融合跨模态特征与颜色引导的遮挡目标6D位姿估计算法

打开文本图片集

中图分类号：TP242 文献标志码：A 文章编号：1001-3695（2026）04-033-1238-07

doi：10.19734/j.issn.1001-3695.2025.07.0288

Occluded object 6D pose estimation using cross-modal feature fusion and color-guided registration

Jia Junying，AiLiang†，Lu Xin （School of Information ScienceandEngineering，Shenyang Universityof Technology，Shenyang11O87O，China）

Abstract：RGBD-based 6Dposeestimation faces challengesof insuficient multimodal feature fusionandlargeregistration errors underoclusionand viewpoint changes.Thisstudyproposedamultimodal 6Dposeestimation algorithm forocclusion scenarios.Thealgorithmbuiltcross-modalrelationshipsbetweenRGBimagesandpointclouddataduring featureextractin.It integratedcolorfeatures intopointcloudregistration to improveaccuracyandrobustnessExperiments ontheOcclusionLineMOD dataset demonstrate superiorperformanceover existing methods acrossmultipleobject categories.The proposedapproach achieves an average accuracy improvement of 2.3% . These results indicate that the algorithm maintains high pose estimation accuracyundersevereocclusionandsignificantviewpointvariations.Itprovidesanefectivesolutionformultimodal 6D pose estimation and supports applications in robotic manipulation and autonomous driving.

Keywords：ocluded scene poseestimation；multimodalfeature fusionnetwork；pointcloudregistration；cross-modal feature modeling；robotic grasping

0引言

6D位姿估计在计算机视觉和机器人领域中作为一个关键的研究方向，其目标是有二维图像的情况下，估计目标物体在三维坐标系下的具体位置和姿态。（剩余17086字）