Martial arts action recognition based on improved method of human pose estimation using stacked hourglass network

In order to improve the effective guidance of martial arts movements, a decomposition method of martial arts movements based on image recognition is proposed. First of all, visual image target sampling of martial arts action is performed, and most of the noise background is eliminated through morphological gradient operation. Then, the contour edge of the human body is obtained, the contour edge of each frame of the video is extracted, and the accumulation is realized in the same image. Using accumulation, the edge image calculates the grid-based HOG to obtain the image action feature vector. Secondly, using the improved dynamic time warping theory combined with the characteristics of the angle change of each joint under the action time sequence, the joint change sequence among various martial arts movements can be identified in order to realize the decomposition process of martial arts actions based on image recognition. The experimental results show that the use of image recognition can effectively decompose martial arts movements.

The absence of local features and global shape constraints severely limits the performance of the hourglass network for facial landmark detection in unconstrained environments. Moreover, diverse feature types and scales may result in low accuracy. This paper proposes a probability‐guided hourglass network to enhance the shape constraints for robust facial landmark detection. Firstly, a multi‐scale pre‐processing module is designed to extract features at different scales. Secondly, based on the heatmaps generated by the stacked hourglass network, the coarse localizations are obtained, while the probability maps are generated with local features. Finally, a probability‐based boundary regression method is proposed and the hausdorff distance is modified as the loss function to constrain the feature shape. Adaptive weights are also added to the loss function, which can help relieve the data imbalance problem. Subjective and objective experimental results on the challenging datasets show that this method outperforms the state‐of‐the‐arts on unconstrained conditions.

Pose estimation has been a hot topic in the field of machine vision in recent years. Animals exist widely in nature, and the analysis of their shape and movement is important in many fields and industries. In the pose estimation task, to improve the detection accuracy, the existing models often need to consume a lot of computing and memory resources. Therefore, it is a key problem for the pose estimation methods to carry out a lightweight model and reduce the computational overhead on the premise of ensuring model accuracy. In this paper, we focus on the structure of the convolutional neural network in animal pose estimation, construct a lightweight and efficient stacked hourglass network model oriented to optimize the balance of model computation and accuracy, and implement the application algorithm design based on it. Aiming at the problem of large parameters in depthwise convolutional neural networks, a lightweight residual module is proposed, that is, based on the lightweight efficient channel attention improved conditional channel-weighted method (ICCW-Bottle), thereby reducing the weight of the network and obtaining the feature information of different scales. Given the problem that a large amount of feature information is easily lost after the network pooling operation, a lightweight dual-branch fusion module is proposed that fully integrates high-level semantic information and low-level detailed features under the condition of a small number of parameters. Finally, the same as the CC-SSL method: the model is trained jointly using synthetic and real animal datasets, but the CC-SSL method does not take into account the computational power of the model, which consumes a lot of time and memory to run. Through experiments, it is known that compared with the CC-SSL method, the PCK@0.05 of this method is increased by 5.5% on the TigDog dataset. The model in this paper reduces the number of parameters and calculations of the network while ensuring less information loss and model accuracy. The ablation experiment verifies the advancement and effectiveness of the overall network.

Traditional research has difficulty in precisely evaluating students’ movements in martial arts teaching, and cannot provide targeted teaching suggestions based on students’ characteristics in personalized guidance. This study constructs a movement evaluation and personalized guidance system for martial arts teaching by integrating the EfficientNet model with a recommendation algorithm based on collaborative filtering. First, EfficientNet is used to perform multi-category classification evaluation on martial arts movements, and the cross-entropy loss function is used to improve the classification accuracy. Then, combined with the students’ historical movement performance, a collaborative filtering algorithm based on the Pearson correlation coefficient is used to recommend personalized training plans for each student. Grid search is used to optimize hyperparameters such as user neighborhood size and similarity measurement to ensure the accuracy and diversity of recommendation results. Experimental data shows that after systematic teaching evaluation and personalized guidance, the average score of students increases by about 15.84%, and the average self-assessed satisfaction (1-5) of students after the recommended content reaches 4.1. The system can accurately evaluate students’ movement performance and provide targeted training suggestions based on individual needs.

Human pose estimation, a research hotspot of computer vision, has been a key and hard task. Recently, significant progress has been achieved using the newly emergent deep convolutional neural network technique. Many methods first apply the bottom-up and top-down structure, which uses first downsampling and then upsampling. This structure can capture multi-scale feature but lead to loss of information due to downsampling. In this work, we propose a Dilated Convolutional Module (DCM) based on dilated convolution and skip connections. The proposed module, which aims to reduce the loss of information, can expand receptive field and reduce the times of downsampling. Using DCMs as building blocks, this work proposes a novel convolutional network named Dilated Hourglass Network, based on Stacked Hourglass Network. The proposed network is compared with the Stacked Hourglass Network on the standard benchmarks MPII dataset for human pose estimation, which shows that the proposed network achieves a notable progress.

While facial biometrics has been widely used for identification purpose, it has recently been researched as medical biometrics for a range of diseases. In this chapter, we investigate the facial landmark detection for atypical 3D facial modelling in facial palsy cases, while potentially such modelling can assist the medical diagnosis using atypical facial features. In our work, a study of landmarks localisation methods such as stacked hourglass networks is conducted and evaluated to ascertain their accuracy when presented with unseen atypical faces. The evaluation highlights that the state-of-the-art stacked hourglass architecture outperforms other traditional methods.

The Interest in this studies of human kinematics goes back very far in human history drove by curiosity or need for the understanding the complexity of human body motion. To find new and accurate information about the human movement as the advance computing technology became available for human movement that can perform. Martial arts (silat) were chose and multiple type of movement was studied. This project has done by using cutting-edge technology which is 3D motion capture to characterize and to measure the motion done by the performers of martial arts (silat). The camera will detect the markers (infrared reflection by the marker) around the performer body (total of 24 markers) and will show as dot in the computer software. The markers detected were analyzing using kinematic kinetic approach and time as reference. A graph of velocity, acceleration and position at time,t (seconds) of each marker was plot. Then from the information obtain, more parameters were determined such as work done, momentum, center of mass of a body using mathematical approach. This data can be used for development of the effectiveness movement in martial arts which is contributed to the people in arts. More future works can be implemented from this project such as analysis of a martial arts competition.

Martial Arts, Dancing and Sports dataset: A challenging stereo and multi-view dataset for 3D human pose estimation

The integration of Explainable Artificial intelligence (XAI) into sports training offers transformative capabilities for enhancing athlete performance, injury prevention, and personalized coaching. This paper presents the design and development of an intelligent sports training assistance system that leverages advanced AI technologies such as computer vision, deep learning, and wearable sensor analytics. The system incorporates real-time pose estimation using models like OpenPose and PoseNet, enabling precise motion tracking and automated technique correction across various sports disciplines including running, basketball, and martial arts. A multimodal dataset consisting of over <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5, 0 0 0}$</tex> annotated video frames and sensor-based motion logs was used to train deep learning models including 3D Convolutional Neural Networks (3D-CNNs) and Long Short-Term Memory (LSTM) networks-a variant of Recurrent Neural Networks (RNNs) for motion classification, fatigue detection, and feedback generation. The system achieved 95.2 % pose estimation accuracy, 93.4 % movement classification accuracy, under 1.8s feedback latency, and a 40 % improvement in training efficiency compared to traditional coaching. The suggested framework is deployed in university training centers, the system demonstrated high reliability and responsiveness, offering real-time guidance, objective performance analytics, and remote coaching support, making it a scalable and intelligent solution for modern sports training and performance enhancement.

Preserving, maintaining and teaching traditional martial arts are very important activities in social life. That helps individuals preserve national culture, exercise and self-defense for people. However, traditional martial arts have many different postures and activities of the body and body parts. The problem of estimating the actions of the human body still has many challenges, such as accuracy, obscurity, and so forth. Especially is 3-D human pose estimation. In this paper, we propose using Convolutional Neural Network (CNN) for estimating key points and joints of actions in traditional martial postures in the 2-D space and then projecting the results to the 3-D space and apply the measurements (length of joints, deviation angle of joints, deviation distance of key points) for evaluating pose estimation. The CNN model that we use is CPM (Convolutional Pose Machine), since then there is a comparative study proposed when CPM is trained on the classic MSCOCO Keypoints Challenge dataset [1] and Human3.6m [2], the results were evaluated on the Martial Arts, Dancing and Sports dataset of Zhang et al. [3]. The quantitatively results evaluated and published.

Convolutional neural networks (CNN) have been attracting attention for accurate scene parsing including a human pose estimation. However, CNN requires a massive amount of floating-point operations, so it is difficult to realize CNN on low-cost devices. Thus, we propose RegionDrop, an annotation-aware spatially sparse network, which skips computations of unnecessary spatial regions in activations. We present a novel loss that directly uses annotations so that important activation regions are retained. We also developed an efficient sparse GPU kernel to accelerate processing speed of both depthwise and general \(K\times K\) convolutional layers. Our RegionDrop is evaluated by using two pose estimation networks, a modified stacked hourglass network, and an HRNet. RegionDrop using an hourglass network archived 3.2 times faster processing speed compared with a non-sparse network, and 1.8 times faster processing speed than a prior spatially sparse network, with no accuracy degradation. Moreover, the processing speed of RegionDrop using HRNet is increased by a factor of 2.0 with negligible accuracy loss.KeywordsConvolutional neural networksPose estimationSpatially sparse networkGPU

Human Pose Estimation is a method of detecting and classifying all the key human body joints. A 2D Human Pose Estimation method predicts the two-dimensional coordinates from an input image. Classical methods used deformable parts models for detecting body joints. This paper discusses the state-of-the-art Deep Learning based approaches for estimating human poses. It can be implemented by either using the top-down or bottom-up technique for pose estimation. Here, the authors have studied the Convolutional Pose Machines, Hourglass Network and MoveNet architectures for Pose Estimation. These architectures are compared against each other in terms of their efficiency by measuring the time taken by each of them and their accuracy by evaluating them using the Percentage of Detected Joints metric on the MPII dataset. The architectures were implemented by using TensorFlow and Open-Source libraries built using PyTorch. The inference results for each joint were compared individually for all the body joints. The Hourglass Network showed the best accuracy but is really slow. The MoveNet architecture is the second best according to accuracy but is very fast and can be used in real-time systems.

Lane detection is the core problem of autonomous driving. After completing lane recognition, the autonomous driving system can realize the active safety and control function of vehicle lateral movement. However, the existing methods cannot adapt well to various environments and generate many unnecessary points, resulting in low detection accuracy. In this paper, Point Instance Network (PINet) based on key points estimation and instance segmentation is used, which is composed of several stacked hourglass networks that are trained at the same time. Compared with existing algorithms, PINet achieves ideal accuracy and false positive rate on CULane, especially in night and dazzle light.

In order to make the elderly care service system intelligent, this paper proposes a human posture recognition algorithm based on big data analysis and artificial intelligence technology. The algorithm framework includes three parts: human target detection, human posture estimation and human action recognition. Firstly, this paper adopts the obscured human body detection algorithm based on the mixed attention mechanism to detect the human bounding box. Then, to solve the problem that the detected human bounding box may have inaccurate positioning leading to the human body position deviation, the space transformation network is used to affine the coordinates of the human bounding box to correct its position. Finally, the stacked hourglass network is used to predict the joints of the detected human body, so as to realize the recognition of human posture. In this paper, the algorithm is trained and tested on the open pedestrian test set. The experimental results show that the algorithm in this paper can effectively and correctly estimate the visible parts of human posture, and reduce the influence of obstacle occlusion on human posture estimation and human fall detection. The evaluation index of human posture recognition (Kps AP) reached 68.7%, indicating that the posture recognition algorithm proposed in this paper can realize the rapid recognition and estimation of human posture without shelter, and the algorithm can complete the fall recognition of the elderly in the elderly service system, and can effectively prevent the elderly from falling after the subsequent big data statistics and analysis of the elderly's posture. It has good practicability.

Monitoring the development trend of martial arts biomechanics, which provides ideas for the scientific research of martial arts. It also proposes discussions and suggestions based on research results in related fields. As a traditional sport in China, martial arts have been around for a long time, and there is still a lot of research space. The biomechanical research of Wushu must be contemporary, forward-looking, and creative, and fully meet the needs of Wushu. This paper adopts the bibliometric method: it conducts statistical analysis on the research status and development trend of core journals and master and doctoral dissertations in the field of sports from 2011 to 2015. It aims to understand the latest developments in the biomechanical research of martial arts routines in the field of Chinese sports. The results show that the BP neural system model effectively improves the accuracy, recall, and F value of test classification. In the Naive Bayes method, the accuracy of the opening report is only 77.1%, the recall rate is 75.8%, and the F value is 76.3%. In the literature method, the accuracy of the opening report is 82.2%, the recall rate is 83.5%, and the F value is 82.8%. In the text method, the accuracy of the opening report is 86.5%, the recall rate is 88.3%, and the F value is 87.2%. In contrast, it outperforms simple Bayesian algorithms and methods in the literature.