Survey on AI Driven Gym Assistant with Pose Recognition
Devi Krishna K R, Devinandana D Devinandana D, Jayalakshmi V Jayalakshmi, Merin Mary Sabu, a b · International Journal of Advances in Engineering and Management · 2024
In our research, we reviewed studies focusing on pose estimation and correction of forms in exercising using deep learning and machine learning approaches. One such model was the CNN-LSTMbased model that used a combination of spatial and temporal attributes for the recognition action with satisfactory accuracy. Another system utilized OpenPose in real time for pose estimation and improved postures of exercises like bicep curls, shoulder presses, etc. Other researches used 3D pose estimation methods, such as VIBE, for the detection of repeated exercises with a high rate of precision. Other researches utilized wearable sensor data and machine learning techniques to monitor exercise posture. One research focused on arm and shoulder exercises of the user. AI-based models such as Bi-GRU were capable of providing real-time feedback while the user was performing squats so that he did not hurt himself. Advancement on tracking and exercise performance improvement capabilities was also evaluated through humanoid fitness trainers and pose-guided graph convolutional networks. Examples of such studies hereby establish the potentiality of AI-driven systems for health monitoring and personalized exercise feedback.