无监督单目图像深度和位姿估计算法优化  

Optimization of algorithm for unsupervised monocular image depth and pose estimation

作　　者：罗志斌 项志宇[1] 刘磊[1] LUO Zhibin;XIANG Zhiyu;LIU Lei(College of Information Science and Electronic Engineering,Zhejiang University,Hangzhou 310007,China)

机构地区：[1]浙江大学信息与电子工程学院,浙江杭州310007

出　　处：《传感器与微系统》2022年第7期110-113,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金-浙江两化融合联合基金资助项目(U1709214)

摘　　要：针对基于深度学习的无监督单目图像深度和位姿估计模型存在的问题,如训练约束条件单一,动态场景难以处理等,在最新模型UnDEMoN的基础上,通过引入视差上采样、深度一致性约束和多种掩模等优化措施,进一步优化了UnDEMoN模型。实验结果表明:提出的优化措施大幅提高了UnDEMoN模型的深度和位姿估计效果,验证了所引入优化措施的有效性。Aiming at the problems of unsupervised monocular depth and pose estimation model based on deep learning,such as single training constraints and difficult to deal with dynamic scenes,on the basis of the latest model UnDEMoN,by introducing some effective optimization measures,including disparity upsampling,depth consistency constraint and various masks,further optimize UnDEMoN model.Experimental results show that the optimization measures greatly improve the performance of UnDEMoN model,which finally verifies the effectiveness of the proposed measures.

关 键 词：无监督学习 深度估计 位姿估计 优化UnDEMoN模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]