Action Recognition

Utilizing pre-trained models to identify key-points of individuals and athletes, subsequently employing them to classify the player’s actions. This process of point extraction is known as pose estimation, where each body part is assigned a unique point. For instance, in certain scenarios, 17 points may be extracted, while other algorithms may yield more.

Our objective in this project was to develop an API that initially extracts points from players in input videos, then identifies their actions, and finally transfers the information to the client.