基于监督学习的单幅图像深度估计综述  

Survey on Supervised Learning Based Depth Estimation from a Single Image

作　　者：毕天腾 刘越[1] 翁冬冬[1] 王涌天[1] Bi Tianteng;Liu Yue;Weng Dongdong;Wang Yongtian(School of Optoelectronics,Beijing Institute of Technology,Beijing 100081)

机构地区：[1]北京理工大学光电学院,北京100081

出　　处：《计算机辅助设计与图形学学报》2018年第8期1383-1393,共11页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家科技支撑计划(2015BAK01B05);国家自然科学基金(61631010)

摘　　要：单幅图像深度估计是三维重建中基于图像获取场景深度的重要技术,也是计算机视觉中的经典问题,近年来,基于监督学习的单幅图像深度估计发展迅速.文中介绍了基于监督学习的单幅图像深度估计及其模型和优化方法;分析了现有的参数学习、非参数学习、深度学习3类方法及每类方法的国内外研究现状及优缺点;最后对基于监督学习的单幅图像深度估计进行总结,得出了深度学习框架下的单幅图像深度估计是未来研究的发展趋势和重点.Depth estimation from a single image is an important technology in the image-based depth acquisitionfor 3D reconstruction, which is also a classical problem in computer vision. Recently, supervisedlearning based depth estimation from a single image develops rapidly. In this paper, the recent related literaturesare reviewed and supervised learning based depth estimation from a single image and its model andoptimization are introduced. The current research situations of the parametric learning method, non-parametriclearning method and deep learning method both in domestic and abroad are analyzed respectively with theiradvantages and disadvantages. At last, summarizing these methods leads to the conclusion that depth estimationfrom a single image in deep learning framework is the development trend and research priority in thefuture.

关 键 词：机器学习 深度估计 三维重建 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]