 3.8 cm, 65.2 ± 7.9 kg, 27.2 ± 4.5 kg/m2) participants. In data analysis, all athletes went through the screening stage. After that, the body composition and ROM of both arms were measured using InBody 270 and Humantrak 3D Human Movement Analysis, with statistical significance determined at a value &lt;0.05 using the Mann-Whitney test in Minitab. Result: The study found a significant difference in the range of motion of the shoulder between athletes who had injuries and those who did not have injuries. Conclusion: This research revealed a decreased range of motion in weightlifting athletes with injuries. Keywords: Weightlifter, Injury, Range of Motion, Shoulder

Enhancing Radar Image Classification with Autoencoder-CNN Hybrid System

The tracking, analysis, and classification of human movements can be crucial, particularly in areas such as elderly care, healthcare, and infant care. Typically, such tracking is done remotely with cameras. However, radar systems have emerged as significant methods and tools for these tasks due to their advantages such as privacy, wireless operation, and the ability to work through walls. By converting reflected radar signals from targets into images, human activities can be classified using powerful classification tools like deep learning. In this study, range-Doppler images of behind-the-wall human movements obtained with a radar system consisting of one transmitter and four receiver antennas were classified. Since the data collected from the four receiver antennas are in different positions, the collected reflection signals also differ. The signals collected with the range-time matrix content were divided into positive and negative parts, resulting in eight images from the four antennas. Instead of using all the data in CNN training, the images were first subjected to a reconstruction process with an autoencoder to reduce differences. As a result, it was observed that reconstructing the images with an autoencoder before classification with CNN increased the classification success. In conclusion, it was observed that the classification success of radar images can be increased by using a hybrid system with an autoencoder to reconstruct the images before classification with CNN.

ISB recommendations on the definition, estimation, and reporting of joint kinematics in human motion analysis applications using wearable inertial measurement technology

There is widespread and growing use of inertial measurement technology for human motion analysis in biomechanics and clinical research. Due to advancements in sensor miniaturization, inertial measurement units can be used to obtain a description of human body and joint kinematics both inside and outside the laboratory. While algorithms for data processing continue to improve, a lack of standard reporting guidelines compromises the interpretation and reproducibility of results, which hinders advances in research and development of measurement and intervention tools. To address this need, the International Society of Biomechanics approved our proposal to develop recommendations on the use of inertial measurement units for joint kinematics analysis. A collaborative effort that incorporated feedback from the biomechanics community has produced recommendations in five categories: sensor characteristics and calibration, experimental protocol, definition of a kinematic model and subject-specific calibration, analysis of joint kinematics, and quality assessment. We have avoided an overly prescriptive set of recommendations for algorithms and protocols, and instead offer reporting guidelines to facilitate reproducibility and comparability across studies. In addition to a conceptual framework and reporting guidelines, we provide a checklist to guide the design and review of research using inertial measurement units for joint kinematics.

A gait phase recognition method for obstacle crossing based on multi-sensor fusion

The analysis of human motion patterns and gait phase detection holds significant importance for both human motion analysis and precise control of exoskeleton devices. Presently, research on gait during obstacle crossing primarily focuses on the medical rehabilitation domain, with limited studies concerning gait phase analysis during obstacle crossing. This study conducts a detailed analysis of gait data during obstacle crossing and proposes a CNN-PCA-LSTM algorithm that fuses multi-sensor data to accurately identify gait phases when crossing obstacles in the horizontal direction. A wearable multi-sensor data acquisition system was designed, utilizing flexible capacitive pressure sensors to collect pressure data from the soles of both feet and IMUs to gather foot motion data. Kinematic simulation analyses of human obstacle-crossing motions were performed using Anybody software to determine various gait phases, validating the efficacy of the simulation model. Subsequently, one-dimensional and two-dimensional convolutional neural networks were separately employed to extract features from sole pressure data and foot motion data. Principal Component Analysis (PCA) was utilized to reduce the dimensionality of these feature datasets, which were then inputted into LSTM networks. Finally, the Softmax function was employed for the classification and recognition of gait phases when crossing obstacles in the horizontal direction. The experimental results indicate that employing the CNN-PCA-LSTM algorithm integrating multiple sensor data achieves a recognition accuracy of 97.91 % for identifying gait phases during horizontal obstacle crossing.

Versatile Patterning of Liquid Metal via Multiphase 3D Printing.

This paper presents a scalable and straightforward technique for the immediate patterning of liquid metal/polymer composites via multiphase 3D printing. Capitalizing on the polymer's capacity to confine liquid metal (LM) into diverse patterns. The interplay between distinctive fluidic properties of liquid metal and its self-passivating oxide layer within an oxidative environment ensures a resilient interface with the polymer matrix. This study introduces an inventive approach for achieving versatile patterns in eutectic gallium indium (EGaIn), a gallium alloy. The efficacy of pattern formation hinges on nozzle's design and internal geometry, which govern multiphase interaction. The interplay between EGaIn and polymer within the nozzle channels, regulated by variables such as traverse speed and material flow pressure, leads to periodic patterns. These patterns, when encapsulated within a dielectric polymer polyvinyl alcohol (PVA), exhibit an augmented inherent capacitance in capacitor assemblies. This discovery not only unveils the potential for cost-effective and highly sensitive capacitive pressure sensors but also underscores prospective applications of these novel patterns in precise motion detection, including heart rate monitoring, and comprehensive analysis of gait profiles. The amalgamation of advanced materials and intricate patterning techniques presents a transformative prospect in the domains of wearable sensing and comprehensive human motion analysis.

Electromyography as a tool to motion analysis for people with Amyotrophic Lateral Sclerosis: A protocol for a systematic review.

Biomechanical analysis of human movement plays an essential role in understanding functional changes in people with Amyotrophic Lateral Sclerosis (ALS), providing information on muscle impairment. Studies suggest that surface electromyography (sEMG) may be able to quantify muscle activity, identify levels of fatigue, assess muscle strength, and monitor variation in limb movement. In this article, a systematic review protocol will analyze the psychometric properties of the sEMG regarding the clinical data on the skeletal muscles of people with ALS. This protocol uses the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodological tool. A specific field structure was defined to reach each phase. Nine scientific databases (PubMed, Web of Science, Embase, Elsevier, IEEE, Google Scholar, SciELO, PEDro, LILACS E CENTRAL) were searched. The framework developed will extract data (i.e. study information, sample information, sEMG information, intervention, and outcomes) from the selected studies using a rigorous approach. The data will be described quantitatively using frequency and trend analysis methods, and heterogeneity between the included studies will be assessed using the I2 test. The risk of bias will be summarized using the most recent prediction model risk of bias assessment tool. Be sure to include relevant statistics here, such as sample sizes, response rates, P values or Confidence Intervals. Be specific (by stating the value) rather than general (eg, "there were differences between the groups"). This protocol will map out the construction of a systematic review that will identify and synthesize the advances in movement analysis of people with ALS through sEMG, using data extracted from articles.

Kinematic synergies show good consistency when extracted with a low-cost markerless device and a marker-based motion tracking system

Recently, markerless tracking systems, such as RGB-Depth cameras, have spread to overcome some of the limitations of the gold standard marker-based tracking systems. Although these systems are valuable substitutes for human motion analysis, as they guarantee higher flexibility, faster setup time and lower costs, their tracking accuracy is lower with respect to marker-based systems. Many studies quantified the error made by markerless systems in terms of body segment length estimation, articular angles, and biomechanics, concluding that they are appropriate for many clinical applications related to motion analysis. We propose an innovative approach to compare a markerless tracking system (Kinect V2) with a gold standard marker-based system (Vicon), based on motor control assessment. We quantified kinematic synergies from the tracking data of fifteen participants performing multi-directional upper limb movements. Kinematic synergy analysis decomposes the kinematic data into a reduced set of motor primitives that describe how the central nervous system coordinates motion at spatial and temporal level. Synergies were extracted with the recently released mixed-matrix factorization algorithm. Four synergies were extracted from both marker-based and markerless datasets and synergies were grouped in 6 clusters for each dataset. Cosine similarity in each cluster was ⩾0.60 in both systems, showing good consistency of synergies. Good matching was found between synergies extracted from markerless and from marker-based data, with a cosine similarity between matched synergies ⩾0.60 in 5 out 6 synergies. These results showed that the markerless sensor can be feasible for kinematic synergy analysis for gross movements, as it correctly estimates the number of synergies and in most cases also their spatial and temporal organization.

Визуализация и классификация движений человека на основе скелетной структуры: Нейросетевой подход к анализу спортивных упражнений и сравнение методологий

The authors of the paper review and compare different existing approaches to Human Action Recognition (HAR), analyze the advantages and disadvantages of platforms for extracting human skeletal structure from video stream, and evaluate the importance of visual representation in the motion analysis process. This paper presents an example implementation of one of the approaches to HAR based on the use of interpretability and visual expressiveness inherent in skeletal structures. In this work, an ad hoc network with Long Short-Term Memory (LSTM) for human activity classification is designed and implemented, which has been trained and tested in the domain of sports exercises. LSTM incorporation of memory cells and gating mechanisms not only mitigates the vanishing gradient problem but also enables LSTMs to selectively retain and utilize relevant information over extended sequences, making them highly effective in tasks with complex temporal dependencies. The problem with a fading gradient is quite common in deep neural networks and is that if the error is back propagated during the training of the network, the gradient can decrease strongly as it travels through the layers of the network to the initial layers. This can lead to the fact that the weights in the initial layers are practically not updated, which makes training of these layers impossible or slows down its process. The resulting solution can be used to create a real-time virtual fitness assistant. The resulting solution can be used to create a real-time virtual fitness assistant. In addition, this approach will make it possible to create interactive training applications with visualization of human skeletal structure, motion analysis and monitoring systems in the field of medicine and rehabilitation, as well as for the development of security systems with access control based on the analysis of visual data on the movement of human body parts.

The virtual pivot point concept improves predictions of ground reaction forces.

Ground reaction forces (GRFs) are essential for the analysis of human movement. To measure GRFs, 3D force plates that are fixed to the floor are used with large measuring ranges, excellent accuracy and high sample frequency. For less dynamic movements, like walking or squatting, portable 3D force plates are used, while if just the vertical component of the GRFs is of interest, pressure plates or in-shoe pressure measurements are often preferred. In many cases, however, it is impossible to measure 3D GRFs, e.g., during athletic competitions, at work or everyday life. It is still challenging to predict the horizontal components of the GRFs from kinematics using biomechanical models. The virtual pivot point (VPP) concept states that measured GRFs during walking intercept in a point located above the center of mass, while during running, the GRFs cross each other at a point below the center of mass. In the present study, this concept is used to compare predicted GRFs from measured kinematics with measured 3D-GRFs, not only during walking but also during more static movements like squatting and inline lunge. To predict the GRFs a full-body biomechanical model was used while gradually changing the positions of the VPP. It is shown that an optimal VPP improves the prediction of GRFs not only for walking but also for inline lunge and squats.