基于卷积神经网络的静态手势识别研究

Research on the Static Hand Gesture Recognition Base on Convolutional Neural Network

近年来,随着深度学习的快速发展,对于卷积神经网络的应用也越发成熟。基于卷积神经网络强大的特征学习能力,针对Thomas Moeslund手势识别数据集,构建卷积神经网络来对24种静态手语手势进行识别。在1440张训练集图像训练好的网络上对600张测试集图像进行识别,识别准确率高达98.67%。该方法可实现背景单一手势图像的精确识别。

Nowadays, the application of convolutional neural network is more and more mature with the rapid development of the deep learning. Construets the convolutional neural network for identify 24 kinds of static sign language gesture in Thomas Moeslund＇s gesture recognition data-base based on the powerful ability of eonvolutional neural network about learning. Tests 600 images on the trained network which train with 1440 images, the recognition accuracy is 98.67%. This method can accurately identify the hand gesture in a single background image.