Video Preprocessing for American Football Formation Recognition

American football analytics and in particular formation identification is important to gaining a competitive edge. Currently, most of this football video annotation process is done by hand. With the recent advancements in computer vision and artificial intelligence, automated sport analysis systems are gaining popularity. One major challenge in current football systems is constraints on camera location. In this work we attempt to overcome issues of camera location by performing a perspective normalization method. This entails using conventional computer vision algorithms to locate various lines and markers on the football field. This information is then used to normalize points into a camera position agnostic perspective. This preprocessing will then allow for a neural network to be trained to identify formations and preform other football analytics.