基于多列深度3D卷积神经网络的手势识别

发布时间：2018-01-04 16:37

本文关键词：基于多列深度3D卷积神经网络的手势识别　出处：《计算机工程》2017年08期 　论文类型：期刊论文

【摘要】：传统2D卷积神经网络对于视频连续帧图像的特征提取容易丢失目标时间轴上的运动信息,导致识别准确度较低。为此,提出一种基于多列深度3D卷积神经网络(3D CNN)的手势识别方法。采用3D卷积核对连续帧图像进行卷积操作,提取目标的时间和空间特征捕捉运动信息。为避免因单组3D CNN特征提取不充分而导致的误分类,训练多组具有较强分类能力的3D CNN结构组成多列深度3D CNN,该结构通过对多组3D CNN的输出结果进行权衡,将权重最大的类别判定为最终的输出结果。实验结果表明,将多列深度3D CNN应用于CHGDs数据集上进行手势识别,识别率达到95.09%,与单组3D CNN及传统2D CNN相比分别提高近7%,20%,对连续图像目标识别具有较好的识别能力。
[Abstract]:The traditional 2D convolution neural network is easy to lose the moving information on the target time axis for feature extraction of video continuous frame image, which leads to low recognition accuracy. This paper presents a method of hand gesture recognition based on multi-column depth 3D convolution neural network (3D CNN), which uses 3D convolution to check continuous frame images for convolution operation. The temporal and spatial features of the target are extracted to capture the moving information. In order to avoid the false classification caused by the insufficient extraction of the single set of 3D CNN features. The multi-group 3D CNN structure with strong classification ability is trained to form the multi-column depth 3D CNN. The structure tradeoffs the output results of the multi-group 3D CNN. The results show that the multi-column depth 3D CNN is applied to the CHGDs dataset for gesture recognition, and the recognition rate is 95.09%. Compared with single group of 3D CNN and traditional 2D CNN, it can be improved by nearly 7 / 20 and has better recognition ability for continuous image target recognition.
【作者单位】：长安大学电子与控制工程学院;
【基金】：国家自然科学基金青年基金(61203374) 陕西省自然科学基金国际合作项目(2014KW01-05)
【分类号】：TP183;TP391.41
【正文快照】： 中文引用格式:易生,梁华刚,茹锋.基于多列深度3D卷积神经网络的手势识别[J].计算机工程,2017,43(8):243-248.英文引用格式:Yi Sheng,Liang Huagang,Ru Feng.Hand Gesture Recognition Based on Multi-column Deep 3DConvolutional Neural Netw ork[J].Computer Engineering,20

【相似文献】