基于CNN和DCGAN的手势识别的研究  

Research on Gesture Recognition Based on CNN and DCGAN

作　　者：曲宝[1] 李荟[1] Qu Bao;Li Hui(Northeast Petroleum University,Daqing 163318)

机构地区：[1]东北石油大学,大庆163318

出　　处：《黑龙江八一农垦大学学报》2021年第3期79-84,共6页journal of heilongjiang bayi agricultural university

摘　　要：近年来,由于硬件的不断改善,深度学习在视觉识别等方向有了突破性的进展,具体表现为人机交互行为频繁地出现在日常中,尤其计算机视觉方面的飞速发展导致以人为本的人机交互技术必将取代以计算机为本的人机交互技术。手势识别的研究提供了一种与人类互动的新方式,它符合上述描述的趋势。传统的手势识别方法采用人力提取特征值,费时费力。因此提出了一种新的手势识别算法,该算法基于深度卷积神经网络和深度卷积生成对抗网络,在表情识别、计算和文本输出等方面都可以使用这一方法,效果良好。实验发现该方法识别时只用了较少的样本训练模型,但依然对手势分类和检测产生了重大的影响。此外,该方法不仅可以进行有效地实时识别,而且还不易受到光照和背景干扰的影响。In recent years,due to the continuous improvement of hardware,deep learning has been a breakthrough in the visual identification of the progress,embodied in the human-computer interaction behavior frequently appeared in the daily,especially the rapid development of computer vision aspect to human-oriented man-machine interaction technology would replace the human-computer interaction technology based on computer.The study of gesture recognition offered a new way of interacting with humans that fits into the trend described above.The traditional gesture recognition method used manpower to extract feature values,which was time-consuming and laborious.Therefore,a new gesture recognition algorithm was proposed based on deep convolutional neural network and deep convolutional generating antagonistic network.This algorithm could be used in facial expression recognition,calculation and text output with good results.It was found that the method only used a small number of sample training models,but still had a significant impact on gesture classification and detection.In addition,the method was not only effective in real time recognition,but also not susceptible to the interference of light and background.

关 键 词：手势识别 人机交互 深度卷积神经网络 深度卷积生成对抗网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]