自动化与信息工程2025，Vol.46Issue(6)：9-16,8.DOI:10.12475/aie.20250602
相对位姿与姿态池优化的类级别6D姿态估计模型
Relative Pose and Pose Pool Optimized Category-level 6D Pose Estimation Model
摘要
Abstract
To address the challenges faced by 6D pose estimation in complex environments—such as heavy reliance on high-precision 3D models or large-scale template libraries,severe occlusion,and lighting interference—this paper proposes a category-level 6D pose estimation model based on relative pose and pose pool optimization.The model constructs a progressive estimation framework encompassing target object segmentation and detection,keypoint matching,relative pose calculation,and pose pool-based optimization.By computing the absolute pose of the target object through relative poses between adjacent frames,it reduces dependency on high-precision 3D models or template libraries.Furthermore,a pose pool optimization strategy is introduced,integrating self-attention and cross-attention mechanisms to fuse historical frame information,thereby enhancing the accuracy and robustness of the current pose estimation.Experimental results on public datasets NOCS and HO3D demonstrate that the model achieves an average 5°5 cm accuracy of 90.35%,an average IoU25 of 99.97%,and an average translation error of 1.72 cm on the NOCS dataset,outperforming three comparative models.The average rotation error is 1.90°,which is comparable to the BundleSDF model.On the HO3D dataset,the model achieves an ADD-S of 96.9%and an ADD of 93.4%,surpassing four comparative models.These results indicate that the model can achieve accurate and stable category-level 6D pose estimation even under weak model dependency,providing crucial support for robotic manipulation in complex environments and practical deployment.关键词
6D姿态估计/相对位姿/姿态池优化/类级别位姿估计/注意力机制Key words
6D pose estimation/relative pose/pose-pool optimization/category-level pose estimation/attention mechanism分类
信息技术与安全科学引用本文复制引用
黄天仑,胡生超,刘笑,王卫军,冯伟..相对位姿与姿态池优化的类级别6D姿态估计模型[J].自动化与信息工程,2025,46(6):9-16,8.基金项目
国家重点研发计划(2023YFB4705002) （2023YFB4705002）
广东省现代控制技术重点实验室开放基金项目(GDKLMCT202402). （GDKLMCT202402）
