Exercise Pose Recognition and Counting System using Robust Topological Landmarks

The paper introduces the exercise recognition and counting system based on vision systems. The current detectable exercises include barbell curls, push-ups, and lateral raises from the "Workout/Exercise Images" dataset, an open-source dataset from Kaggle. The system can predict and count live streams in real-time using a mobile phone or laptop camera. We improve the robustness of the system by using centering and scaling. Furthermore, this research reduces the complexity of the trained data by half without losing the accuracy of the machine learning model by implementing Principal Component Analysis (PCA). The implications of this research extend beyond the realm of fitness tracking, with potential applications for in-home exercise, sports analysis, physical therapy, and interactive fitness technologies.