DeepCLR: Correspondence-Less Architecture for Deep End-to-End Point Cloud Registration

@article{Horn2020DeepCLRCA,
  title={DeepCLR: Correspondence-Less Architecture for Deep End-to-End Point Cloud Registration},
  author={Markus Horn and Nico Engel and Vasileios Belagiannis and Michael Buchholz and Klaus C. J. Dietmayer},
  journal={2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC)},
  year={2020},
  pages={1-7}
}
  • Markus HornNico Engel K. Dietmayer
  • Published 22 July 2020
  • Computer Science, Environmental Science
  • 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC)
This work addresses the problem of point cloud registration using deep neural networks. We propose an approach to predict the alignment between two point clouds with overlapping data content, but displaced origins. Such point clouds originate, for example, from consecutive measurements of a LiDAR mounted on a moving platform. The main difficulty in deep registration of raw point clouds is the fusion of template and source point cloud. Our proposed architecture applies flow embedding to tackle… 

Figures and Tables from this paper

D3dlo: Deep 3d Lidar Odometry

A network architecture that learns LO by directly processing 3D point clouds is proposed that is trained on the KITTI dataset in an end-to-end manner without the necessity of pre-defining corresponding pairs of points.

LiDAR Odometry Methodologies for Autonomous Driving: A Survey

This survey examines the existing LiDAR odometry methods and summarize the pipeline and delineate the several intermediate steps, and compares the accuracy and the running speed among these methodologies evaluated over the KITTI odometry dataset and outline promising future research directions.

Zero NeRF: Registration with Zero Overlap

Zero-NeRF is presented, a projective surface reg- istration method that offers the first general solution capable of alignment between scene representations with minimal or zero visual correspondence, and enforces consistency between visible surfaces of partial and complete reconstructions.

VPRNet: Virtual Points Registration Network for Partial-to-Partial Point Cloud Registration

A self-supervised virtual point generation network (VPGnet), which utilizes the attention mechanism of Transformer and Self-Attention to fuse the geometric information of two partial point clouds, combined with the Generative Adversarial Network (GAN) structure to produce missing points.

Indirect Point Cloud Registration: Aligning Distance Fields Using a Pseudo Third Point Set

Inspired by the high precision of deep learning global feature registration, the algorithm is generalized to a non-Deep Learning setting while retaining the accuracy and achieves a competitive performance and faster speed, compared to Deep Learning-based registration models.

IPCR: an End-to-end Iterative Point Cloud Registration Network

An end-to-end point cloud registration network IPCR is proposed, which extracts the global and local features by PointNet, sends them into the localization estimation network, obtains the transformation matrix through SVD, and finally improves the registration accuracy by iterative.

PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization

A novel end-to-end learning-based LiDAR sensor relocalization framework, termed PointLoc, which infers 6-DoF poses directly using only a single point cloud as input, which can achieve accurate relocalized performance on recently released challenging Oxford Radar RobotCar dataset and real-world robot experiments.

Graph Attention based Point Cloud Registration

This paper uses a mini-ConvNet to extract point-wise features for sampled points and their neighbors, and graph attention is applied to aggregate the local information from neighbor points to center points and a kind of cross attention method for point cloud registration.

ENCODE: a dEep poiNt Cloud ODometry nEtwork

This paper replaces hand-crafted features with a neural network and directly estimate the relative pose between two adjacent scans from a LiDAR sensor using ENCODE: a dEep poiNt Cloud ODometry nEtwork to provide accurate poses with low drift in various kinds of scenarios.

References

SHOWING 1-10 OF 23 REFERENCES

DeepPCO: End-to-End Point Cloud Odometry through Deep Parallel Neural Network

This work investigates how to exploit deep learning to estimate point cloud odometry (PCO), which may serve as a critical component in point cloud-based downstream tasks or learning-based systems, and proposes a novel end-to-end deep parallel neural network called DeepPCO, which can estimate the 6-DOF poses using consecutive point clouds.

DeepVCP: An End-to-End Deep Neural Network for Point Cloud Registration

DeepVCP is presented - a novel end-to-end learning-based 3D point cloud registration framework that achieves comparable registration accuracy to prior state-of-the-art geometric methods and the low registration error and high robustness of the method make it attractive to the substantial applications relying on the point cloudRegistration task.

PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet

It is argued that PointNet itself can be thought of as a learnable "imaging" function, and classical vision algorithms for image alignment can be brought to bear on the problem -- namely the Lucas & Kanade (LK) algorithm.

3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration

The 3DFeat-Net is proposed which learns both 3D feature detector and descriptor for point cloud matching using weak supervision and obtains state-of-the-art performance on these gravity-aligned datasets.

FlowNet3D: Learning Scene Flow in 3D Point Clouds

  • Xingyu LiuC. QiL. Guibas
  • Computer Science
    2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
  • 2019
This work proposes a novel deep neural network named FlowNet3D that learns scene flow from point clouds in an end-to-end fashion and successfully generalizes to real scans, outperforming various baselines and showing competitive results to the prior art.

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

This paper designs a novel type of neural network that directly consumes point clouds, which well respects the permutation invariance of points in the input and provides a unified architecture for applications ranging from object classification, part segmentation, to scene semantic parsing.

LO-Net: Deep Real-Time Lidar Odometry

A novel deep convolutional network pipeline, LO-Net, for real-time lidar odometry estimation that outperforms existing learning based approaches and has similar accuracy with the state-of-the-art geometry-based approach, LOAM.

DeepLocalization: Landmark-based Self-Localization with Deep Neural Networks

This work proposes DeepLocalization, a deep neural network that regresses the vehicle’s translation and rotation parameters from unordered and dynamic input landmarks and shows that it can be combined with regular GPS signals and filtering algorithms such as the extended Kalman filter.

Fast Point Feature Histograms (FPFH) for 3D registration

This paper modifications their mathematical expressions and performs a rigorous analysis on their robustness and complexity for the problem of 3D registration for overlapping point cloud views, and proposes an algorithm for the online computation of FPFH features for realtime applications.

RegNet: Multimodal sensor registration using deep neural networks

RegNet is presented, the first deep convolutional neural network to infer a 6 degrees of freedom (DOF) extrinsic calibration between multimodal sensors, exemplified using a scanning LiDAR and a monocular camera.