对偶树复小波与空域信息的手势识别分类研究

PDF下载HTML

Title:

Research on gesture recognition and classification of dual-tree complex wavelet and spatial information

作者:

贾鹤鸣¹, 朱传旭¹, 张森¹, 杨泽文², 何东旭²

1. 东北林业大学 机电工程学院, 黑龙江 哈尔滨 150040;
2. 哈尔滨工程大学 自动化学院, 黑龙江 哈尔滨 150001

Author(s):

JIA Heming¹, ZHU Chuanxu¹, ZHANG Sen¹, YANG Zewen², HE Dongxu²

1. College of Mechanical and Electrical Engineering, Northeast Forestry University, Harbin 150040, China;
2. College of Automation, Harbin Engineering University, Harbin 150001, China

关键词:

手势识别; 空域特征; 对偶树复小波; 特征融合; 分类器优化; BD-SVM; 径向基核函数; 静态测试

Keywords:

gesture recognition; spatial feature; dual-tree complex wavelet; feature fusion; classifier optimization; BD-SVM; radial basis kernel function; static test

分类号:

TP273

DOI:

10.11992/tis.201708003

摘要:

为提高手势识别中特征获取的有效性，本文提出空域特征与对偶树复小波变换特征相结合的融合特征，主要包括水平位置、竖直位置、长宽比、矩形度、Hu矩7个分量，及11维空域特征与对偶树复小波变换的16维特征进行融合后得到的27维特征。针对分类器优化算法，提出进行训练样本优选的最优距离-支持向量机(BD-SVM)分类方法。最后的实验结果表明，对“1~9”手势进行测试，当采用径向基核函数时，平均识别精度最高，为90.33%，平均识别时间为0.026 s，说明所提出的方法能够较好地进行静态手势识别，具有较高的训练速度和辨识精度。

Abstract:

To improve the validity of features obtained in gesture recognition, in this paper, we propose a fusion feature that combines spatial and dual-tree complex wavelet transform features. These features mainly include seven components (horizontal position, vertical position, aspect ratio, rectangular degree, Hu moments, etc.) and 27 dimensional features, comprising 11 dimensional spatial features and 16 dimensional dual-tree complex wavelet transform features. We employ the optimal distance support vector machine (BD-SVM) classification method to optimize training samples for the classifier optimization algorithm. The experimental results show that, in a test of gestures “1~9” using the RBF kernel function, the highest average recognition accuracy is 90.33% and the average recognition time is 0.026 s. These results reveal that the proposed method demonstrates excellent static gesture recognition, a high training speed, and accuracy in identification.