面向广电节目的虚拟人手语合成显示平台研究

基于中国手语合成技术的虚拟人手语视频显示平台技术是一个全新的课题.为了满足广电新闻节目对手势运动流畅性要求,实现了一种上下文相关的手势运动平滑算法,该方法能够充分利用前后两帧的差异来实现手势运动的平滑过渡,其视觉感观效果较传统插值算法更加平滑自然;同时提出了一种基于统计和规则相结合的手势运动重定向算法,在统计方法的基础上针对不同骨架大小以及运动特性进行规则约束,使得标准模型手势运动数据应用到新模型上而不失其准确性;最后,通过扩展基本手语词表达形式并基于alpha融合技术实现了面向广电新闻节目的虚拟人手语合成显示平台并取得很好的结果.

Gesture Synthesis Based on Virtual Newscaster for Broadcast TV News Programs

How to develop avatar displaying system based on Chinese sign language synthesis technology for TV news programs is a novel task. However, the majority of existing displaying systems working on Chinese sign language synthesis concentrate on mapping system which automatically converts Chinese text to some sign that the deaf can understand well and it doesn't implement such displaying system. In this paper, an improved context-sensitive algorithm of gesture animation is presented. It can effectively utilize the differentiae of the consecutive key frames and make the avatars' gesture motion more natural and smooth than traditional algorithms. The authors propose a motion retargeting algorithm based on statistics and regulations for avatars' gesture animation. It uses regulations based on statistics, and considers the differentiae of skeleton in size and the features of gesture motion between virtual models, which have the similar skeleton topology. It can improve the accuracy of gesture motion data when transferring the standard avatar's gesture motion data to the new model without rectifying the standard model's gesture motion data by people using specified software. Finally, by extending the forms of basic gesture words' representation and adopting the alpha blending technique, a displaying system based on Chinese sign language synthesis technology for TV news programs has been effectively implemented.