无标记手势识别中基于混合特征的手部分割研究

Hand Segment Using Multi-feature Fusion Method in Unmarked Hand Gesture Recognition

结合多种算法对无标记手势动作的识别进行了研究。根据人体肤色分布特性,采用了高斯肤色模型对肤色进行了建模;针对外界光照问题,采用HSV颜色空间来表示不同的肤色;针对手部运动过程中会出现背景信息融入手部的情况,使用背景去除的Haar-like特征手部描述算法,同时研究了Ada Boost分类器进行特征分类。实验结果表明:在无标记手部分割中,采用多特征融合的方法较可以得到更好的分割效果。

Identification of non-mark gesture movement was studied by multi-algorithms. Gaussian skin model was used to model human complexion accoding to human comlexion distribution properties. HSV color space was applied to represent different skin colors. For in hand movement process, background information was incorporated in hand, algorithm of Haar- like which described hand characters by removing background was applied. Meanwhile AdaBoost classifier was explored to classify characters. The results of experiment show that in non-mark hand segmenting, muhi-feature combination method can achieve better segment results.