基于改进卷积神经网络结构的机器视觉室内定位算法

Indoor localization algorithm based on improved convolutional neural network architecture

针对当前基于卷积神经网络的机器视觉室内定位算法定位精度低及需要庞大的图像数据库训练神经网络等问题,提出一种改进卷积神经网络结构及基于多层迁移学习的复杂神经网络模型训练方法。新的卷积神经网络结构由一个完整的UNet、两个VGG16Net的前13层、一个VGG16Net的后3层全连接层及ArcFace分类器构成。U-Net的作用是从RGB图像中提取出语义分割图像;两个VGG16Net的前13层用于分别从RGB图像及语义分割图像中提取位置特征;VGG16Net的后3层全连接层用于融合这两类位置特征;ArcFace分类器用于获得最终的分类结果。为实现仅使用小的数据集就能训练复杂深度卷积神经网络,设计了一种基于多层迁移学习的复杂卷积神经网络训练方法。实验证明,所提算法能准确定位室内移动机器人,较基于RGB图像及基于语义分割图像的方法,定位准确率分别提高了10.7%及11.8%。

Aimed at improving the localization accuracy of current vision-based indoor localization algorithm,a new vision based indoor localization method using a creative convolutional neural network architecture and a novelty training method is proposed.It is composed by an intact U-Net,two first 13 layers of VGG16 Net,a fully connection layers of VGG16 Net and an ArcFace classifier.U-Net is applied to extract semantically segmented image from RGB image,two first 13 layers of VGG16 Nets are used to extract location features from RGB images and semantically segmented images,respectively.These location features are then combined together by the fully connection layers of VGG16 Net,ArcFace classifier is applied to obtain the final classification results.What’s more,a multi-layer transfer learning training method for complex convolutional neural networks is designed,transfer learning decreases the number of training set and the layered strategy makes the model easy to be trained.Experimental results show that the proposed algorithm can localize indoor mobile robot accurately,compared to using RGB image method and using semantic image method,the accuracy of our method increased by 10.7%and 11.8%.