基于不确定性单目图像自监督场景深度估计  

Self-supervised scene depth estimation for monocular images based on uncertainty

作　　者：柴国强[1] 薄祥仕 刘海军[2] 芦宾 王大为[1] CHAI Guoqiang;BO Xiangshi;LIU Haijun;LU Bin;WANG Dawei(School of Physics and Information Engineering,Shanxi Normal University,Taiyuan 030000,China;School of Microelectronics and Communication Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]山西师范大学物理与信息工程学院,太原030000 [2]重庆大学微电子与通信工程学院,重庆400044

出　　处：《北京航空航天大学学报》2024年第12期3780-3787,共8页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家自然科学基金(62201333,62001063);山西省基础研究计划(20210302124647);山西省高等学校科技创新项目(2021L269)。

摘　　要：深度信息对于准确理解场景三维结构、分析图像中物体之间的三维关系具有重要作用。结合运动恢复结构、图像重投影和不确定性理论,以端到端的形式提出一种基于不确定性单目图像自监督深度估计算法。利用基于改进稠密连接模块的编码器-解码器深度估计网络得到目标图像的深度图,利用位姿估计网络计算出拍摄目标图像和源图像2个时刻相机位置转换矩阵;根据图像重投影对源图像进行逐像素采样,得到重构目标图像;结合重构目标函数、不确定性目标函数和平滑目标函数对所提算法网络进行优化训练,通过使重构图像和真实目标图像差异最小化实现自监督的深度信息估计。实验结果表明:所提算法在客观指标与主观视觉对比上均取得了比竞争合作估计算法(CC)、Monodepth2、Hr-depth等主流算法更好的深度估计结果。Depth information plays an important role in accurately understanding the three-dimensional scene structure and the three-dimensioual relationship between objects in images.An end-to-end self-supervised depth estimation algorithm based on uncertainty for monocular images was proposed in this paper by combining structure-from-motion,image reprojection,and uncertainty theory.The depth map of the target image was obtained by the encoder-decoder depth estimation network based on an improved densely connected module,and the transformation matrix of camera positions for shooting the target image and source image was calculated by the pose estimation network.Then,the source image was sampled pixel by pixel according to the image reprojection to obtain the reconstructed target image.The proposed algorithm was optimized by the reconstructed objective function,uncertain objective function,and smooth objective function,and the self-supervised depth information estimation was realized by minimizing the difference between the reconstructed image and the real target image.Experimental results show that the proposed algorithm achieves better depth estimation effects than the mainstream algorithms such as competitive collaboration estimation algorithm(CC),Monodepth2,and Hr-depth in terms of both objective indicators and subjective visual comparison.

关 键 词：深度估计 深度学习 自监督 图像重投影 不确定性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]