基于碎片模型的弱纹理物体位姿估计
Pose Estimation of Weak Textured Objects Based on Fragment Model
摘要: 位姿估计已广泛应用于智能机器人，无人驾驶以及增强现实等多种应用场景，现有基于神经网路的点对匹配方法大多未能处理好前景遮挡，弱纹理以及对称。本文提出了一种基于碎片模型的弱纹理物体位姿估计方法，该方法利用现有的3D碎片模型从一幅RGB输入图像中估计出刚体物体的6D姿态；物体由紧凑表面碎片表示，能够系统地处理物体对称；使用编码器–解码器网络预测密集采样的像素和片段之间的对应关系；还提出了一种通用的数据合成方案，创建了高相似度的弱纹理物体数据集；最后，设计并改进了一种PNP-RANSAC算法稳健而有效地估计可能多个对象实例的姿态，并在3个弱纹理数据集上进行对比实验并验证了该方法的有效性。
Abstract: Pose estimation has been widely used in intelligent robot, unmanned driving, augmented reality and other application scenarios. Most of the existing point pair matching methods based on neural network failed to deal with foreground occlusion, weak texture and symmetry. In this paper, a pose estimation method for weakly textured objects based on the fragment model is proposed. This method uses the existing 3D fragment model to estimate the 6D pose of rigid bodies from an RGB input image. Objects are represented by compact surface fragments, which can systematically deal with object symmetry. The encoder-decoder network is used to predict the correspondence between pixels and fragments of dense sampling. A general data synthesis scheme is proposed to create a weak textured object dataset with high similarity. Finally, a PNP-RANSAC algorithm is designed and improved for robust and efficient attitude estimation of multiple object instances, and the effectiveness of the method is verified by comparison experiments on three weakly textured datasets.
文章引用：李耀, 程良伦, 王涛. 基于碎片模型的弱纹理物体位姿估计[J]. 计算机科学与应用, 2022, 12(1): 252-261. https://doi.org/10.12677/CSA.2022.121025

参考文献

[1] 涂文哲. 基于合成样本的弱纹理物体6D位姿估计[D]: [硕士学位论文]. 成都: 电子科技大学, 2020.
[2] Roberts, L.G. (1963) Machine Perception of Three-Dimensional Solids. Massachusetts Institute of Technology, Cam-bridge.
[3] Fischler, M.A. and Bolles, R.C. (1981) Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography. Communications of the ACM, 24, 381-395.
[4] Lepetit, V., Moreno-Noguer, F. and Fua, P. (2009) Epnp: An Accurate O(n) Solution to the PnP Problem. International Journal of Computer Vision, 81, Article No. 155. [Google Scholar] [CrossRef] 
[5] Lowe, D.G. (1999) Object Recognition from Local Scale-Invariant Features. Proceedings of the Seventh IEEE International Conference on Computer Vision, Kerkyra, 20-27 September 1999, 1150-1157. [Google Scholar] [CrossRef] 
[6] 张昊若. 面向机器人抓取的弱纹理物体六自由度位姿估计方法研究[D]: [博士学位论文]. 上海: 上海交通大学, 2019.
[7] Du, G.G., Wang, K., Lian, S.G., et al. (2020) Vi-sion-Based Robotic Grasping from Object Localization, Object Pose Estimation to Grasp Estimation for Parallel Grippers: A Review. Artificial Intelligence Review, 54, 1677-1734. [Google Scholar] [CrossRef] 
[8] Xiang, Y., Schmidt, T., Narayanan, V., et al. (2017) PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes. [Google Scholar] [CrossRef] 
[9] Hinterstoisser, S., Holzer, S., Cagniart, C., et al. (2011) Mul-timodal Templates for Real-Time Detection of Texture-Less Objects in Heavily Cluttered Scenes. 2011 International Conference on Computer Vision, Barcelona, 6-13 November 2011, 858-865. [Google Scholar] [CrossRef] 
[10] Denninger, M., Sundermeyer, M., Winkelbauer, D., et al. (2019) BlenderProc.
https://arxiv.org/abs/1911.01911
[11] Liu, Z., Lin, Y., Cao, Y., et al. (2021) Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows.
https://arxiv.org/abs/2103.14030
[12] Hodan, T., Michel, F., Brachmann, E., et al. (2018) BOP: Benchmark for 6D Object Pose Estimation. Computer Vision—ECCV 2018, Munich, 8-14 September 2018, 19-35. [Google Scholar] [CrossRef] 
[13] Li, Z.G., Wang, G. and Ji, X.Y. (2019) CDPN: Coordi-nates-Based Disentangled Pose Network for Real-Time RGB-Based 6-Dof Object Pose Estimation. Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, 27 October-2 November 2019, 7678-7687. [Google Scholar] [CrossRef] 
[14] Labbé, Y., Carpentier, J., Aubry, M., et al. (2020) Cosypose: Consistent Multi-View Multi-Object 6D Pose Estimation. European Conference on Computer Vision, Glasgow, 23-28 August 2020, 574-591. [Google Scholar] [CrossRef] 
[15] Hodan, T., Barath, D. and Matas, J. (2020) EPOS: Estimating 6D Pose of Objects with Symmetries. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, 13-19 June 2020, 11703-11712. [Google Scholar] [CrossRef]