Movie shot classification using machine learning

Summary

In movie production, we call shot the time unity upon which a specific and continuous type of camera is dominated until a subsequent module with a different type of movement or from another camera begins. A shot is not related to the scene that has wider time, spatial and semantic characteristics. A movie consists of a large number of shots where they can differ from each other-directed. This work aims to detect shots presented in a film and classify each of them in the class to which it be- longs. The classification process is through supervised machine learning algorithms. In addition, a new dataset containing shots is presented classified in various directorial categories and a corresponding labeling process that data from at least three users have been required. Finally, this work presents a demo to evaluate trained algorithms in real movies through statistical associations of the categorization results of the individual shots.