MTUSI has developed a system for detecting a human skeletal model during a fitness class

The research department for the development of application and system software at MTUSI has developed a system for detecting a human skeletal model during fitness using HPE technology – HumanPoseEstimation.

Human Pose Estimation (HPE) is a technology for identifying and classifying nodes of the human body. In fact, this is a way of determining the coordinates of each node (arm, head, torso, etc.), called a key point, which determines the position of the human body. HPE is used to evaluate a person's performance during their workout: whether they do a given exercise correctly, how many times they do it, and how effectively they perform it.

The system uses a developed lightweight convolutional neural network GL-Pose for human pose estimation, adapted to output results in real time on different types of devices and trained on the collected DataSet. This model is one of the leaders in terms of accuracy for HPE tasks and shows results of 74% according to the mAP metric, as well as 97.5% according to the metric PCK@0.2.

The development of such a system will allow the addition of personalization functions that will help create a completely individual training plan based on the user’s physical level. All this makes training more effective and accessible, allowing everyone to exercise at home, adjust their workouts in real time and achieve optimal results at no extra cost.