利用多模态机器注意改善视频会议用户感受  

Improve the Experience of Users of Video Conference System with Multi-modal Machine Attention

作　　者：周敏[1] 张丽清[1] 

机构地区：[1]上海交通大学计算机科学与工程系

出　　处：《微型电脑应用》2006年第12期1-4,64,共5页Microcomputer Applications

基　　金：中国国家自然科学基金(60375015)支持

摘　　要：远程视频会议的应用已很多年了,但是,目前的视频会议系统对使用者的限制较多,往往需要使用者对系统保持关注方可正常使用。如何通过改善视频会议系统的人机交互模式来改善用户感受是视频会议系统进一步完善的重要途径。多模态机器注意技术来自于仿生学的研究成果,本文通过对基于多模态信息集成和同步来实现机器注意技术的讨论,将这种技术用于改进视频会议系统的人机交互模式,通过使视觉、听觉信号集成和同步的方法实现人工系统中视觉关注于感兴趣的对象——发言人的脸部及其发言。改进后的系统大大减少了发言人受到的约束,能够准确地自动跟踪发言人,而且在画面上更加突出视频会议画面中感兴趣的对象。Remote video conferences have been existing for many years. However, there are still quite a few restrictions on users in present video conference systems. e. g. a user has to keep paying attention to a computer system. One important approach to further improve the system is to develop a better mode of human computer interaction. The Multi-modal Machine Attention Technology, based on bionics, is introduced. By integrating and synchronizing multi-modal information includingvideo and audio signals, we realize the technology, which can automatically trace a speaker in a conference ?focus the attention of an artificial system on anyobject of interest, such as the face of a speaker and his speech, and display the concerned object with a proper ratio in the center of pictures to equally attract the audience in the remote conference. The improved human computer interaction reduces the restrictions on speakers.

关 键 词：多模态机器注意 声音定位 人脸检测 人脸跟踪 

分 类 号：TN948.63[电子电信—信号与信息处理]