Over-the-shoulder shot detection in art films

The ability to characterize a film, in terms of its narrative and style, is becoming a necessity especially for developing personal video recommendation systems to better deliver on-demand Internet streaming media. Among the set of identifiable stylistic features which play an important role in the film's emotional effects, the use of Over-the-shoulder (OtS) shots in movies is able to convey a big dramatic tension on the viewers. In this work we propose a methodology able to automatically detect this kind of shots by combining in a SVM learning scheme some state-of-the-art human presence detectors, with a set of saliency features based on colour and motion. In the experimental investigation, the comparison of obtained results with manual annotations made by cinema experts proves the validity of the framework. Experiments are conducted on two art films directed by Michelangelo Antonioni belonging to his famous "tetralogy on modernity and its discontent'', one in shades of gray (L'avventura, 1960), and the other in colour motion (Il deserto rosso, 1964).