Spotting Athletes: A New Way with Deep Learning

In sports like running, jumping, or soccer, coaches often need to study athletes' moves to help them improve. One big challenge is quickly spotting athletes, their gear, and field boundaries in videos or photos. Existing methods struggle with this, mainly because they lose important time details, mix up targets, have overlapping objects, and combine two tough tasks: guessing what something is (classification) and where it is (regression).

To fix this, a team created a new way to use deep learning for this job. They added a special tool called a TSM module that helps networks understand time better in sports scenes. They also made a clever attention mechanism to focus on individual athlete actions and split the tough tasks into separate parts. They tested their new method on public datasets and compared it to seven other common tools. Their method won with a 92. 298% accuracy score in a map_0. 5 test. Even when they removed parts of their method, it still scored well, showing it works great for sports analysis.

Actions