Taylor and Francis Book Chapter

2024

Read Taylor and Francis Book Chapter (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2411-00833.txt.

Description

Abstract \chapterauthor M. M. Akash*visie.tech, Dhaka, Bangladesh Email: [email protected], *Corresponding Author \chapterauthor Rahul Deb MohalderKhulna University, Khulna, Bangladesh Email: [email protected] \chapterauthor Md. Al Mamun Khanvisie.tech, Dhaka, Bangladesh Email: [email protected] \chapterauthor Laboni PaulKhulna University, Khulna, Bangladesh Email: [email protected] \chapterauthor Ferdous Bin Alivisie.tech, Dhaka, Bangladesh Email: hridoyferdous@yahoo.

Catalog record

YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2411-00833

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Taylor and Francis Book Chapter Abstract \chapterauthor M. M. Akash*visie.tech, Dhaka, Bangladesh Email: [email protected], *Corresponding Author \chapterauthor Rahul Deb MohalderKhulna University, Khulna, Bangladesh Email: [email protected] \chapterauthor Md. Al Mamun Khanvisie.tech, Dhaka, Bangladesh Email: [email protected] \chapterauthor Laboni PaulKhulna University, Khulna, Bangladesh Email: [email protected] \chapterauthor Ferdous Bin Alivisie.tech, Dhaka, Bangladesh Email: [email protected] Chapter 0 Yoga Pose Classification Using Transfer Learning \chaptermark Yoga Pose Classification Using Transfer Learning \chapterinitial Yoga has recently become an essential aspect of human existence for maintaining a healthy body and mind. People find it tough to devote time to the gym for workouts as their lives get more hectic and they work from home. This kind of human pose estimation is one of the notable problems as it has to deal with locating body key points or joints. Yoga-82, a benchmark dataset for large-scale yoga pose recognition with 82 classes, has challenging positions that could make precise annotations impossible. We have used VGG-16, ResNet-50, ResNet-101, and DenseNet-121 and finetuned them in different ways to get better results. We also used Neural Architecture Search to add more layers on top of this pre-trained architecture. The experimental result shows the best performance of DenseNet-121 having the top-1 accuracy of 85% and top-5 accuracy of 96% outperforming the current state-of-the-art result. 1 Introduction Human Activity Recognition (HAR) is a field in computer vision that involves identifying and classifying human activities from sensor data or video streams. This can be done using various techniques, such as machine learning algorithms, pattern recognition, and computer vision. Yoga poses are a specific type of human activity that can be recognized using HAR techniques. By analyzing sensor data or video streams of a person performing yoga, a HAR system can identify and classify the specific yoga poses being performed. This can be useful for a variety of applications, such as yoga instruction, and injury prevention. HAR system could be used to track a person’s progress as they perform a series of yoga poses, providing feedback and guidance on proper form and technique. It could also be used to identify any incorrect or potentially harmful movements, helping to prevent injuries. Machine learning is being used to help people exercise more effectively through the use of virtual personal trainers. These are systems that use machine learning algorithms to provide personalized exercise routines and coaching based on the user’s goals, fitness level, and other factors. By analyzing data such as heart rate, body composition, and exercise performance, these systems can adapt and modify the user’s routine in real-time to optimize their results. Wearable technology such as fitness trackers and smartwatches are increasingly using machine learning to improve their accuracy and effectiveness. For example, some fitness trackers use machine learning algorithms to analyze data such as steps taken, calories burned, and heart rate variability to provide more accurate and personalized feedback to the user. During the COVID-19 pandemic, Yoga is a popular practice that has been shown to reduce stress, anxiety, and depression. However, to get the maximum benefits of yoga, it is crucial to perform the poses correctly, which requires access to a yoga trainer. Unfortunately, not everyone has access to a trainer. For these concerned people, a software application can play the role of an instructor. Traditional pose estimation methods for recognizing and correcting poses are at a disadvantage here [ 15 ] because of the diversity of poses on human bodies. Therefore, this paper proposes a novel solution, a classification system that captures the similarities and differences between poses. Instead of producing body key points for human subjects, this method recognizes the overall pose and classifies it. The proposed approach can improve access to yoga for individuals who do not have access to a trainer and enhance the benefits of yoga for all practitioners. 1 Our Contribution: The key contributions of this research are: We have experimented with different types of image preprocessing techniques. We have used transfer learning in different stages like using fully imagenet weight, finetuning different layers of the model, and finetuning the whole model. We have defined the network using Random Search for architecture. The paper is organized as follows, the very next section (section 2 ) contains the previous work description. After that, we introduced our methodology in section 3 followed by result analysis in section 4 and the conclusion and future work in section 5 . 2 Related Work Chen [ 5 ] conducted various deep learning (DL) based techniques to deduce the human pose estimation from images. The usage of Deep learning for human pose estimation was established when AlexNet used CNN and achieved state-of-the-art results. The researchers pointed out that there are now a variety of pose estimate techniques, including pixel-level analysis, body joint point mapping, body-based and model-based human body estimation, and heat-map mapping. Faisal [ 7 ] continued on Chen’s body-joint estimation research [ 5 ] and mentioned the current state-of-the-art, where they repeated that the body-joint points are identified by employing different gyroscope and joint angle methods for determining joint point angle, as well as multiple sensor fusion. The authors of [ 2 ] repeated Faisal and Chen’s findings from [ 5 ] , [ 7 ] by estimating human activity based on poses. It should be noted that prominent posture estimation techniques include template-based, generative model-based, and discriminative model-based techniques. These techniques were later verified in research on human pose estimation published in [ 6 ] . Yoga has been more well-liked among people because of its believed benefits of good health and fitness. Nagalakshmi examined the effects of yoga and discovered that remote yoga activities have become significantly more popular since the COVID outbreak, in addition to alleviating musculoskeletal pain and also inspiring a healthy lifestyle [ 13 ] . For this goal, Agarwal [ 1 ] experimented with various Machine Learning (ML) methods. They created a dataset of 5,500 photos for ten distinct yoga postures and then they used the open pose estimation algorithm to create the skeleton images. On their dataset, they acknowledged that random forest classifiers performed better. Liaqat et al. [ 11 ] created a hybrid posture recognition approach by combining deep neural networks with conventional machine learning techniques. The weight learned by the deep learning model and the prediction from the conventional model are combined to produce the final class prediction. In a different paper [ 10 ] , the authors first extracted key points using OpenPose, then classified the data using a hybrid CNN-LSTM layer. The researchers on this team worked for sitting pose estimation [ 9 ] and concluded a Deep Recurrent Hierarchical Network based on MobileNetV2 and gained 91.47% accuracy. Similarly, the use of SVM and boosting for stance detection is addressed in [ 14 ] , and Nagalakshmi’s work [ 12 ] specifically focused on yoga. Keeping pace with deep neural networks pose estimation models are getting more effective day by day. The mostly accurate keypoints and skeleton annotations are the strength of state-of-the-art pose estimation models. Pose estimation models are widely used in human surveillance [ 4 ] , and human activity recognition [ 8 ] like computer vision problems. So far we are working with yoga poses as a result we can say

Read the complete text