Tools
Tools
Rajendran, Aarthi (2024) Physical Exercise Pose Detection using BlazePose and Machine Learning Framework. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (936kB) | Preview |
Abstract
Pose detection involves identification of specific body postures and movements. This approach can be applied to various exercise activities, providing insights with widespread applications in fitness, sports, healthcare, and human-computer interaction. This research proposes automated pose detection and classification algorithms to automate the identification of exercise activities, providing a scalable and effective solution for real-time monitoring and analysis in various factors such as enhancing fitness tracking, improving sports performance engaging more people into the physical activity. This research presents a comprehensive approach to detecting human poses and identifying specific exercise activities using machine learning techniques. By utilizing MediaPipe for pose detection and the classification algorithm able to detect the key points accurately on the human body and classify them into predefined exercise categories. The proposed framework enhances real-time monitoring of human exercise poses by integrating pose detection models with classification algorithms. Using pose detection models like MediaPipe and BlazePose, which extract body positioning features in the form of coordinates (x, y, and z). These coordinates are subsequently used to generate custom geometric features for model training. A large-scale dataset from UCF 101 consists of realistic action videos having 101 different action categories. This data set is an extension of UCF50 data set which includes 50 action categories and features 13320 videos across the 101 action categories, where 7 groups are taken for our research such as pushups, pullups, Body weight squats, bench press, and jumping jacks. Using three algorithms to classify the exercise activities. Among these, XGBoost is identified as the best model, achieving an accuracy of 94%. followed by GBM with an accuracy of 93% and Distributed Random Forest (DRF) with an accuracy of 91%. This framework enhances real-time monitoring of human exercise poses, offering significant potential for applications in automated fitness tracking and sports analytics. This research aligns with the cutting edge in pose detection and classification, providing real time monitoring solution by using advanced algorithms and machine learning for high accuracy by making it suitable for various fitness applications. This research has potential for practical applications such as fitness tracking, and health monitoring. Future work could address the integration into the wearable devices to provide real time feedback and insights.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Makki, Ahmed UNSPECIFIED |
| Subjects: | G Geography. Anthropology. Recreation > GV Recreation Leisure Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning R Medicine > RA Public aspects of medicine > RA0421 Public health. Hygiene. Preventive Medicine > Personal Health and Hygiene |
| Divisions: | School of Computing > Master of Science in Data Analytics |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 25 Aug 2025 10:23 |
| Last Modified: | 25 Aug 2025 10:23 |
| URI: | https://norma.ncirl.ie/id/eprint/8615 |
Actions (login required)
 |
View Item |