How do the authors prevent the linear growth in the number of points?

To prevent the linear growth in the number of points, the authors decimate the elements of the local map by voxelizing them at a resolution of 2.5 cm.

What are the datasets that are used for benchmarking?

The datasets cover 3 environments of increasing complexity (low, high, medium), with 3 types of motions (rotational, translational, fly) at 3 different speeds (slow, medium, fast).

How does the algorithm minimize Eq. 13?

Their algorithm minimizes Eq. 13 by using a local parametrization of the perturbation in the following form: ∆T = ( ∆tx ∆ty ∆tz ∆qx ∆qy ∆qz )T .

What is the simplest way to compute the transformation between the two frames?

Once the authors have a set of candidate correspondences C = 〈i, j〉1:M , the authors compute the transformation between the two frames by using an iterative least squares formulation.

How do the authors compute the point cloud?

Given the raw sensor data (3D scan or depth image), the authors compute the point cloud by applying an unprojectionfunction, as discussed in Section IV-A.

What is the operator on points with normals?

The ⊕ operator on points with normals is:p̃′ = T⊕ p̃ = (Rp + t Rn) (11)Given a correspondence and a current transform T , the error eij(T) is a 6D vectoreij (T) = ( p̃ci −T⊕ p̃rj ) .

What is the general formulation of ICP?

In its most general formulation, ICP iteratively refines an initial transform T by alternating the search for correspondences and the solution of Eq. 1.

How do the authors overcome the issue of the correspondence between points?

The authors overcome this issue by reformulating the correspondence between points as a partial overlap between small surfaces centered in the points, and introducing an error metric that takes into account this aspect.

What is the way to compare NICP to other state-of-the-art?

The authors performed statistical experiments comparing their approach to other state-of-the-art methods showing that their algorithm offers better results and higher robustness to poor initial guesses.

Why did it take so long to process each new point cloud?

KINFU tookabout 50 ms to process each new point cloud but this low frame rate is probably mainly due to the low end video card in their system.

What is the difference in time between the two poses?

Note that, since these datasets are recorded with a high frame rate, the RPE is computed on poses with a difference in time of 1 second.

Open AccessProceedings Article10.1109/IROS.2015.7353455

NICP: Dense normal based point cloud registration

Jacopo Serafin, +1 more

- 17 Dec 2015

- pp 742-749

329

TL;DR: A novel on-line method to recursively align point clouds by considering each point together with the local features of the surface (normal and curvature) that minimizes an error metric depending on these surface characteristics.

Abstract: In this paper we present a novel on-line method to recursively align point clouds. By considering each point together with the local features of the surface (normal and curvature), our method takes advantage of the 3D structure around the points for the determination of the data association between two clouds. The algorithm relies on a least squares formulation of the alignment problem, that minimizes an error metric depending on these surface characteristics. We named the approach Normal Iterative Closest Point (NICP in short). Extensive experiments on publicly available benchmark data show that NICP outperforms other state-of-the-art approaches.

AI Agents for this Paper

Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps

Most frequently asked questions

1. What are the contributions mentioned in the paper "Nicp: dense normal based point cloud registration" ?

In this paper the authors present a novel on-line method to recursively align point clouds.

2. What have the authors stated for future works in "Nicp: dense normal based point cloud registration" ?

In this paper the authors presented a novel system to register point clouds that takes into account an extended measurement vector.. In the future the authors plan to provide a GPU implementation of their system.

3. What is the problem of registering two point clouds?

ICPThe problem of registering two point clouds consists in finding the rotation and the translation that maximize the overlap between the two clouds.

4. What is the metric for the evaluation of visual odometry?

The RPE measures the pairwise alignment error between successive poses and it is one of the best metric for the evaluation of visual odometry or camera tracking systems.

Fig. 1. Side views of the reconstruction of the ETH Hauptgebaude dataset. From top to bottom: ground-truth, NICP, GICP and NDT reconstruction. NICP was able to reconstruct consistently the scene while both NDT and GICP generated a shortened corridor (count the arcs).

TABLE II AVERAGE CLOUD REGISTRATION TIME FOR EACH ALGORITHM.

TABLE I MEAN OF RELATIVE TRANSLATIONAL AND ROTATIONAL ERROR FOR THE DEPTH CAMERA DATASETS. GREEN CELLS IN THE TABLE HIGHLIGHT THE BEST RESULT FOR EACH SPECIFIC DATASET AMONG ALL THE COMPARED ALGORITHMS (GICP, DVO, NDT AND NICP).

Fig. 4. Mean relative translational (left) and rotational (right) error obtained with KINFU and NICP+ on the depth camera datasets.

Fig. 5. Side views of the reconstruction of the ETH Stairs dataset. Top-left: ground-truth. Top-right: NICP reconstruction. Bottom-left: GICP reconstruction. Bottom-right: NDT reconstruction. Both NICP and NDT consistenly reconstructed the scene, GICP failed in aligning the part on the back of the stairs.

Fig. 6. Evolution of the mean of the RPE, and its standard deviation, when incrementally increasing the weight of the surface normals in the algorithm. The first and second plot illustrate respectively the translational and the rotational part of the error for the ETH Hauptgebaude dataset. The third and fourth plot illustrate respectively the translational and the rotational part of the error for the ETH Stairs dataset.

Citations

Proceedings Article•10.1109/ICRA48506.2021.9560835

Voxelized GICP for Fast and Accurate 3D Point Cloud Registration

Kenji Koide, +3 more

- 30 May 2021

TL;DR: The voxelized generalized iterative closest point (VGICP) algorithm for fast and accurate three-dimensional point cloud registration is presented, and it is confirmed that the accuracy of the proposed algorithm is comparable to GICP, but is substantially faster than existing methods.

...read moreread less

320

•Proceedings Article•10.1109/CVPR.2019.00867

LO-Net: Deep Real-Time Lidar Odometry

Qing Li, +6 more

- 15 Jun 2019

TL;DR: Li et al. as discussed by the authors proposed a novel deep convolutional network pipeline, LO-Net, for real-time lidar odometry estimation, which can effectively learn feature representation for LO estimation, and implicitly exploit the sequential dependencies and dynamics in the data.

...read moreread less

266

Journal Article•10.1109/TRO.2016.2609395

Effective Background Model-Based RGB-D Dense Visual Odometry in a Dynamic Environment

Deok-Hwa Kim, +1 more

- 01 Dec 2016

- IEEE Transactions on Robotics

TL;DR: Experimental results demonstrate that the ego-motion is robustly obtained by BaMVO in a dynamic environment.

...read moreread less

201

•Journal Article•10.1109/ACCESS.2020.2986470

Evaluation of the ICP Algorithm in 3D Point Cloud Registration

Peng Li, +3 more

- 08 Apr 2020

- IEEE Access

TL;DR: This paper takes the overlap ratio, angle, distance, and noise as the influencing factors of ICP and evaluates the validity, robustness, accuracy, and efficiency of point- to-point and point-to-plane ICP by using four datasets.

...read moreread less

148

•Proceedings Article•10.1109/CVPR.2017.595

Convex Global 3D Registration with Lagrangian Duality

Jesus Briales, +1 more

- 01 Jul 2017

TL;DR: This paper addresses finding the globally optimal transformation in various 3D registration problems by a unified formulation that integrates common geometric registration modalities by introducing a strengthened Lagrangian dual relaxation, which surpasses previous similar approaches in effectiveness.

...read moreread less

136

...

Expand

References

Journal Article•10.1109/34.121791

A method for registration of 3-D shapes

Paul J. Besl, +1 more

- 01 Feb 1992

- IEEE Transactions on Pattern Analysis an...

TL;DR: In this paper, the authors describe a general-purpose representation-independent method for the accurate and computationally efficient registration of 3D shapes including free-form curves and surfaces, based on the iterative closest point (ICP) algorithm, which requires only a procedure to find the closest point on a geometric entity to a given point.

...read moreread less

20.6K

•Proceedings Article

ROS: an open-source Robot Operating System

Morgan Quigley

- 01 Jan 2009

TL;DR: This paper discusses how ROS relates to existing robot software frameworks, and briefly overview some of the available application software which uses ROS.

...read moreread less

10.2K

•Proceedings Article•10.1109/ICRA.2011.5980567

3D is here: Point Cloud Library (PCL)

Radu Bogdan Rusu, +1 more

- 09 May 2011

TL;DR: PCL (Point Cloud Library) is presented, an advanced and extensive approach to the subject of 3D perception that contains state-of-the art algorithms for: filtering, feature estimation, surface reconstruction, registration, model fitting and segmentation.

...read moreread less

5.5K

•Proceedings Article•10.1109/ISMAR.2011.6092378

KinectFusion: Real-time dense surface mapping and tracking

Richard Newcombe, +9 more

- 26 Oct 2011

TL;DR: A system for accurate real-time mapping of complex and arbitrary indoor scenes in variable lighting conditions, using only a moving low-cost depth camera and commodity graphics hardware, which fuse all of the depth data streamed from a Kinect sensor into a single global implicit surface model of the observed scene in real- time.

...read moreread less

4.9K

•Proceedings Article•10.1109/IROS.2012.6385773

A benchmark for the evaluation of RGB-D SLAM systems

Jrgen Sturm, +4 more

- 24 Dec 2012

TL;DR: A large set of image sequences from a Microsoft Kinect with highly accurate and time-synchronized ground truth camera poses from a motion capture system is recorded for the evaluation of RGB-D SLAM systems.

...read moreread less

4.3K

...

Expand

NICP: Dense normal based point cloud registration

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What are the contributions mentioned in the paper "Nicp: dense normal based point cloud registration" ?

2. What have the authors stated for future works in "Nicp: dense normal based point cloud registration" ?

3. What is the problem of registering two point clouds?

4. What is the metric for the evaluation of visual odometry?

5. How do the authors prevent the linear growth in the number of points?

6. What are the datasets that are used for benchmarking?

7. How does the algorithm minimize Eq. 13?

8. What is the simplest way to compute the transformation between the two frames?

9. How do the authors compute the point cloud?

10. What is the operator on points with normals?

11. What is the general formulation of ICP?

12. How do the authors overcome the issue of the correspondence between points?

13. What is the way to compare NICP to other state-of-the-art?

14. Why did it take so long to process each new point cloud?

15. What is the difference in time between the two poses?

Figures

Citations

Voxelized GICP for Fast and Accurate 3D Point Cloud Registration

LO-Net: Deep Real-Time Lidar Odometry

Effective Background Model-Based RGB-D Dense Visual Odometry in a Dynamic Environment

Evaluation of the ICP Algorithm in 3D Point Cloud Registration

Convex Global 3D Registration with Lagrangian Duality

References

A method for registration of 3-D shapes

ROS: an open-source Robot Operating System

3D is here: Point Cloud Library (PCL)

KinectFusion: Real-time dense surface mapping and tracking

A benchmark for the evaluation of RGB-D SLAM systems

Related Papers (5)

A method for registration of 3-D shapes

Efficient variants of the ICP algorithm

Fast Point Feature Histograms (FPFH) for 3D registration

Are we ready for autonomous driving? The KITTI vision benchmark suite

LOAM: Lidar Odometry and Mapping in Real-time