Development of virtual information space system for visualizing and editing mul-tidimensional structural information.

我们提出了一个系统，用于可视化和编辑多媒体内容在一个虚拟的3D空间。系统输入多媒体内容并提取其属性。然后，它构建了一个虚拟空间，其中的内容和它们之间的相互关系是可视化的，参考他们的属性。可视化的内容是根据用户的需求，通过使用人性化的界面，如手势编辑。编辑的结果输出后，被重组表示。我们开发了一个原型的建议系统。电视新闻作为多媒体内容被使用，因为它们是典型的例子之一。广播的电视新闻被存储并划分成片段。选择良好地表示每个片段的静止图像作为代表图像。针对图像、telop和语音计算段之间的相关性。片段中的字幕和语音被转换成文本。名词是从课文中摘出的。通过计算名词的出现频率，将电视新闻片段分为政治、经济、社会事件、国际事件、体育和其他6类，并设计了一个基于手势的人机交互界面，提出了一种从检测到的手的三维运动中识别指令的算法，假设我们使用这些指令访问虚拟三维空间。在该系统中，手势的连续操作被分解为各个操作单元，并且这些单元被分类为有意义的单元或无意义的单元。然后，通过将单元与模型匹配来识别手势。

We proposed a system for visualizing and editing multimedia contents in a virtual 3D space. The system inputs multimedia contents and extracts their attributes. Then, it constructs a virtual space where the contents and their mutual relationship are visualized referring to their attributes. Visualized contents are edited according to user's demand by using human friendly interface such as hand gesture. Edited results are output after being restructured for presentation.We developed a prototype of the proposed system. TV news are used as multimedia contents since they are one of the typical examples. Broadcasted TV news are stored and divided into segments. A still image that represents each segment well is selected as a representative image. Correlation between the segments is calculated for images, telops and voice. Telops and voice in the segments are transformed into text. Nouns are extracted from the text. By calculating the frequency of the extracted nouns, the segments of TV news are classified into six categories : politics, economy, social events, international affairs, sports and the others.We developed a man-machine interface based on hand gesture and the algorithm to recognize commands from three dimensional movement of a detected hand on the assumption that we use them for accessing the virtual three dimensional space. In this system, a continuous operation of the gesture is decomposed into individual operation units, and these units are classified into meaningful units or meaningless units. Then, the gesture is recognized by matching the unit with the model.