Understanding human movements often starts with recognizing actions from videos. Unlike images, videos offer a wealth of information, helping to pinpoint exactly what's happening. Over the past decade, researchers have made significant strides in video action recognition by focusing on datasets, innovative models, and learning techniques. However, sports analytics present unique hurdles. Capturing and labeling data requires expertise from sports professionals, and the actions can be lightning-fast and complex, involving multiple players in team sports like football and basketball.

To shed light on this field, let's dive into a survey of video action recognition in sports. We'll explore over ten types of sports, from team sports such as football, basketball, volleyball, and hockey, to individual sports like figure skating, gymnastics, table tennis, tennis, diving, and badminton. We'll compare existing frameworks used for sports analysis, showing the current state of video action recognition in both team and individual sports.

One of the key challenges lies in the complexity of recognizing actions in team sports. A single action can involve many players, making it difficult to accurately identify each player's role. To tackle these issues, we'll discuss the unresolved problems in this area and introduce a toolbox developed using PaddlePaddle. This toolbox supports action recognition in football, basketball, table tennis, and figure skating, aiming to simplify sports analytics.

Unraveling Sports Actions: A Video Recognition Journey

Actions