Posture Recognition Research Articles

Seeking optimal and explainable deep learning models for inertial-based posture recognition

Deep Learning (DL) models, widely used in several domains, are often applied for posture recognition. This work researches five DL architectures for posture recognition: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Transformer, hybrid CNN-LSTM, and hybrid CNN-Transformer. Agriculture and construction working postures were addressed as use cases, by acquiring an inertial dataset during the simulation of their typical tasks in circuits. Since model performance greatly depends on the choice of the hyperparameters, a grid search was conducted to find the optimal hyperparameters. An extensive analysis of the hyperparameter combinations’ effects is presented, identifying some general tendencies. Moreover, to unveil the black-box DL models, we applied the Gradient-weighted Class Activation Mapping (Grad-CAM) explainability method on CNN’s outputs to better understand the model’s decision-making, in terms of the most important sensors and time steps for each window output. Innovative hybrid architectures combining CNN and LSTM or Transformer encoder were implemented, by using the convolution feature maps as LSTM’s or Transformer’s inputs and fusing both subnetworks’ outputs with weights learned during the training. All architectures successfully recognized the eight posture classes, with the best model of each architecture exceeding 91.5% F1-score in the test. A top F1-score of 94.33%, with an inference time of just 0.29 ms (in a regular laptop), was achieved by a hybrid CNN-Transformer.

Skin-core-fiber-based fabric integrated with pressure sensing and deep learning for posture recognition

The aggravation of subhealth caused by contemporary lifestyles boosts the innovation of sensing systems that can acquire and analyze physiological signals in a real-time, high-efficiency, and even smart way. However, reliable and universal fabrication for these smart sensing units in a bulk or film form remains challenging and thus restricts their widespread applications. Herein, we developed a smart fabric system for accurate real-time sitting posture recognition based on the combination of wet-spun skin-core aerogel fibers, a signal conversion module, and deep learning model. The sensing fibers show a high sensitivity of 23.59 kPa-1, fast response/recovery time of 45/40 ms, and excellent stability over 10,000 cycles, suggesting good pressure-sensing ability. After being assembled them with a signal conversion module, the resultant fabric sensing system could capture physiological signals from human motions and sitting posture in a precise and real-time fashion. Such systems function well even integrated into different parts of clothing and provide long-term wearing comfort. When introducing a deep learning model, 98% sensing accuracy is demonstrated. This fabric sensing system not only provides reliable solutions to subhealth status monitoring and unhealth lifestyle correction, but also inspires the mass production of next-generation smart textiles.

Distance and Angle Insensitive Radar-Based Multi-Human Posture Recognition Using Deep Learning

Human posture recognition has a wide range of applicability in the detective and preventive healthcare industry. Recognizing posture through frequency-modulated continuous wave (FMCW) radar poses a significant challenge as the human subject is static. Unlike existing radar-based studies, this study proposes a novel framework to extract the postures of two humans in close proximity using FMCW radar point cloud. With radar extracted range, velocity, and angle information, point clouds in the Cartesian domain are retrieved. Afterwards, unsupervised clustering is implemented to segregate the two humans, and finally a deep learning model named DenseNet is applied to classify the postures of both human subjects. Using four base postures (namely, standing, sitting on chair, sitting on floor, and lying down), ten posture combinations for two human scenarios are classified with an average accuracy of 96%. Additionally, using the centroid information of human clusters, an approach to detect and classify overlapping human participants is also introduced. Experiments with five posture combinations of two overlapping humans yielded an accuracy of above 96%. The proposed framework has the potential to offer a privacy-preserving preventive healthcare sensing platform for an elderly couple living alone.

Virtual Yoga Assistant using Machine Learning and Artificial Intelligence

In recent years, yoga has become an integral part of life for people worldwide. This growing interest has created a demand for scientific analysis of yoga postures. Pose detection techniques offer a promising approach to identify and assist people in performing yoga poses with greater accuracy. However, posture recognition remains a challenging task due to the limited availability of datasets and the difficulty of real-time posture detection. To address this, a large dataset has been developed with over 5,500 images representing ten different yoga poses. A tf-pose estimation algorithm is used to create a skeletal overlay on each image in real time, drawing the human body’s skeleton to extract joint angles. These joint angles serve as features for training various machine learning models. The dataset is split, with 80% used for training and 20% for testing. This approach has been tested on multiple machine learning classification models, achieving an accuracy of 99.04% with a Random Forest Classifier.

A Wushu Leg Gesture Recognition Algorithm Based on Random Forest and Bone Feature Extraction (RF-SFE)

Addressing the challenges of complex movement trajectories and rapid action changes in martial arts performances, this study introduces a novel posture recognition algorithm based on Random Forest and Skeletal Feature Extraction (RF-SFE) for martial arts leg movements. Unlike traditional posture recognition methods that struggle with accuracy, RF-SFE aims to provide intelligent analysis of training postures to assist practitioners in efficient training. The algorithm initially employs advanced skeletal feature extraction techniques to identify and articulate the relative positions and movements specific to martial arts. These extracted spatial features enhance the flexibility in modeling the unique dynamics of martial arts. Subsequently, Random Forest classification is utilized to categorize different leg movements, leveraging its strength in handling high-dimensional data and providing robust classification. Comparative experiments on diverse martial arts posture datasets demonstrate a significant improvement in recognition rates over baseline methods. This validates the effectiveness of the RF-SFE method in recognizing martial arts postures, offering scientific guidance for practitioners’ training regimens.

DeepLPos: a comprehensive hybrid deep learning model for lying position recognition using a tactile sensor array system

Abstract Unpredictable limb movements or turning motions can significantly disrupt the accurate extraction of physiological signals, such as respiratory and heart rates. In clinical environments, reliable detection of lying positions is crucial for continuous patient monitoring, particularly during sleep. In this paper, a smart sleeping position recognition system is proposed, which employs a tactile pressure sensor array based on the unique structure of ‘the electrostatic double-layer capacitors’. The sensor array, comprising 64 rows and 32 columns (2048 nodes), captures four types of healthy lying positions using an 8-bit AD module. Despite challenges arising from limited experimental samples for accurate training, we propose DeepLPos, a hybrid deep learning approach combining generative adversarial networks and the you only look once network. To tackle the differentiation challenge between supine and prone positions, we introduce an SPD Conv attention module to enhance the resolution of detailed descriptions in pressure images. The model is further pruned to optimize both structure and parameters, enabling efficient real-time detection. Evaluated on the SLP dataset, the proposed system achieves an accuracy of 97.5% with a real-time processing speed of 0.069 s per frame, demonstrating its potential for practical, high-precision measurement and monitoring applications in healthcare.

Human-machine interaction with intelligent machine learning algorithms for smart classrooms

ABSTRACT The modern technology of learning and teaching in education has moved on to smart classes. The introduction of artificial intelligence (AI) in the education sector has essentially taken over traditional classrooms and transformed how education leads to admiration. Various data types, such as videos, images, sound, text, and biometric data like eye-tracking potentially captured through smart classrooms, pose new issues and challenges for artificial intelligence algorithms. Hence, this paper suggests the Machine Learning Assisted Intelligent Human-Machine Interaction (MLIHMI) model for the smart classroom monitoring system. The real-time interactive dimension signifies the capability to support the Smart classroom’s teaching and human-computer interactions, including smooth interaction, convenient operation, and interactive monitoring. This paper used ML and HMI models to introduce a new effective database for online learning and smart classroom environments utilising the learners’ body postures, facial expressions, and hand gestures. The numerical findings show that the suggested system achieves less error rate of 10.1% and enhances the accuracy rate of 96.5%, with the highest classification ratio of 95.6%, prediction rate of 93.6%, the detection rate of 97.3%, and engagement rate of 81.3% in student gesture and postures recognition and the identification of facial expressions with the high performance of 93.3% when compared to other existing models.

Bridging the Appearance Domain Gap in Elderly Posture Recognition with YOLOv9

Accurate posture detection of elderly people is crucial to improve monitoring and provide timely alerts in homes and elderly care facilities. Human posture recognition is experiencing a great leap in performance with the incorporation of deep neural networks (DNNs) such as YOLOv9. Unfortunately, DNNs require large amounts of annotated data for training, which can be addressed by using virtual reality images. This paper investigates how to address the appearance domain that lies between synthetic and natural images. Therefore, four experiments (VIRTUAL–VIRTUAL; HYBRID–VIRTUAL; VIRTUAL–REAL; and HYBRID–REAL) were designed to assess the feasibility of recognising the postures of virtual or real elderly people after training with virtual and real images of elderly people. The results show that YOLOv9 achieves the most outstanding accuracy of 98.41% in detecting and discriminating between standing, sitting, and lying postures after training on a large number of virtual images complemented by a much smaller number of real images when testing on real images.

Deep Learning Models for Anatomical Location Classification in Esophagogastroduodenoscopy Images and Videos: A Quantitative Evaluation with Clinical Data.

During gastroscopy, accurately identifying the anatomical locations of the gastrointestinal tract is crucial for developing diagnostic aids, such as lesion localization and blind spot alerts. This study utilized a dataset of 31,403 still images from 1000 patients with normal findings to annotate the anatomical locations within the images and develop a classification model. The model was then applied to videos of 20 esophagogastroduodenoscopy procedures, where it was validated for real-time location prediction. To address instability of predictions caused by independent frame-by-frame assessment, we implemented a hard-voting-based post-processing algorithm that aggregates results from seven consecutive frames, improving the overall accuracy. Among the tested models, InceptionV3 demonstrated superior performance for still images, achieving an F1 score of 79.79%, precision of 80.57%, and recall of 80.08%. For video data, the InceptionResNetV2 model performed best, achieving an F1 score of 61.37%, precision of 73.08%, and recall of 57.21%. These results indicate that the deep learning models not only achieved high accuracy in position recognition for still images but also performed well on video data. Additionally, the post-processing algorithm effectively stabilized the predictions, highlighting its potential for real-time endoscopic applications. This study demonstrates the feasibility of predicting the gastrointestinal tract locations during gastroscopy and suggests a promising path for the development of advanced diagnostic aids to assist clinicians. Furthermore, the location information generated by this model can be leveraged in future technologies, such as automated report generation and supporting follow-up examinations for patients.

The application of improved DTW algorithm in sports posture recognition

Sports posture recognition plays a crucial role in modern sports science and training. Posture recognition and analysis plays a positive role in improving sports quality and ensuring sports safety. However, existing recognition technologies still have poor recognition and accuracy in large amounts of posture data. Therefore, to further improve the performance of the existing posture recognition techniques, this study assumes that postures during movement can be effectively represented through the time series of skeletal key points, and the local similarity of these postures can be captured through the Dynamic Time Warping (DTW) algorithm. Based on this assumption, the existing DTW algorithm is improved by introducing the K-Nearest Neighbor (KNN) algorithm and combining it with Principal Component Analysis (PCA) for feature dimensionality reduction. A novel algorithmic model for postures recognition is proposed. The experimental results showed that the improved algorithm performed well in postures recognition rate and accuracy. Especially, when the k value was 5, the recognition rate reached up to 89%, and the accuracy reached 87%. Compared with the existing algorithm, the improved KNN-DTW algorithm has significant improvement in accuracy and computational efficiency. In summary, the new algorithm shows significant advantages in terms of accuracy and stability, providing a powerful tool for the analysis of athletic postures in the field of sports. Meanwhile, this research result has important application prospects in fields such as sports training, sports medicine, and virtual reality.

Laser-inertial tightly coupled SLAM system for indoor degraded environments

Purpose To address the issues of low localization and mapping accuracy, as well as map ghosting and drift, in indoor degraded environments using light detection and ranging-simultaneous localization and mapping (LiDAR SLAM), a real-time localization and mapping system integrating filtering and graph optimization theory is proposed. By incorporating filtering algorithms, the system effectively reduces localization errors and environmental noise. In addition, leveraging graph optimization theory, it optimizes the poses and positions throughout the SLAM process, further enhancing map accuracy and consistency. The purpose of this study resolves common problems such as map ghosting and drift, thereby achieving more precise real-time localization and mapping results. Design/methodology/approach The system consists of three main components: point cloud data preprocessing, tightly coupled inertial odometry based on filtering and backend pose graph optimization. First, point cloud data preprocessing uses the random sample consensus algorithm to segment the ground and extract ground model parameters, which are then used to construct ground constraint factors in backend optimization. Second, the frontend tightly coupled inertial odometry uses iterative error-state Kalman filtering, where the LiDAR odometry serves as observations and the inertial measurement unit preintegration results as predictions. By constructing a joint function, filtering fusion yields a more accurate LiDAR-inertial odometry. Finally, the backend incorporates graph optimization theory, introducing loop closure factors, ground constraint factors and odometry factors from frame-to-frame matching as constraints. This forms a factor graph that optimizes the map’s poses. The loop closure factor uses an improved scan-text-based loop closure detection algorithm for position recognition, reducing the rate of environmental misidentification. Findings A SLAM system integrating filtering and graph optimization technique has been proposed, demonstrating improvements of 35.3%, 37.6% and 40.8% in localization and mapping accuracy compared to ALOAM, lightweight and ground optimized lidar odometry and mapping and LiDAR inertial odometry via smoothing and mapping, respectively. The system exhibits enhanced robustness in challenging environments. Originality/value This study introduces a frontend laser-inertial odometry tightly coupled filtering method and a backend graph optimization method improved by loop closure detection. This approach demonstrates superior robustness in indoor localization and mapping accuracy.

Decoding the silent language: A framework for advanced visual emotion recognition to enhance worker well-being on IIoT-enabled manufacturing

The Industrial Internet of Things (IIoT), pivotal in the Industry 4.0 revolution, has transformed manufacturing, particularly assembly lines, by enabling real-time process monitoring and data analysis to enhance productivity. Yet, its potential impact on worker well-being has been less explored. As the National Association of Manufacturers anticipates an opening of 2.1 million unfilled jobs by 2030, the necessity for frameworks that not only understand but also improve employee well-being becomes apparent. We introduce a novel, non-intrusive visual emotion recognition framework that incorporates facial emotion recognition (FER) and emotional body posture recognition (EBPR) to assess workers’ psychological states. This initiative aims to mitigate human errors and accidents linked to stress or high cognitive load, thereby enhancing the manufacturing environment. Strategies such as optimized break time planning, line balancing, procedure modifications, and improved training are proposed to ensure both employee well-being and stakeholder benefits. This framework indirectly supports the industry’s efforts to address the looming challenge of unfilled jobs by fostering a supportive work environment that enhances job satisfaction, potentially making positions more appealing and retaining valuable talent. By bridging the gap between technological advancements and human-centric considerations, our approach underscores the significance of integrating IIoT solutions with employee welfare initiatives to address both productivity and workforce sustainability challenges.

Validity Of An Image-based Posture Recognition System For Monitoring Daily Physical Activity

Validity Of An Image-based Posture Recognition System For Monitoring Daily Physical Activity

Classification of Sleeping Position Using Enhanced Stacking Ensemble Learning.

Sleep position recognition plays a crucial role in enhancing individual sleep quality and addressing sleep-related disorders. However, the conventional non-invasive technology for recognizing sleep positions tends to be limited in its widespread application due to high production and computing costs. To address this issue, an enhanced stacking model is proposed based on a specific air bag mattress. Firstly, the hyperparameters of the candidate base model are optimized using the Bayesian optimization algorithm. Subsequently, the entropy weight method is employed to select extreme gradient boosting (XGBoost), support vector machine (SVM), and deep neural decision tree (DNDT) as the first layer of the enhanced stacking model, with logistic regression serving as the meta-learner in the second layer. Comparative analysis with existing machine learning techniques demonstrates that the proposed enhanced stacking model achieves higher classification accuracy and applicability.

Using posture recognition algorithms based on machine learning to identify senior health

Faced with the situation that the elderly people at home have dangerous behaviors, the study explores various aspects of motion target detection, real-time target tracking and behavioral pose recognition and classification, using behavioral poses in videos as samples. To tackle the challenges in detecting motion targets, a target detection method based on Gaussian mixture model (GMM) and four frame difference method is proposed; A tracking technique incorporating Kalman filter (KF) is investigated to trail the behavioral changes of the elderly in actual time. A seven-layer convolutional neural network (CNN) is constructed to face the problem of inaccurate behavioral pose recognition. Through relevant experimental analyses, the outcomes show that the increased GMM detection way has a complete profile and the accuracy is significantly improved. The KF target tracking technique can trail the object trajectory in actual time and steadily, with the smallest trailing error value of 0.19. The classification accuracy of the CNN pose recognition model is 95.87%, and the pose classification time is 27 seconds. Its performance is superior to the mean shift algorithm, particle filter algorithm, and Cam Shift algorithm in all aspects. When applied in practice, it can accurately identify whether the elderly’s behavior is abnormal and ensure their daily health.

YOLOv5-POS: research on cabbage pose prediction method based on multi-task perception technology.

Accurate and rapid identification of cabbage posture is crucial for minimizing damage to cabbage heads during mechanical harvesting. However, due to the structural complexity of cabbages, current methods encounter challenges in detecting and segmenting the heads and roots. Therefore, exploring efficient cabbage posture prediction methods is of great significance. This study introduces YOLOv5-POS, an innovative cabbage posture prediction approach. Building on the YOLOv5s backbone, this method enhances detection and segmentation capabilities for cabbage heads and roots by incorporating C-RepGFPN to replace the traditional Neck layer, optimizing feature extraction and upsampling strategies, and refining the C-Seg segmentation head. Additionally, a cabbage root growth prediction model based on Bézier curves is proposed, using the geometric moment method for key point identification and the anti-gravity stem-seeking principle to determine root-head junctions. It performs precision root growth curve fitting and prediction, effectively overcoming the challenge posed by the outer leaves completely enclosing the cabbage root stem. YOLOv5-POS was tested on a multi-variety cabbage dataset, achieving an F1 score of 98.8% for head and root detection, with an instance segmentation accuracy of 93.5%. The posture recognition model demonstrated an average absolute error of 1.38° and an average relative error of 2.32%, while the root growth prediction model reached an accuracy of 98%. Cabbage posture recognition was completed within 28 milliseconds, enabling real-time harvesting. The enhanced model effectively addresses the challenges of cabbage segmentation and posture prediction, providing a highly accurate and efficient solution for automated harvesting, minimizing crop damage, and improving operational efficiency.

Sitting Posture Recognition Systems: Comprehensive Literature Review and Analysis

Sitting posture recognition systems have gained significant attention due to their potential applications in various domains, including healthcare, ergonomics, and human-computer interaction. This paper presents a comprehensive literature review and analysis of existing sitting posture recognition systems. Through an extensive examination of relevant research articles and conference papers, we identify and analyze the underlying technologies, methodologies, datasets, performance metrics, and applications associated with these systems. The review encompasses both traditional methods, such as vision-based approaches and sensor-based techniques, as well as emerging technologies such as machine learning and deep learning algorithms. Additionally, we examine the challenges, constraints, and future trends in the field of sitting posture recognition systems. Researchers, practitioners, and policymakers who want to comprehend the most recent developments and latest trends in sitting posture recognition technology will find great value in this study.

American Football Play Type and Player Position Recognition

American Football Play Type and Player Position Recognition

GFI-YOLOv8: Sika Deer Posture Recognition Target Detection Method Based on YOLOv8.

As the sika deer breeding industry flourishes on a large scale, accurately assessing the health of these animals is of paramount importance. Implementing posture recognition through target detection serves as a vital method for monitoring the well-being of sika deer. This approach allows for a more nuanced understanding of their physical condition, ensuring the industry can maintain high standards of animal welfare and productivity. In order to achieve remote monitoring of sika deer without interfering with the natural behavior of the animals, and to enhance animal welfare, this paper proposes a sika deer individual posture recognition detection algorithm GFI-YOLOv8 based on YOLOv8. Firstly, this paper proposes to add the iAFF iterative attention feature fusion module to the C2f of the backbone network module, replace the original SPPF module with AIFI module, and use the attention mechanism to adjust the feature channel adaptively. This aims to enhance granularity, improve the model's recognition, and enhance understanding of sika deer behavior in complex scenes. Secondly, a novel convolutional neural network module is introduced to improve the efficiency and accuracy of feature extraction, while preserving the model's depth and diversity. In addition, a new attention mechanism module is proposed to expand the receptive field and simplify the model. Furthermore, a new pyramid network and an optimized detection head module are presented to improve the recognition and interpretation of sika deer postures in intricate environments. The experimental results demonstrate that the model achieves 91.6% accuracy in recognizing the posture of sika deer, with a 6% improvement in accuracy and a 4.6% increase in mAP50 compared to YOLOv8n. Compared to other models in the YOLO series, such as YOLOv5n, YOLOv7-tiny, YOLOv8n, YOLOv8s, YOLOv9, and YOLOv10, this model exhibits higher accuracy, and improved mAP50 and mAP50-95 values. The overall performance is commendable, meeting the requirements for accurate and rapid identification of the posture of sika deer. This model proves beneficial for the precise and real-time monitoring of sika deer posture in complex breeding environments and under all-weather conditions.

Scalable wet-spinning multilevel anisotropic structured PVDF fibers enhanced with cellulose nanocrystals-exfoliated MoS2 for high-performance piezoelectric textiles

The high-performance fabric-based piezoelectric nanogenerators (PENGs) is the ideal choice for the next generation of wearable electronics. However, large-scale fabrication of high-performance fabric-based PENGs remains a great challenge. In this study, cellulose nanocrystals (CNCs) exfoliated MoS2 (C@M) are integrated into the poly (vinylidene fluoride) (PVDF) matrix, which are successfully fabricated PVDF/C@M fibers (PCF) by wet-spinning and post-drawing processes. Accompanied by the stress-induced anisotropic alignment of C@M and PVDF molecular chains during post-drawing, the interfacial interactions between C@M and PVDF dipoles greatly promote the formation of the electroactive β-phase of PVDF as well as the orientation of the dipoles. The PCF demonstrate excellent mechanical performance with a tensile strength of 301 MPa and toughness of 68.5 MJ·m−3. Subsequently, the PCF are woven into fabrics and subsequently assembled into a flexible PENG (PCF-PENG), exhibiting excellent piezoelectric coefficient (d33 = 40.3 pC·N−1), power density of 56.25 μW·cm−2. Furthermore, scenarios such as energy harvesting, motion detection, and posture recognition of the PCF-PENG are demonstrated, providing a feasible proposal for the large-scale fabrication of electronic textiles.