A Computer Vision Based Framework for Posture Analysis and Performance Prediction in Athletes

In the previous several decades, there have been tremendous improvements in the analysis of human movement in the field of sports biomechanics and rehabilitation. Driven by the demand for faster and more complex ways to capture movement across a variety of situations, ranging from clinical gait analysis to video game animation, motion analysis techniques have undergone significant development. This research explores the previous work done on the sports performance analytics systems based on computer vision and how they affect the development of athletic performance but the prior work lacks in accuracy and versatility. Therefore, in this research, a framework has been proposed which uses 3D pose estimation technique and parameters like distance offset and inclination between neck and torso to analyse athletes' postures and motions during any sporting activity in order to predict the correctness of an athlete's posture. In this work, an analysis has been carried out with the help of machine learning models on the sports videos having a frame rate of 25 fps. To improve the robustness of the model, variations in blur, illumination, and clarity are incorporated into the dataset. This dataset has been trained on various machine learning models like Random Forest, MLP, and SVM, and procured the accuracy rates of
$85.03{{\% }},71.42{{\% }}$
, and
$63.26{{\% }}$
, and
$F1$
-scores of
$84.99{{\% }}$
,
$71.57{{\% }}$
, and
$61.80{{\% }}$
, respectively. Our results show how well these technologies work to provide coaches and athletes with performance information and real-time feedback.