End-to-End 3D Point Cloud Learning for Registration Task Using Virtual Correspondences

Wei, Huanshu; Qiao, Zhijian; Liu, Zhe; Suo, Chuanzhe; Yin, Peng; Shen, Yueling; Li, Haoang; Wang, Hesheng

doi:10.1109/iros45743.2020.9341249

Cited by 16 publications

(4 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deng et al [17] uses a PointNet [18] model to learn point-wise features and trains the model using an N -tuple loss. VCR-Net [19] learns point-wise feature vectors using multi-layer-perceptrons (MLPs) to extract local features, which are refined using global attention and used to identify correspondences between point clouds. In contrast, [12] uses sparse fully convolutional networks to obtain voxelwise features and trains the model using variations of triplet loss with hard negative mining.…”

Section: B Learning-based Methodsmentioning

confidence: 99%

“…Secondly, we improve the quality of the correspondences using a novel graph-based attention network that allows to efficiently combine self-and cross-information across point clouds. Differently from the feature-based attention in [19], the proposed graph attention leverages both spatial and feature dimensions of local neighbourhoods to refine point-wise features. Finally, we extend our analysis beyond existing datasets and evaluate our model performance in challenging low overlapping point clouds using a novel dataset where the translation between sensor poses vary uniformly up to 30 meters.…”

Section: B Learning-based Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Fast and Robust Registration of Partially Overlapping Point Clouds

Arnold

Mozaffari

Dianati

2022

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

Section: B Learning-based Methodsmentioning

confidence: 99%

Section: B Learning-based Methodsmentioning

confidence: 99%

Fast and Robust Registration of Partially Overlapping Point Clouds

Arnold

Mozaffari

Dianati

2022

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

“…A lot of point cloud registration approaches can be used to obtain the pose estimation results, including the traditional ICP-based approaches (Wang and Solomon, 2019) and the learning-based approaches (Wang et al , 2021). In this paper, we use the point cloud learning based approach presented in our previous work (Wei et al , 2020) for pose registration and estimation.…”

Section: System Designmentioning

confidence: 99%

Map-less long-term localization in complex industrial environments

Liu

Qiao

Suo

et al. 2021

Self Cite

View full text Add to dashboard Cite

Purpose This paper aims to study the localization problem for autonomous industrial vehicles in the complex industrial environments. Aiming for practical applications, the pursuit is to build a map-less localization system which can be used in the presence of dynamic obstacles, short-term and long-term environment changes. Design/methodology/approach The proposed system contains four main modules, including long-term place graph updating, global localization and re-localization, location tracking and pose registration. The first two modules fully exploit the deep-learning based three-dimensional point cloud learning techniques to achieve the map-less global localization task in large-scale environment. The location tracking module implements the particle filter framework with a newly designed perception model to track the vehicle location during movements. Finally, the pose registration module uses visual information to exclude the influence of dynamic obstacles and short-term changes and further introduces point cloud registration network to estimate the accurate vehicle pose. Findings Comprehensive experiments in real industrial environments demonstrate the effectiveness, robustness and practical applicability of the map-less localization approach. Practical implications This paper provides comprehensive experiments in real industrial environments. Originality/value The system can be used in the practical automated industrial vehicles for long-term localization tasks. The dynamic objects, short-/long-term environment changes and hardware limitations of industrial vehicles are all considered in the system design. Thus, this work moves a big step toward achieving real implementations of the autonomous localization in practical industrial scenarios.

show abstract

“…Specifically, the non-convexity of the optimization problem poses a significant challenge to the attainment of a globally optimal solution . When dealing with sparse and non-uniform data, traditional methods like nearest-neighbor search may not be effective, resulting in higher registration errors Wei et al 2020).…”

Section: Introductionmentioning

confidence: 99%

PosDiffNet: Positional Neural Diffusion for Point Cloud Registration in a Large Field of View with Perturbations

She,

Wang,

Kang

et al. 2024

AAAI

View full text Add to dashboard Cite

Point cloud registration is a crucial technique in 3D computer vision with a wide range of applications. However, this task can be challenging, particularly in large fields of view with dynamic objects, environmental noise, or other perturbations. To address this challenge, we propose a model called PosDiffNet. Our approach performs hierarchical registration based on window-level, patch-level, and point-level correspondence. We leverage a graph neural partial differential equation (PDE) based on Beltrami flow to obtain high-dimensional features and position embeddings for point clouds. We incorporate position embeddings into a Transformer module based on a neural ordinary differential equation (ODE) to efficiently represent patches within points. We employ the multi-level correspondence derived from the high feature similarity scores to facilitate alignment between point clouds. Subsequently, we use registration methods such as SVD-based algorithms to predict the transformation using corresponding point pairs. We evaluate PosDiffNet on several 3D point cloud datasets, verifying that it achieves state-of-the-art (SOTA) performance for point cloud registration in large fields of view with perturbations. The implementation code of experiments is available at https://github.com/AI-IT-AVs/PosDiffNet.

show abstract

End-to-End 3D Point Cloud Learning for Registration Task Using Virtual Correspondences

Cited by 16 publications

References 15 publications

Fast and Robust Registration of Partially Overlapping Point Clouds

Fast and Robust Registration of Partially Overlapping Point Clouds

Map-less long-term localization in complex industrial environments

PosDiffNet: Positional Neural Diffusion for Point Cloud Registration in a Large Field of View with Perturbations

Contact Info

Product

Resources

About