Abstract:
"The presented research addresses the need for an advanced fitness application that integrates AI and computer vision technologies to provide real-time pose estimation and pose class classification. Traditional fitness apps often lack personalized guidance and real-time feedback, leaving users struggling with maintaining correct exercise form and staying motivated. This study aims to develop an AI-powered personal trainer app that provides real-time feedback and corrections with personalized workout plans, catering to the unique needs of fitness enthusiasts. To address the identified problem, the research employed a methodology centered around the integration of computer vision models and advanced AI techniques. Leveraging state-of-the-art models along with alternative architectures like LSTM and CNN, the system redefined real-time posture detection and tracking during exercises. Additionally, personalized feedback and user profiles were incorporated to offer customized guidance based on individual fitness goals and preferences along with rep counting. The approach focused on overcoming limitations in current fitness applications, ensuring accuracy, accessibility, and ease of use presenting a web application.
The initial results of the study demonstrated promising outcomes, particularly in terms of model accuracy. Using LRCN, the classification model achieved an accuracy of 96% across four distinct classes. This indicates a significant improvement in exercise recognition compared to existing fitness apps. However, further evaluation metrics such as confusion matrices and classification report will provide a more comprehensive understanding of the model's performance. Overall, these preliminary findings underscore the potential of the proposed AI-powered personal trainer app to revolutionize the fitness industry by providing tailored guidance and real-time feedback to users, thereby enhancing their overall fitness experience."
