结合肤色模型和卷积神经网络的手势识别方法

发布时间：2018-03-26 18:38

本文选题：手势识别　切入点：高斯肤色模型　出处：《计算机工程与应用》2017年06期

【摘要】：在手势识别研究过程中,人工选取特征难以适应手势的多变性。提出了一种结合肤色模型和卷积神经网络的手势识别方法,对采集的不同背景下的手势图像,首先用肤色高斯模型分割出手势区域,然后采用卷积神经网络建立手势的识别模型,该模型融合了手势特征提取和分类过程,模拟视觉传导和认知,有效避免了人工特征提取的主观性和局限性。识别模型以手势区域的灰度信息为输入,同时利用权值共享和池化等技术减少网络权值个数,降低了模型的复杂度。实验结果表明,卷积神经网络(CNN)方法能够有效进行特征学习,在不同数据集下对手势的平均识别率都达到95%以上,与传统方法进行对比实验,表明该方法具有较高的识别率和实时性。
[Abstract]:In the process of gesture recognition, artificial feature selection is difficult to adapt to the variety of gestures. A combination of skin color model and convolutional neural network method of gesture recognition, gesture image acquisition under the background of different, firstly divided the gesture area with color Gauss model, then the model recognition gesture convolutional neural network. The model combines the feature extraction and classification process, simulation of visual conduction and cognition, effectively avoids the subjectivity and limitation of artificial feature extraction. Recognition model by gray information of the gesture area as input, using weight sharing and pooling technology to reduce the number of network weights, reduce the complexity of model experiment. The results show that the convolution neural network (CNN) method can effective learning characteristics in different data sets, the average for the gesture recognition rate of over 95%, and the traditional party The comparison experiment shows that the method has high recognition rate and real time.

【作者单位】：昆明理工大学信息工程与自动化学院;
【基金】：国家自然科学基金(No.61263017) 云南省自然科学基金(No.2011FZ060,No.KKSY201303120)
【分类号】：TP391.41;TP183

【相似文献】