摘要
为了提高手势图像识别率,建立一种基于卷积神经网络的手势识别网络。将更快的区域神经网络(faster region convolutional neural network,Faster R-CNN)修改为残差网络(residual network 50,ResNet50),利用区域建议网络生成的候选框和特征图进行兴趣区域操作,提取目标手势的手势分割框;将手势分割框输入到全连接层为1×1卷积核的视觉几何组(visual geometry group,VGG)16中,同时修改激活函数Relu为LeakyRelu,经过参数调节和测试训练进行手势图像特征提取和识别。实验结果表明,该网络在手势识别上的效果更好,识别率高达97.57%。
In order to improve the recognition rate of gesture images,a gesture recognition network based on convolutional neural network was established.Faster region convolutional neural network(Faster R-CNN)is modified to residual network(residual network 50,ResNet50),using candidate boxes and feature maps generated by the region suggestion network to perform region of interest operations and extract targets Gesture segmentation box for gestures;input the gesture segmentation box into the visual geometry group(VGG)16 with a fully connected layer of 1×1 convolution kernel,and modify and activate the function Relu to LeakyRelu at the same time,after parameter adjustment and testing Training for gesture image feature extraction and recognition.Experimental results show that the network has a better effect on gesture recognition,and the recognition rate is as high as 97.57%.
作者
官巍
马俊峰
马力
GUAN Wei;MA Junfeng;MA Li(School of Computer Science and Technology, Xi'an University of Posts and Telecommunications, Xi'an 710121,China)
出处
《西安邮电大学学报》
2019年第6期80-84,共5页
Journal of Xi’an University of Posts and Telecommunications
基金
陕西省自然科学基金资助项目(2016JM6085)
陕西省产学研协同创新计划资助项目(2017XT-028)。
关键词
人机交互
机器学习
神经网络
手势识别
human-computer interaction
machine learning
neural network
gesture recognition
