Automatic virtual camera control using a style imitation technique for 3D game based animation

This research aims to develop a model of automatic virtual camera control for 3D animation film making based on 3D game. The model consists of two main parts, which are rules of shot as representation of control camera knowledge, and inference engine to generate virtual camera control in capturing ideal shot. Style imitation technique commonly used to generate music composition automatically, is used to imitate film materials of reference film. Film material are defined in components of number of actors in a shot, type of action, type of control camera, shot duration, and shot sequence. Shots of reference film are analyzed based on these components to build perspective packages of camera control knowledge base. Type of control camera, perspective packages, shot sequence is used as constraint to define ideal shot. Genetic algorithm is use to generate virtual camera control, with data of shot duration of each perspective package is used as parameter of iteration. Experiment is conducted by imitating shots of 3D animation film entitled how to train your dragon 2 produced by DreamWorks Animation, and developing a 3D game in which the proposed model is implemented. Evaluation of the performance of genetic algorithm results 82.2% shots of total number of shots can be generated by the algorithm. The implementation of style imitation is measured by comparing shots produced in this research to shots of reference film based on point of view. Three experts of film and multimedia are asked to value the similarity, and Mean Opinion Score technique is used to measure the expert's opinion. The result shows that in scale 1-5, where score 1 is for poor, and 5 is for excellent, 3D animation produced in this research are similar to shots of reference film at score 3.9.