Abstract

The articial intelligence is nowadays a constant topic of conversation witha eld of research that is pushed forward by some of the world’s largestcompanies and universities. Deep learning is a branch of machine learningwithin articial intelligence based of learning representation of data suchas images and texts by processing the data through deep neural networks.Sports are competitive businesses that over the years have become more datadriven. Statistics play a big role in the development of the practitioners andthe tactics in order to win. Sport organizations have big statistic teams sincestatistics are manually obtained by these teams. To learn a machine to recognize patterns and actions with deep learning would save a lot of time. In this thesis a deep learning approach is used to examine how well it can performto classify actions in American football games. A deep learning architectureis first trained and developed on the UCF101 video dataset and then trainedto classify run and pass plays of a new, American football dataset called theA22 dataset. Results and earlier research show that deep learning has potentialto automatize sport statistic but is not yet ready to overtake the rolestatistic teams have. Further research, bigger and more task specific datasetsand more complex architectures are required to enhance the performance ofthis specific type of deep learning based video recognition.