Multisport Dynamics Explored Through Advanced Activity Recognition Techniques in Athletic Performance

Recognition of human activities has drawn a lot of interest lately in the field of computer vision and machine learning. Group activity recognition is a significant subcategory in which several people participate in a common activity. The primary obstacle in these tasks is learning the relationships between individuals in a scene and understanding their evolution over time. The suggested study offers a new taxonomy to classify state-of-the-art (SOTA) group activity recognition approaches and subcategorizes the current literature. It also critically analyzes these techniques. To understand scenes involving multiple people, models must describe individual actions in context and infer collective activities. Accurately capturing relationships between actors and performing relational reasoning is crucial for comprehending group activities. However, modelling these relationships is challenging due to the limited availability of interaction information, relying only on individual action labels and collective activity labels. Inferring relationships from other aspects is thus essential. Group activity recognition has garnered increased research attention recently due to its importance in video understanding. The difficulties lie not only in recognizing individual actions but also in exploring scene information and collaborative relations among people. This research addresses these challenges by providing a comprehensive review of existing techniques, offering a structured approach to categorize them, and highlighting the importance of relational reasoning in understanding group activities.