基于自适应卷积和潜在表征的航天器管路位姿估计方法

doi:10.3901/JME.2025.14.150

摘要/Abstract

摘要： 为保证航天器中管路系统装配的高精度、高可靠性和高效率，精确的目标检测和位姿估计是实现管路系统自动化装配的根本前提。然而，管路无纹理且形状复杂，易受光照、遮挡影响，传统基于纹理和简单几何特征的方法精度低。为解决该问题，提出一种基于自适应卷积和潜在表征的航天器管路位姿估计方法。首先，利用域随机化生成丰富多样且真实的合成数据，有效解决管路大规模数据集获取困难和标注耗时费力的问题。然后，针对管路形状复杂导致的分割精度低的问题，提出基于自适应卷积的管路实例分割网络，根据管路的复杂几何形状进行动态调整，更好地捕捉局部特征和空间关系，并结合结构化剪枝方法优化网络结构，有效提高管路分割的准确与效率。最后，针对传统位姿特征表达有限、容易受到环境变化影响的问题，设计仅对位姿变换敏感而对其他因素鲁棒的潜在位姿表征，融合监督学习和自监督学习机制完成管路位姿的初始估计，并进一步采用基于边缘特征的位姿优化算法提高位姿估计精度。试验结果显示，对比现有的位姿估计方法，所提出的方法能实现更精确的管路分割与位姿估计，分割准确率为99.2%，速度为32.8 帧/s，位姿精度为2.445 mm和1.074°，满足管路自动化装配中目标检测和位姿估计的要求。

关键词: 位姿估计, 实例分割, 域随机化, 自适应卷积, 潜在位姿特征

Abstract: To ensure high precision, reliability and efficiency in spacecraft pipe system assembly, accurate target detection and pose estimation are essential. However, the untextured and complex geometry, as well as susceptibility to light and occlusion, leads to low accuracy of traditional texture- and geometry-based methods. To solve these challenges, a novel pipe pose estimation method based on adaptive convolution and latent representation is proposed. Firstly, diverse and realistic synthetic datasets are generated using domain random, mitigating challenges in large-scale data acquisition and annotation. Then, to address the problem of low segmentation accuracy due to the complex geometry of pipes, an adaptive convolution-based pipe instance segmentation network is proposed, dynamically adjusting to pipe complex geometry to better capture key features and spatial relationships. Structured pruning further optimizes network efficiency and segmentation accuracy. Finally, to address the problem that traditional positional features are limited in expression and susceptible to environmental changes, a novel latent pose representation is designed, which is only sensitive to pose transformation but robust to other factors. Supervised and self-supervised learning are integrated for initial pose estimation, followed by an edge feature-based optimization algorithm to enhance accuracy. Experimental results show that compared with the existing pose estimation methods, our proposed methods achieve more accurate pipe segmentation and pose estimation, with a segmentation accuracy of 99.2%, a speed of 32.8 FPS, and pose accuracies of 2.445 mm and 1.074°, which meets the requirements of target detection and pose estimation in automated pipe assembly.

Key words: pose estimation, instance segmentation, domain randomization, adaptive convolution, latent pose representation

中图分类号:

TH164
TH166

胡佳, 刘检华, 刘少丽, 刘金山, 王家修, 孙连胜. 基于自适应卷积和潜在表征的航天器管路位姿估计方法[J]. 机械工程学报, 2025, 61(14): 150-165.

HU Jia, LIU Jianhua, LIU Shaoli, LIU Jinshan, WANG Jiaxiu, SUN Liansheng. Pose Estimation Method for Spacecraft Pipe Based on Adaptive Convolution and Latent Representation[J]. Journal of Mechanical Engineering, 2025, 61(14): 150-165.

参考文献

[1] LIU Qiang，WANG Chengen. Pipe-assembly approach for aero-engines by modified particle swarm optimization[J]. Assembly Automation，2010，30(4):365-377.
[2] ZHANG Z，SAADAT M. Multi-objective grasp pose optimisation for robotic 3D pipe assembly manipulation[J]. Robotics and Computer-Integrated Manufacturing，2022，76:102326.
[3] HU Jia，LIU Jianhua，LIU Shaoli，et al. Pipe pose estimation based on machine vision[J]. Measurement，2021，182:109585.
[4] 原晓佩，陈小锋，廉明. 基于Haar-like和LBP的多特征融合目标检测算法[J]. 计算机科学，2021，48(11):219-225. YUAN Xiaopei，CHEN Xiaofeng，LIAN Ming. Improved multi-feature fusion algorithm for target detection based on Haar-like and LBP[J]. Computer Science，2021，48(11):219-225.
[5] 宋敏敏，周泽亚，邱燕，等. 基于HOG特征SVM分类器的红外图像智能检测与分类方法[J]. 红外，2022，43(4):25-32. SONG Minmin，ZHOU Zeyan，QIU Yan，et al. Intelligent detection and classification method of infrared image based on HOG feature and SVM classifier[J]. Infrared，2022，43(4):25-32.
[6] 陈泽瑜，李向国，曹登锋，等. 面向无序抓取的DPC聚类多目标检测方法研究[J]. 计算机工程与应用，2023，59(23):175-182. CHEN Zeyu，LI Xiangguo，CAO Dengfeng，et al. Reaserch on DPC clustering multi-objective detection method for disorderly grasping[J]. Computer Engineering and Applications，2023，59(23):175-182.
[7] 郑联语，付强，樊伟，等. 基于双目视觉和先验加工数据的大型筒件原位位姿感知方法[J]. 机械工程学报，2023，59(11):129-146. ZHENG Lianyu，FU Qiang，FAN Wei，et al. In-situ pose measurement method for large cylinders based on binocular vision and prior processing data[J]. Journal of Mechanical Engineering，2023，59(11):129-146.
[8] 张亚炜，付东翔. 基于双向融合纹理和深度信息的目标位姿检测[J]. 数据采集与处理，2024，39(5):1214-1227. ZHANG Yawei，FU Dongxiang. Target position detection based on bidirectional fusion of texture and depth information[J]. Journal of Data Acquisition and Processing，2024，39(5):1214-1227.
[9] GAIDON A，WANG Q，CABON Y，et al. Virtual worlds as proxy for multi-object tracking analysis[C]// IEEE Conference on Computer Vision and Pattern Recognition，Las Vegas:CVPR，2016:4340-4349.
[10] HU Y T，CHEN H S，HUI K，et al. SAIL-VOS:Semantic amodal instance level video object segmentation-a synthetic dataset and baselines[C]// IEEE Conference on Computer Vision and Pattern Recognition，Long Beach:CVPR，2019:3105-3115.
[11] FABBRI M，LANZI F，CALDERARA S，et al. Learning to detect and track visible and occluded body joints in a virtual world[C]// European Conference on Computer Vision，Munich:ECCV，2018:430-446.
[12] TREMBLAY J，PRAKASH A，ACUNA D，et al. Training deep networks with synthetic data:bridging the reality gap by domain randomization[C]// IEEE International Conference on Computer Vision Pattern Recognition，Salt Lake City:CVPR，2018:969-977.
[13] 郑义桀，罗健欣，陈卫卫，等. 基于Unity3D三维多视角虚拟数据集构建[J]. 计算机技术与发展，2023，33(5):173-179. ZHENG Yijie，LUO Jianxin，CHEN Weiwei，et al. 3D multi-view virtual dataset construction based on Unity3D[J]. Computer Technology and Development，2023，33(5):173-179.
[14] ABOUELYAZID M. Reinforcement learning-based approaches for improving safety and trust in robot-to-robot and human-robot interaction[J]. Advances in Urban Resilience and Sustainable City Design，2024，16(2):18-29.
[15] 郎宁，王德成，程鹏. 基于集成自适应欠采样的铝管表面缺陷检测方法研究[J]. 机械工程学报，2023，59(6):18-31. LANG Ning，WANG Decheng，CHENG Peng. Research on surface defect detection method of aluminum tube based on ensemble adaptive undersampling[J]. Journal of Mechanical Engineering，2023，59(6):18-31.
[16] DANG L M，WANG H，LI Y，et al. Lightweight pixel-level semantic segmentation and analysis for sewer defects using deep learning[J]. Construction and Building Materials，2023，371:130792.
[17] LI Y，WANG H，DANG L M，et al. A robust instance segmentation framework for underground sewer defect detection[J]. Measurement，2022，190:110727.
[18] 唐若仪，陈成军，王金磊，等. 基于TD-Mask R-CNN的机械装配体图像实例分割[J]. 组合机床与自动化加工技术，2024(4):135-140. TANG Ruoyi，CHEN Chengjun，WANG Jinlei，et al. Instance segmentation of mechanical assembly image based on TD-Mask R-CNN[J]. Modular Machine Tool & Automatic Manufacturing Technique，2024(4):135-140.
[19] JIANG Yang，CHANG Jiaming，ZHAO Bin. Improved cascade rcnn model-based research on multi-object grasping technique for robotic arms[C]// International Symposium on Robotics & Intelligent Manufacturing Technology，Changzhou:ISRIMT，2024:139-143.
[20] 徐胜军，李康平，韩九强，等. 基于多尺度特征注意Yolact网络的堆叠工件分拣算法[J]. 计算机测量与控制，2022，30(9):184-192，200. XU Shengjun，LI Kangping，HAN Jiuqiang，et al. Stacking workpieces sorting algorithm based on multi-scale feature attention yolact network[J]. Computer Measurement & Control，2022，30(9):184-192，200.
[21] MO Lidong，QI Naiming，ZHAO Zhengqing. Spacecraft pose estimation based on different camera model[J]. Chinese Journal of Mechanical Engineering，2023，36:63.
[22] OBERWEGER M，RAD M，LEPETIT V. Making deep heatmaps robust to partial occlusions for 3d object pose estimation[C]// European Conference on Computer Vision，Munich:ECCV，2018:125-141.
[23] PENG Sida，ZHOU Xiaowei，LIU Yuan，et al. PVNet:Pixel-wise voting network for 6DoF object pose estimation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence，2022，44(6):3212-3223.
[24] KENDALL A，GRIMES M，CIPOLLA R. PoseNet:a convolutional network for real-time 6-DOF camera relocalization[C]// IEEE International Conference on Computer Vision，Santiago:ICCV，2015:2938-2946.
[25] WALCH F，HAZIRBAS C，LEAL-TAIXE L，et al. Image-based localization using LSTMs for structured feature correlation[C]// IEEE International Conference on Computer Vision，Venice:ICCV，2017:627-637.
[26] MELEKHOV I，YLIOINAS J，KANNALA J，et al. Image-based localization using hourglass networks[C]// IEEE International Conference on Computer Vision，Venice:ICCV，2017:870-877.
[27] HUNG W C，JAMPANI V，LIU S，et al. SCOPS:Self-supervised co-part segmentation[C]// IEEE Conference on Computer Vision Pattern Recognition，Long Beach:CVPR，2019:869-878.
[28] DENG X，XIANG Y，MOUSAVIAN A，et al. Self-supervised 6d object pose estimation for robot manipulation[C]// IEEE International Conference on Robotics and Automation，Paris:ICRA，2020:3665-3671.
[29] 刘志赟，肖伟毅，任若飞. 基于Cook-Torrance光照模型改进BRDF计算方式的研究实现[J]. 电脑知识与技术，2024，20(17):123-130. LIU Zhiyun，XIAO Weiyi，REN Ruofei. Research implementation of improved BRDF calculation based on Cook-Torrance light model[J]. Computer Knowledge and Technology，2024，20(17):123-130.
[30] LINNERT D，STAVRIDIS M，NEUSCHAEFER R U，et al. Physically based rendering method to derive realistic simulation of chromatic confocal measurements[J]. Modeling Aspects in Optical Metrology IX，2023(12619):205-215.
[31] HODAN T，HALUZA P，OBDRZALEK S，et al. T-LESS:An RGB-D dataset for 6D pose estimation of texture-less objects[C]// IEEE International Conference on Computer Vision and Pattern Recognition，Honolulu:CVPR，2017:880-888.
[32] PARK S Y，SON C M，JEONG W，et al. Relative pose estimation between image object and ShapeNet CAD model for automatic 4-dof annotation[J]. Applied Sciences，2023(13):693.
[33] LIN T，DOLLAR P，GIRSHICK R B，et al. Feature pyramid networks for object detection[C]// IEEE International Conference on Computer Vision and Pattern Recognition，Las Vegas:CVPR，2016:936-944.
[34] HE K，GKIOXARI G，DOLLAR P，et al. Mask r-cnn[C]// IEEE International Conference on Computer Vision，Venice:ICCV，2017:2980-2988.
[35] CAI Z，VASCONCELOS N. Cascade R-CNN:High quality object detection and instance segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence，2021，43(5):1483-1498.
[36] VU T，KANG H，YOO C D. SCNet:Training inference sample consistency for instance segmentation[C]// AAAI Conference on Artificial Intelligence，Vancouver:AAAI，202135(3):2701-2709.
[37] WANG Xinlong，KONG Tao，SHEN Chunhua，et al. SOLO:segmenting objects by locations[C]// European Conference on Computer Vision，Glasgow:ECCV，2020:649-665.
[38] MAO A，MOHRI M，ZHONG Y. Cross-entropy loss functions:Theoretical analysis and applications[C]// International Conference on Machine Learning，Hawaii Convention Center:ICML，2023:23803-23828.
[39] MILLETARI F，NAVAB N，AHMADI S A. V-Net:Fully convolutional neural networks for volumetric medical image segmentation[C]// IEEE International Conference on Computer Vision and Pattern Recognition，Las Vegas:CVPR，2016:565-571.
[40] WOLD S，ESBENSEN K，GELADI P. Principal component analysis[J]. Chemometrics and Intelligent Laboratory Systems，1987(2):37-52.
[41] CHEN C，ZHANG C，WANG J，et al. Semantic segmentation of mechanical assembly using selective kernel convolution UNet with fully connected conditional random field[J]. Measurement，2023，209:112499.
[42] YI Qin，XI Dejun，CHEN Weiwei，et al. Gear Pitting Measurement by Multi-Scale Splicing Attention U-Net[J]. Chinese Journal of Mechanical Engineering，2023，36:50.
[43] MING Q，XIAO X. Towards accurate medical image segmentation with gradient-optimized dice loss[J]. IEEE Signal Processing Letters，2024(31):191-195.