Research On Story Understanding And Scene Planning In Automatic Generation Of Computer Animation

The Automatic Generation of Computer Animation (AGCA) technology studies the design and production process, focuses on the process from natural language to the final animation product, aimed at increasing the degree of automation and intelligence of animation product. As a new topic in Computer Science, AGCA combines many disciplines of the existing research and has great theoretical and practical importance.Story understanding and scene planning are both important modules of AGCA. As the first module, the main task of story understanding is to analyze and understand the input natural language text and then generate the formal description of the input story. As the beginning of scripts generation, the main purpose of scene planning, based on the formal description of the input story, is to invoke the appropriate models and automatic planning algorithm to obtain the formal description of the scene according to certain knowledge base and rules base, and finally render 3D scenes.This paper first analyzes and studies the story understanding and scene planning techniques of AGCA, then designs and implements a system called Text2Scene, which transforms natural language input to 3D scene, with the processes of understanding and planning. The system is an important part of AGCA and can be easily integrated into the whole AGCA system in the future.For story understanding module, this paper designs and implements a parser which is based on Probabilistic Context Free Grammars and cognitive linguistics dictionary WordNet. Lexical Visual Semantic Representation which defines the visual semantic representation of events, nouns, adjectives and prepositions, is introduced to link natural language and visual information and finally obtain the formal description of the input story.For scene planning, the paper first introduces the design of model library, which includes object library, spatial relation and action posture. The description information, especially the spatial tag, is very import for the generation of scenes. Then the design of knowledge base, including scenes base, rules base and constrains base, and the process of scene planning algorithm is analyzed. Finally, the paper generates MEL scripts according to the scene description information and render 3D scenes on Maya model building platform.At last, this paper summarizes the conclusions and discusses the future work.