基于误差分量模型的两阶段深度校正算法  

Two-stage depth correction algorithm based on error component modeling

作　　者：李平 喻洪流[1] Li Ping;Yu Hongliu(Institute of Rehabilitation Engineering&Technology,University of Shanghai for Science&Technology,Shanghai 200093,China;Dept.of Biomedical Engineering,Changzhi Medical College,Changzhi Shanxi 046000,China)

机构地区：[1]上海理工大学康复工程与技术研究所,上海200093 [2]长治医学院生物医学工程系,山西长治046000

出　　处：《计算机应用研究》2025年第2期523-529,共7页Application Research of Computers

基　　金：国家重点研发计划资助项目(2022YFC3601400);山西省高等学校科技创新资助项目(2022L376);长治医学院博士科研启动基金资助项目(2024BS12)。

摘　　要：为了提高消费级RGB-D相机的深度精度,提出了一种基于误差分量模型的两阶段深度校正算法。该算法根据误差特性建立误差分量模型,引入从短距离到长距离迭代计算思想,设计了两阶段深度校正算法。探究了像素离散化对校正效果的影响,将算法应用于洗浴机器人中,并与其他深度校正算法进行对比。结果表明,该算法可减小深度误差,且像素离散化采样越密集,效果越好。在实际应用中,该算法仍能有效减小深度误差,与其他校正算法相比,在远距离处具有优势。该算法能有效提升消费级RGB-D相机的深度精度,且数据收集场景简单,适用于可以产生RGB图像、深度图像和点云的传感器。To improve the depth accuracy of consumer-grade RGB-D cameras,this paper proposed a two-stage depth correction algorithm based on an error component model.The algorithm established an error component model by analyzing error characteristics,employed an iterative calculation approach that progressed from short to long distances,and designed a two-stage depth correction algorithm.This paper explored the effect of pixel discretization on correction accuracy,applied the algorithm to a bathing robot,and compared its performance with that of another depth correction algorithm.The results indicate that this method effectively reduces depth errors,with better performance observed as pixel discretization becomes denser.In practical applications,the algorithm consistently minimizes depth errors and demonstrates superior performance at longer distances compared to alternative correction methods.The algorithm not only enhances the depth accuracy of consumer-grade RGB-D cameras but also features a simple data collection setup,making it suitable for sensors that generate RGB images,depth images,and point clouds.

关 键 词：消费级RGB-D相机 深度校正 局部误差 全局误差 像素离散化 深度质量评价 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]