Automatic Generation of 3D Animations from Text and Images

Abstract

The understanding of information in a text description can be improved by visually accompanying it with images or videos. This opportunity is particularly relevant for books and other traditional instructional material. Videos or, more in general, (interactive) graphics contents, can help to increase the effectiveness of this material, by providing, e.g., an animated representation of the steps to be performed to carry out a given procedure. The generation of 3D animated contents, however, is still very labor-intensive and time-consuming. Systems able to speed up this process offering flexible and easy-to-use interfaces are becoming of paramount importance. Hence, this paper describes a system designed to automatically generate a computer graphics video by processing a text description and a set of associated images. The system combines Natural Language Processing and image analysis for extracting information needed to visually represent the procedure depicted in an instruction manual using 3D animations. It relies on a database of 3D models and preconfigured animations that are activated according to the information extracted from the said input. Moreover, by analyzing the images, the system can also generate new animations from scratch. Promising results have been obtained assessing the system performance in a specific use case focused on printers maintenance.