多模态人机交互中基于笔输入的手势识别

Pen-Based Gesture Recognition in Multimodal Human-Computer Interaction

为研究多模态人机交互系统的理论及构造方法 ,提出了一种快速的、单笔划手势识别方法 .该方法通过提取手势轨迹的关键点及各关键点的运移方向 ,形成特征码 ,然后与标准手势符号的各种可能的特征码进行匹配 .其中方向特征用于预分类 ,而关键点位置信息用于细分类 .实验结果表明该方法速度快、识别率高 .

This paper proposes a fast, one stroke pen gesture recognition approach to the studying of multimodal human computer interaction theory and building method. In the approach, a pen gesture is characterized by a sequence of dominant points along the gesture trajectory and a sequence of writing directions between consecutive dominant points. The recognition result can be obtained by matching the feature code of the input gesture with the various possible feature codes of each standard gesture. The directional feature is used for gesture pre-classification and the positional information is used for fine classification. Experimental results show that this approach is fast and can get a high recognition rate.