基于多尺度形状描述子的手势识别

Gesture Recognition Based on Multi-scale Shape Descriptor

随着计算机性能的提高和人机交互技术的发展,手势识别越来越受到人们的重视,尤其是基于视觉通道的手势识别,使人机交互变得更加便捷。但由于人手是复杂形变体,现有的方法对运动过程中手的形变的描述不够充分。该文从视觉角度提出了一种新的手势建模的方法-多尺度形状描述子。从分析手的基本形状入手,利用了圆形的轴对称和中心对称的几何特点,具有旋转和尺度不变性。该描述子从多个尺度对手势进行形状描述,在一定程度上解决了手势的精细区分问题。

With the improvement of computer performance and development of human computer interaction technology,gesture recognition attracts more and more people's attention,especially vision based approach which makes HCI more convenient than before.But because of the complex variance of human hand shape,people can not fully describe the hand in movement by current methods.A new hand modelling method-multiscale descriptor is presented from the vision perspective in this paper.This paper begins by analysing the basic hand shape,then implements the axis and center symmetrical property of a circle.The descriptor is rotation and scaling invariant.By describing the gesture from different scales,the descriptor solves the problem of differentiating hand shapes to some extent.