Personalized system for human gym activity recognition using an RGB camera

Abstract

Human Activity Recognition is one of the most researched topics in the field of computer vision. It is a powerful tool mainly used to aid medical systems, smart homes, surveillance, and many more areas. In this paper, an RGB camera was used to record gym activities such as push-up, squat, plank, forward lunge, and sit-up. Features were extracted from the recorded videos and were fed into classification algorithms such as Support Vector Machines, Decision Tree classifier, K-Nearest Neighbor classifier, and Random Forest classifier. The developed models were evaluated using metrics such as accuracy, balanced accuracy, precision score, recall score, and F1 score. The Random Forest Classifier outperformed all the other attempted methods with an accuracy of 98.98%. A repetition counter was developed, which splits workouts based on local minima analysis, and correctness of the workout was calculated for each skeletal point using dynamic time warping. An interactive android application was built for the user to gain insights on the performed workouts.