Three years ago the world was hit by a crisis related to the new coronavirus which had a big impact on human habits of physical exercises and activities outdoors and in fitness centers. Since then more and more people are opting for different types of physical exercises at home. Many people are ashamed of their bodies and do not want to go to fitness centers or the subscriptions are too much for them. In the meantime, a lot of fitness applications were implemented which offer digital or virtual fitness instructor which lead the users through the exercise, count the reps, and warn them about their posture so that it can be corrected. Most of these applications are based on machine learning and computer vision where under the hood, recognizing human posture and body parts is happening. In our thesis, we analyze different implementations of libraries that allow such pose estimation, evaluate them, and compare them. We analyze the use of these implementations in fitness applications that are already in use in public and evaluate their efficiency and usability from the perspective of a user. In the end, we present our implementation of such an application that allows pose estimation
|
