基于三维模型的双目手势特征提取

Three-dimensional 3d model-based feature extraction of binocular gestures

提出一种基于三维模型的双目手势特征提取算法.该算法使用双摄像头,在手势的三维几何模型基本框架下,将得到的几何体的旋转角度作为提取出的手势特征.使用图像的极半径不变矩对比图像之间的相似度,并用遗传算法逐代改变三维模型中几何体的旋转角度,使其最大程度地接近真实手势,克服三维重建中速度与精度之间的平衡问题.实验证明,在简单手势的条件下,该算法可以比较精确地对手势实现三维重建,进而提取手势特征,是手势识别的基础.

A algorithm of the binocular hand gesture feature extraction was presented on a three-dimensional model.In double camera conditions as well as in the frame of three-dimensional geometrical model of hand gesture,the rotation angles of the geometrical body were used as the extracted feature of the hand gesture.The invariant moment of the polar radius of the images was used to compare the similarity among the images and a genetic algorithm was used to change the rotation angle of the geometric body from generation to generation to make it approach to the real gestures to a great extent.This method provided a good solution for balance problem between match precision and match speed during three-dimensional rebuilding.Experiments showed that,in simple gestures conditions,this algorithm could accurately reconstruct the hand gesture and extract the hand gesture features,being a basis for hand gesture recognition.