MTUSI: Human Pose Estimation (HPE)

Main article: Neural networks (neural networks)

2024: Representation of the human skeletal model detection system during fitness activities

The Research Department of Applied and System Development software MTUSI has developed a system for detecting a human skeletal model during a lesson fitness using technology - HPE HumanPoseEstimation. The university announced this on July 16, 2024.

Human Pose Estimation (HPE) is a technology for identifying and classifying nodes in the human body. In fact, this is a way to determine the coordinates of each node (arm, head, torso, etc.), called the key point and determining the position of the human body. HPE is used to evaluate a person's action during their training: whether they are doing the exercise properly, how many times they have done it, and how effectively they perform it.

The system uses the developed lightweight convolutional neural network GL-Pose to evaluate human posture, adapted to output real-time results on different types of devices and trained on the assembled DataSet. This model is one of the leaders in accuracy indicators for HPE problems and shows results in 74% by mAP metric, as well as 97.5% by PCK @ 0.2 metric.

The development of such a system will allow you to add personalization functions that will help you draw up a completely individual training plan based on the physical level of the user. All this makes training more efficient and affordable, allowing everyone to play sports at home, adjust their activities in real time and achieve optimal results without unnecessary costs.