基于RGB-D图像的语义场景补全研究综述  

Survey on Semantic Scene Completion Based on RGB-D Images

作　　者：张康 安泊舟 李捷 袁夏[1] 赵春霞[1] ZHANG Kang;AN Bo-Zhou;LI Jie;YUAN Xia;ZHAO Chun-Xia(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学计算机科学与工程学院,江苏南京210094

出　　处：《软件学报》2023年第1期444-462,共19页Journal of Software

基　　金：国家自然科学基金(61773210)。

摘　　要：近年来随着计算机视觉领域的不断发展,三维场景的语义分割和形状补全受到学术界和工业界的广泛关注.其中,语义场景补全是这一领域的新兴研究,该研究以同时预测三维场景的空间布局和语义标签为目标,在近几年得到快速发展.对近些年该领域提出的基于RGB-D图像的方法进行了分类和总结.根据有无使用深度学习将语义场景补全方法划分为传统方法和基于深度学习的方法两大类.其中,对于基于深度学习的方法,根据输入数据类型将其划分为基于单一深度图像的方法和基于彩色图像联合深度图像的方法.在对已有方法分类和概述的基础上,对语义场景补全任务所使用的相关数据集进行了整理,并分析了现有方法的实验结果.最后,总结了该领域面临的挑战和发展前景.In recent years,with the continuous development of computer vision,semantic segmentation and shape completion of 3D scene have been paid more and more attention by academia and industry.Among them,semantic scene completion is emerging research in this field,which aims to simultaneously predict the spatial layout and semantic labels of a 3D scene,and has developed rapidly in recent years.This study classifies and summarizes the methods based on RGB-D images proposed in this field in recent years.These methods are divided into two categories based on whether deep learning is used or not,which include traditional methods and deep learning-based methods.Among them,the methods based on deep learning are divided into two categories according to the input data type,which are the methods based on single depth image and the methods based on RGB-D images.Based on the classification and overview of the existing methods,the relevant datasets used for semantic scene completion task are collated and the experimental results are analyzed.Finally,the challenges and development prospects of this field are summarized.

关 键 词：三维场景 语义场景补全 环境理解 计算机视觉 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]