摘要
随着物联网技术的发展,手势识别在当今的人机交互中起着至关重要的作用。针对复杂背景下手势识别率低、算法鲁棒性差的问题,提出了一种基于神经网络手势识别方法对26个英文字母实现静态手势识别,该算法由手势检测和特征提取及识别3部分构成。在手势检测部分,解决手势区域提取困难的问题;在手势特征提取部分,通过肤色检测提取出手的轮廓信息的二值图像;在识别阶段,使用从LeNet-5改进的CNN来识别手势。在自己制作的数据集下对神经网络进行训练,最终获得较高的识别率;并在NUS-II和Marcel两个复杂背景的公共数据集上进行了验证实验,识别率分别达到95.31%和98.10%。结果表明,该方法可以在复杂环境下对手势进行精确识别具有较高的稳定性。
With the development of Internet of things technology, gesture recognition plays a vital role in today’s human-computer interaction. Aiming at the problem of low recognition rate and poor algorithm robustness in complex background, this paper proposes a gesture recognition method based on neural network to achieve static gesture recognition of 26 English letters. The algorithm consists of gesture detection and feature extraction and recognition. In the gesture detection section, the problem of difficulty in extracting the gesture area is solved. In the gesture feature extraction section, the binary image of the contour information of the hand is extracted by the skin color detection. In the recognition phase, the gesture is recognized using the CNN improved from LeNet-5. The neural network was trained under the data set produced by ourselves, and finally the comparatively higher recognition rate was obtained. The verification experiments were carried out on common datasets of NUS-II and Marcel, which have complex background, and the recognition rates reached 95.31% and 98.10% respectively. The results show that the method can achieve high stability in the accurate recognition of gestures in complex environments.
作者
包兆华
高瑜翔
夏朝禹
郭春妮
BAO Zhaohua;GAO Yuxiang;XIA Chaoyu;GUO Chunni(College of Communication Engineering College of Microelectronics,Chengdu University of Infonnation Technology,Chengdu 610225,China)
出处
《成都信息工程大学学报》
2019年第6期606-609,共4页
Journal of Chengdu University of Information Technology
