【摘要】 手势识别是人机交互,智能语义识别和远程人机交流领域的热门研究课题。目前基于视觉的手势识别问题仍是研究的难点,在多变背景下的手势姿态识别仍然存在较大问题。近年来,随着深度神经网络技术的快速发展,利用网络自主学习的方法来提取手势姿态有关特征得到了广泛关注。由于卷积神经网络具有较强的学习能力和个体特征的表达能力,本文针对传统手势识别算法精度低,鲁棒性差的问题,提出了基于卷积神经网络的TensorFlow框架下加入扁平卷积模块的FD-CNN网络手势识别算法。在预处理数据集后,基于FD-CNN网络的手势识别方法可以直接将预处理后的图像输入网络进行训练,最终输出测试结果的识别精度为99.0%。与传统方法和经典卷积神经网络方法相比,本文方法提高了网络系统对样本数据的多样性和复杂性的有效识别,具有较高的识别率和较好的鲁棒性效果。更多还原

【Abstract】 Gesture recognition is a hot research topic in the field of human-computer interaction.At present,the problem of gesture recognition based on vision is still a difficult point of research.There are still big problems in gesture recognition under the changing background.In recent years,with the rapid development of deep neural network technology,using network autonomous learning methods to extract gesture-related features has received extensive attention.Because the convolutional neural network has strong learning ability and individual feature expression ability,for the traditional gesture recognition algorithm with low precision and poor robustnesst,this paper based TensorFlow framework to add flattened convolution module proposes a simple convolutional neural network gesture recognition algorithm FD-CNN network.After the data set is preprocessed,the gesture recognition method based on FD-CNN network can directly input the preprocessed image into the network for training,and the recognition accuracy of the final output test result is 99.0%.Compared with the traditional method and the classical neural network,this paper method improves the network system to the sample data diversity and the complexity effective discrimination,has the higher recognition rate and the better robustness effect.更多还原