基于编解码卷积神经网络的单张图像深度估计  

Single Image Depth Estimation Based on Encoder-Decoder Convolution Neural Network

作　　者：贾瑞明 刘立强 刘圣杰 崔家礼 JIA Rui-ming;LIU Li-qiang;LIU Sheng-jie;CUI Jia-li(School of Information Science and Technology,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学信息学院

出　　处：《图学学报》2019年第4期718-724,共7页Journal of Graphics

基　　金：北京市教委面上基金(KM201510009005);北方工业大学学生科技活动项目(110051360007)

摘　　要：针对传统方法在单目视觉图像深度估计时存在鲁棒性差、精度低等问题,提出一种基于卷积神经网络(CNN)的单张图像深度估计方法。首先,提出层级融合编码器-解码器网络,该网络是对端到端的编码器-解码器网络结构的一种改进。编码器端引入层级融合模块,并通过对多层级特征进行融合,提升网络对多尺度信息的利用率。其次,提出多感受野残差模块,其作为解码器的主要组成部分,负责从高级语义信息中估计深度信息。同时,多感受野残差模块可灵活地调整网络感受野大小,提高网络对多尺度特征的提取能力。在NYUD v2数据集上完成网络模型有效性验证。实验结果表明,与多尺度卷积神经网络相比,该方法在精度δ<1.25上提高约4.4%,在平均相对误差指标上降低约8.2%。证明其在单张图像深度估计的可行性。Focusing on the poor robustness and lower accuracy in traditional methods of estimating depth in monocular vision,a method based on convolution neural network (CNN) is proposed for predicting depth from a single image.At first,fused-layers encoder-decoder network is presented.This network is an improvement of the end-to-end encoder-decoder network structure.Fused-layers block is added to encoder network,and the network utilization of multi-scale information is improved by this block with fusing multi-layers feature.Then,a multi-receptive field res-block is proposed,which is the main component of the decoder and used for estimating depth from high-level semantic information.Meanwhile,the network capacity of multi-scale feature extraction is enhanced because the size of receptive field is flexible to change in multi-receptive field res-block.The validation of proposed network is conducted on NYUD v2 dataset,and compared with multi-scale convolution neural network,experimental results show that the accuracy of proposed method is improved by about 4.4% in δ<1.25 and average relative error is reduced by about 8.2%.The feasibility of proposed method in estimating depth from a single image is proved.

关 键 词：CNN 编码器-解码器 深度估计 单目视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]