Text Based 3D Scene Generation

Natural language is a ubiquitous and vital part of human society and intelligence. It is reasonable to believe that the technology of processing and understanding natural language automatically could facilitate the everyday life of ordinary people. Indeed, there exist many other applications in this area. For example, automatic language translation technologies help people to acquire the information presented in unfamiliar language; automatic grammar analysis technologies help people avoid making grammatical mistakes during writing; and automatic summarization technologies condense large amounts of information into key contents to save people time. In addition to these areas, natural language processing technology in other fields also has important applications, one of which is text to 3D scene generating automatically.In early time, artists create works of art mostly done manually. This manual process can be tedious, time consuming and labor intensive. The generation of 3D scene based on text is just to make this creation process automatically and free people in complex software learning and operating. In addition to art creation, the text to 3D scene also has many applications in other fields. For example, the simulation of car accident, agent roaming in 3D virtual world, and so on. This application background makes it not only having enormous commercial value for the entertainment and advertising industries, but also the potential to become a useful illustration tool in education and communication.The first of the difficulties of the text to the 3D scene is the understanding of the text. Because of ambiguity of the human language, a word or a phrase that is in a different context has different part of speech and meanings. As a result if we want to full understanding of the text, we must look up to the context of the described text which results in a large knowledge base that we need. The second difficulty is the basic idea of text to visualized scene is very simple, but the concrete method and the step are undefined and the algorithm and the data structure of every step are undefined also. Every reasonable approach is possible. The third, as the types of the real-world objects are diverse, the actions and shapes of objects are very different. Therefore, to accurately express all kinds of objects, there must keep a large number of models in the model database and the models must be preprocessed. And this workload is very huge.To solve the problem mentioned above, this paper do a research in automatically generating 3D scene based on text. This research based on related theory presents the implementation process which differs from previous work and implements the system. In the text analysis phrase, this paper presents the querying method of a word's hypernym and hyponym and the algorithm of removing the pronoun in the text. In the phrase of rendering scene, this research defines the scene description file, collects 3D models, proposes the method of preprocessing models, the property matching algorithm according to the principle of full text retrieval, the algorithm of path planning in rendering scene.