Deep learning approaches for workout repetition counting and validation

Abstract

The study of the human body motion throughout different activities is one of the most challenging and long-standing problems in Computer Vision. With the recent advances in Deep Learning algorithms, the information acquired from conventional frame sensors can be used to infer the human body pose for further analysis. Specifically, one can contemplate the performance of body movements throughout physical exercises to provide feedback to the individual. Therefore, we propose a system for workout repetition counting and validation based on a set of skeleton-based and deep semantic features that are obtained from a 2D human pose estimation network. To this end, we have acquired over 130 participants performing five popular Cross-Fit exercises in order to train a Convolutional Neural Network to predict the exercises’ moments at a frame level. Hence, this underlying idea of inferring the moment of the exercise is two-fold: (i) to provide information about the exercise execution with a fine-level of detail; (ii) the ability to detect invalid repetitions promptly. Finally, a repetition counting and validation module receives the predicted moment and outputs the current number of valid repetitions that one has been performing with over 92% precision scores for 4 out of 5 considered exercises.