基于卷积神经网络的手势识别研究

Research on CNN-based Gesture Recognition

在手势识别的过程中,手势的多样性和复杂性会对识别的可靠性和准确性带来较大影响。为了能够提高手势识别的识别速度和准确率。本文使用Google的开源Tensorflow框架构建手势识别模型,介绍了Tensorflow的平台特征,并提出了基于Tensorflow框架的卷积网络模型。该实验的数据集是结合已有的的数据集和自收集的数据集进行设计的。实验结果表明,该模型具有较高的识别精度,较高的计算效率,较强的鲁棒性等特点,可以轻松调整网络结构,快速找到最优模型,较好地完成手势识别任务。

In the process of gesture recognition,reliability and accuracy of the diversity and complexity of gesture recognition will bring a greater impact. In order to improve the recognition speed and accuracy of the gesture recognition. In this paper,the use of Google’ s newest open-source framework for buildingTensorflow gesture recognition model,introduced the platform features Tensorflow,and proposed a convolution Tensorflow network model based on the framework. The experimental data set is combined with the existing data sets and data sets collected from the design. Experimental results show that the model has higher recognition accuracy,higher calculation efficiency,and stronger robustness. It can easily adjust the network structure,quickly find the optimal model,and complete the gesture recognition task well.