Please use this identifier to cite or link to this item: http://dspace.dtu.ac.in:8080/jspui/handle/repository/20802
Full metadata record
DC FieldValueLanguage
dc.contributor.authorREHMAN, AMAN-
dc.date.accessioned2024-08-05T08:56:03Z-
dc.date.available2024-08-05T08:56:03Z-
dc.date.issued2024-05-
dc.identifier.urihttp://dspace.dtu.ac.in:8080/jspui/handle/repository/20802-
dc.description.abstractPhysical exercise is a key part of a healthy lifestyle, encompassing a wide array of activities from dance to weightlifting and sports. Proper form and posture are critical to the ensure the safety of these exercises and making them effective. Accurate assessment of workout poses can provide invaluable feedback to individuals and fitness professionals alike, enabling adjustments for optimal performance. Performing exercises correctly also reduces the risk of injury, especially for people new to exercises. Additionally, fitness professionals can utilize this to remotely guide clients in virtual training sessions. The emergence and evolution of deep learning techniques has revolutionized computer vision tasks, offering excellent performance in various tasks. In the context of workout pose estimation, this can be used for automating the process of assessing body positions during exercises. Multiple machine and deep learning techniques have been used in this domain with excellent results. Although a comparative study has not been performed, and the datasets chosen have seen extreme variety. Our approach involves leveraging advanced architectures, machine learning methods, and ensemble learning methods like Random Forest, XGBoost, to develop an efficient and accurate system capable of precise pose estimation across a diverse range of exercises with quick inference. Further objectives of the research include implementing and comparing the perfor mance of the above mentioned models with fine-tuning for workout pose estimation, eval uating their performance to understand the strengths and weaknesses of each architecture, collect and present the information gained, and suggest models that give high performance and quick inference. We also aim to shed light of some of the less utilized methods to explore the above problem. Our approach also captures the difference in using plain deep learning networks, as opposed to keypoint based machine learning and ensemble methods.en_US
dc.language.isoenen_US
dc.relation.ispartofseriesTD-7320;-
dc.subjectWORKOUT POSE RECOGNITIONen_US
dc.subjectDEEP LEARNINGen_US
dc.subjectMACHINE LEARNINGen_US
dc.subjectPHYSICAL EXERCISEen_US
dc.titleWORKOUT POSE RECOGNITIONen_US
dc.typeThesisen_US
Appears in Collections:M.E./M.Tech. Computer Engineering

Files in This Item:
File Description SizeFormat 
AMAN REHMAN M.Tech..pdf10.3 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.