APP下载

基于特征融合的手势识别

2021-12-07陈忠辉王等准万广方洪波黄以卫谢本亮

智能计算机与应用 2021年7期

陈忠辉 王等准 万广 方洪波 黄以卫 谢本亮

摘 要: 在手势识别的过程中,手势的多样性和复杂性会对识别的可靠性和准确性带来较大影响。基于视觉的手势识别通常采取单一特征用于手势分类,但是单一特征无法较好地描述整个图像。因此本文提出多种特征融合的方法,分别提取改进后的梯度方向直方图(HOG)特征和MB-LBP特征,并进行特征融合,结合支持向量机(SVM)分类器完成手势图像的识别。实验结果表明,提取的融合特征包含手势图像的局部区域梯度信息和图像的纹理信息,可以更加全面地描述图像的手势特征。相较于单一特征识别方法而言,基于特征融合的方法有着更高的识别率。

关键词: 手势识别; MB-LBP; 改进HOG特征

文章编号: 2095-2163(2021)07-0212-05中图分类号:TP391.4文献标志码: A

Gesture recognition based on feature fusion

CHEN Zhonghui1, WANG Dengzhun1, WAN Guang1, FANG Hongbo1, HUANG Yiwei2, XIE Benliang1

(1 College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China;

2 West China  Modernization Research Center, Guizhou University of Finance and Economics, Guiyang 550025, China)

【Abstract】In the process of gesture recognition, the diversity and complexity of gestures will have a greater impact on the reliability and accuracy of recognition. Vision-based gesture recognition usually adopts a single feature for gesture classification, but a single feature cannot describe the entire image well. Therefore, this paper proposes a variety of feature fusion methods, respectively extracting the improved histogram of gradient (HOG) features and MB-LBP features, and performing feature fusion, combined with SVM classifier to complete the recognition of gesture images. Experimental results show that the extracted fusion features include the local area gradient information of the gesture image and the texture information of the image, which can describe the gesture features of the image more comprehensively. Compared with the single feature recognition method, the method based on feature fusion has a higher recognition rate.

【Key words】gesture recognition; MB-LBP; improved HOG features

0 引 言

手势是一种十分重要的沟通交流方式,若能对手势进行成功识别就必将为人类与机器、设备和计算机的交互开创出崭新局面。随着计算机技术、尤其是增强现实[1],辅助驾驶[2]等技术的飞速发展,通过手势识别进行人机交互[3]的需求越来越广泛。手势识别已然成为当前的热点研究方向之一,因此手势识别技术的研究具有十分重要的现实意义。

研究可知,基于视觉的手势识别包括着3个基本步骤:手势分割[4]、手势特征提取[4]、手势识别[5]。其中,特征提取对手势识别的准确率有着至关重要的作用。当前,常见的手势特征通常包括了形状特征和纹理特征。具体来说,形狀特征主要有:梯度直方图[6](HOG)和Haar-like[7]特征;纹理特征有局部二值图[8](LBP)和Gabor[9]特征。……

登录APP查看全文