Robust 3D Point Cloud Registration Exploiting Unique LiDAR Scanning

Pattern

Ahmad Kamal Aijazi

and Paul Checchin

Institut Pascal, UMR 6602 CNRS Universit

e Clermont Auvergne, F-63000 Clermont-Ferrand, France

Keywords:

LiDAR, Non-Repetitive Scanning, 3D Point Cloud and Scan Registration.

Abstract:

The task of 3D point cloud registration is fundamentally about aligning multiple scans or point clouds ob-

tained from one or more LiDAR sensors to create a uniﬁed and accurate representation of the scanned scene.

This process serves as the cornerstone for applications such as map building, autonomous navigation, land

surveying and many others. While 3D registration techniques have made signiﬁcant advancements, several

persistent challenges continue to warrant research attention and innovation. Recently, non-repetitive scanning

LiDAR sensors are emerging as a promising alternative for 3D data acquisition. In this paper, a novel 3D

point cloud registration method is presented that exploits the unique scanning pattern of the sensor to register

successive 3D scans. The sensor is ﬁrst characterized and then, using the characteristic equation of the unique

scanning pattern, a perfect scan is reconstructed at the target distance. The real scan is then compared with this

reconstructed scan to extract objects in the scene. The displacements of these extracted objects in successive

scans, with respect to the center of the unique scanning pattern, are compared in successive scans to determine

the transformations that are then used to register the successive scans. The proposed method is evaluated on

two real and different datasets and compared with other state-of-the-art registration methods. The results show

that the method is comparable with other methods in terms of accuracy but surpasses them in performance in

terms of processing time.

1 INTRODUCTION

The advancement of LiDAR (Light Detection and

Ranging) technology has played a pivotal role in rev-

olutionizing our ability to capture three-dimensional

representations of the environment with unprece-

dented accuracies and details. The task of 3D point

cloud registration is fundamentally about aligning

multiple scans or point clouds obtained from one

or more of these LiDAR sensors to create a uni-

ﬁed and accurate representation of the scanned scene.

However, the process of registering multiple LiDAR-

generated point clouds into a single, coherent 3D

model is not without its challenges.

Most of the state-of-the-art methods employ repet-

itive scanning LiDAR sensors which have their inher-

ent limitations. Recently, non-repetitive scanning Li-

DAR sensors are emerging as a promising alternative

for 3D data acquisition. In this paper, we explore the

use of such a LiDAR sensor to improve the 3D point

https://orcid.org/0009-0000-6349-128X

https://orcid.org/0000-0002-2930-3393

cloud registration process using its unique scanning

pattern.

2 RELATED WORK

Point cloud registration is an important task for lo-

calization and mapping applications. In this section,

we review some key research in the ﬁeld of 3D point

cloud registration.

The Iterative Closest Point algorithm (ICP) pro-

posed by Besl and McKay (1992) has been an ele-

mentary method in point cloud registration. In this

method, the distance between corresponding points

in two point clouds is iteratively minimized, allowing

for rigid transformation estimation. Extensions, such

as the Generalized ICP (GICP) by Segal et al. (2009),

dense normal based ICP (NICP) (Seraﬁn and Grisetti,

2015) and Sample Consensus ICP (Xie et al., 2022)

have improved robustness to various scenarios.

Tazir et al. (2018) modiﬁed the method for select-

ing 3D points for the matching process in the standard

ICP algorithm to propose Cluster Iterative Closest

Aijazi, A. and Checchin, P.

Robust 3D Point Cloud Registration Exploiting Unique LiDAR Scanning Pattern.

DOI: 10.5220/0012342600003654

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 13th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2024), pages 679-686

ISBN: 978-989-758-684-2; ISSN: 2184-4313

679

Point (CICP) that successfully register point clouds

of different densities obtained form different sensors.

Another variant Global ICP (G-ICP) presented by Ai-

jazi et al. (2019) employs bundle adjustment to ﬁne

tune the registration process, at the end, in addition

to the scan by scan registration using point to plane

matching. However, such method is not suitable for

real-time processing.

Feature-based registration methods focus on ex-

tracting distinctive features from point clouds and

matching them to establish correspondences. The

SIFT-based approach by Rusu et al. (2009) and the

Fast Point Feature Histograms (FPFH) by Rusu and

Cousins (2011) are notable examples.

One of the important concerns for point cloud

registration is scalability. The LeGO-LOAM system

by Xue et al. (2022) offers a real-time solution for

large-scale point cloud registration and mapping. Re-

cent advances in deep learning have led to the emer-

gence of deep neural networks for point cloud regis-

tration (Pais et al., 2019; Choy et al., 2020).

Some works have also employed recently devel-

oped non-repetitive scanning LiDARs. The work has

also been done to register their point clouds. An ex-

tension of classical feature-based methods, the work

by Wang et al. (2021) introduced a feature-based reg-

istration algorithm employing point clouds generated

by non-repetitive scanning LiDARs leveraging dis-

tinct features to establish correspondences. Wang

et al. (2023) proposed a novel method that leverages

deep learning techniques for sparse point cloud reg-

istration, achieving impressive results in various sce-

narios. He et al. (2021) introduced a highly efﬁcient

algorithm designed for non-repetitive scanning Li-

DARs, combining geometric and learning-based ap-

proaches to improve speed and accuracy. Li et al.

(2023) presented a registration method for combin-

ing non-repetitive scanning LiDAR point clouds with

RGB-D data, enabling precise 3D scene reconstruc-

tion. Zou et al. (2023) presented a plane-based global

registration (PGR) approach for coarse registration

followed by an ICP algorithm to register point clouds.

In this paper, we present a fast 3D scan registra-

tion method exploiting the unique scanning pattern of

a LiDAR sensor. To the best of our knowledge, no

work has so far exploited the non-repetitive scanning

pattern to register point clouds.

In the following sections, we ﬁrst explain the Li-

DAR sensor and its characterization (Section 3) and

then our method (Section 4). The experiments and re-

sults are presented in Section 5 while we conclude in

Section 6.

3 CHARACTERIZATION OF

NON-REPETITIVE SCANNING

LIDAR

The recently developed Mid-70 sensor by LIVOX is

a solid state 3D LiDAR. The sensor is capable of

collecting up to 100 000 points per second within a

circular ﬁeld of view directly in front of the scan-

ner’s glass window. With a small form factor, it

has a range of about 260m with precision of up to

0.02m (see Table 1 for more speciﬁcations). How-

ever, what really sets this sensor apart from other laser

scanners is its scanning mechanism. Rather than ro-

tate the laser scanner/receiver pair themselves (as is

done in a multi-line scanner like Velodyne LiDARs),

this holds the two in a “solid state”, immobilized in-

side the scanner’s housing while the outbound and

inbound light is directed through a pair of Risley

prisms (Brazeal et al., 2021). These wedge-shaped

glass lenses act as optical steering mechanisms as they

rotate in opposite directions, directing the beam in

a rosette-style scan pattern in the circular scan area

(Figure 1).

Table 1: Main speciﬁcations of the used sensor.

Main speciﬁcations

Laser Wavelength 905nm

Detection Range 260m @ 80% reﬂectivity

Field of view 70

◦

circular

Range Precision 0.02m

Beam divergence 0.28

◦

(vert.) × 0.03

◦

(horiz.)

Dimension 97 × 64 × 62.7 mm

A rosette is a set of points in polar coordinates

(r, θ) speciﬁed by the polar equation:

r = a cos(kθ) (1)

or in Cartesian coordinates (x, y) using the parametric

equations:

x = r cos(θ) = a cos(kθ) cos(θ) (2)

y = r sin(θ) = a cos(kθ)sin(θ) (3)

where a is the amplitude or size factor, k the angular

frequency. The rose speciﬁed by r = a sin(kθ) is iden-

tical to that speciﬁed by r = a cos(kθ) rotated counter-

clockwise by

radians, which is one-quarter the pe-

riod of either sinusoid. If k is odd, the rosette is k

petaled. If k is even, the rosette is 2k petaled.

In order to exploit the scanning pattern of the Li-

DAR sensor, we need to characterize its scanning pat-

tern by determining the value of a and k. To determine

these parameters, we positioned the sensor on a tripod

stand in front of a plane surface as shown in Figure 2.

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

680

(a) 100 ms (b) 200 ms (c) 400 ms

Figure 1: Rosette – style scanning pattern of the sensor scans at different integration time.

Figure 2: Experimental setup.

In order to determine the value of k, the differ-

ent scans obtained from the sensor were analyzed (the

form and number of petals) at different integration

times, as shown in Figure 1. The number of petals

found was 200 per second. Being an even number,

this equates to a value of k = 100.

We considered the value of a as a function of scan-

ning distance. The distance of sensor was varied from

1 to 50m at regular intervals and the successive scan

patterns obtained were studied. Figure 3 presents the

size (in terms of diameter) of the rosette scan in terms

of distance.

Using a linear approximation, the value of a was

found to be:

a = 1.4 × d (4)

where d is the target distance.

4 3D SCAN REGISTRATION

The scans are registered one by one in succession.

In order to register two successive scans s

and s

where i and j are the number of scans, we ﬁrst recon-

struct each scan (Rosette-style) at the speciﬁed target

Figure 3: The size (in terms of diameter) of the rosette scan

with respect to the target distance.

distance d applying the equations (2) and (3) using

the values of k and a determined in the previous sec-

tion. The distance d in (4) is taken as the distance of

the farthest vertical plane, perpendicular to the sensor

plane, detected in each scan.

The centres c

and c

of the successive recon-

structed Rosette-style scans s

and s

are ﬁrst aligned

and then in the second step their orientations are

aligned and rotations around the three axes, i.e. θ

and θ

respectively, are estimated as shown in Fig-

ure 4.

In order to align the orientation, the surface

normal vector is estimated for each of the recon-

structed scan using Principal Component Analysis

(PCA) (Klasing et al., 2009). Given a set of points in

each reconstructed scan D = x

(i=1)

, the PCA surface

normal approximation for a given data point p ∈ D is

typically computed by ﬁrst determining the K-Nearest

Neighbors, x

∈ D, of p. Given the K neighbors, the

approximate surface normal is then the eigenvector

associated with the smallest eigenvalue of the sym-

metric positive semi-deﬁnite matrix:

V =

∑

k=1

− p)

− p) (5)

where p corresponds to the centers c

and c

of the

Robust 3D Point Cloud Registration Exploiting Unique LiDAR Scanning Pattern

681

(a) (b)

Figure 4: Figure (a) shows the two original successive scans with objects. Figure (b) presents the reconstructed scans cor-

responding to the two original scans without objects. In (c), the centers of the two scans are aligned, and normal vectors

estimated while in (d) the second scan is rotated to align with the ﬁrst scan and the difference of normal vectors give θ

, θ

successive reconstructed Rosette-style scans s

and

The rotation around the three axis is then esti-

mated as the difference between the orientation of the

two normal vectors. As the sensor remains on the

ground, the rotation in successive scans around the

axe perpendicular to the sensor plane is considered

zero.

= (θ

− θ

)

= (θ

− θ

) (6)

= (θ

− θ

) ≈ 0

where θ

, θ

and θ

, θ

are the orienta-

tions of the two vectors with respect to x, y and z axis,

respectively.

The original scans s

and s

are then matched

with the corresponding reconstructed scans s

and s

As the reconstructed scans are free of objects, this al-

lows the extraction of objects present in the original

scans leaving behind holes. The reconstructed scans

help complete the farthest and largest vertical plane

in the original scans occluded by objects in the scene.

All 3D points lying outside this vertical plane are then

considered as object points and are extracted out leav-

ing behind holes. The centroid P of the resulting holes

in the original scans are calculated using a bounding

box around each hole (or extracted object) with re-

spect to c

and c

as shown in Figure 5. The dis-

placement of these centroids with respect to the centre

of the scans in the successive scans provides us with

the translation, i.e. t

x,y

, between the two successive

scans:

x,y

∑

m=1



− c

) − (P

− c

)



(7)

where n is the total number of holes (due to ex-

tracted objects) in the scans, P

and P

are the cen-

troids of the m

hole in the scans s

and s

, respec-

tively. The displacement along the z axis is denoted

= (d

− d

), where d

and d

are the distances of

the farthest vertical plane, perpendicular to the sensor

plane, detected in each scan, respectively.

These rotations R(θ

, θ

) and translations

T(t

, t

) are then used to transform the original

scans, with the ﬁrst scan taken as reference, using (8).

The transformed scan and the reference scan are then

aggregated (9) to form the new reference scan for the

next scan.

ˆs

= R.s

+ T (8)

i+1

= ˆs

+ s

(9)

The method is summarized in Algorithm 1.

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

682

(a) (b)

(e) (f)

Figure 5: Figures (a) and (b) show the two original successive scans with objects. (c) and (d) present the two scans with

holes after extraction of objects. (e) and (f) show the bounding box around the different holes in both scans, along with their

centroids and the vectors from the center of the scans and the centroids. The difference in these vectors (displacement of

objects with respect to the scan center) provides us the estimates of t

and t

5 EXPERIMENTS AND RESULTS

In order to evaluate the proposed method, the sen-

sor was employed in two different modes. In the ﬁrst

mode, it was mounted on a tripod and moved around

different positions to completely scan an indoor room

equipped with furniture (Experiment 1) while in the

second mode, it was mounted on a small mobile

robot moving on a predeﬁned trajectory, at a speed

of about 1 ms

−1

, scanning an enclosed hangar/parking

Robust 3D Point Cloud Registration Exploiting Unique LiDAR Scanning Pattern

683

Data: 3D scans s

and s

i+1

Result: s

i+1

= ˆs

i+1

+ s

for i ← 1 to total number of scans do

Reconstruct Rosette style scan s

and

i+1

at distance of farthest vertical plane

detected;

Align centers of s

and s

i+1

;

Estimate the surface normal vector V for

and s

i+1

;

Estimate the rotations θ

, θ

difference of orientation between the

two normal vectors;

Compare s

and s

i+1

with s

and s

i+1

respectively, to extract m number of

objects;

Compute bounding box around the m

holes in each scan and their centroids

and P

i+1

;

Compute vectors from centroids P

and

i+1

to the centers of s

and s

i+1

, i.e.

and c

i+1

, respectively;

Compare the vectors in the successive

scans to determine t

and t

;

Calculate t

= d

− d

i+1

;

Transform s

i+1

with respect to s

ˆs

i+1

= R.s

i+1

+ T;

end

Algorithm 1: 3D Scan Registration.

(Experiment 2), as shown in Figure 6. Scanning

was done at an integration time of 200 ms. Experi-

ment 1 contained 35 scans while Experiment 2 con-

tains 357 scans. In both cases, loop closure was

ensured to help in the evaluation process while the

ground truth was obtained via physical measurements

of the different room dimensions.

(a) (b)

Figure 6: (a) and (b) show the experimental setup for Ex-

periment 1 and Experiment 2.

Using these data, the proposed method is com-

pared with a state-of-the-art Global Iterative Closest

Point (G-ICP) (Aijazi et al., 2019) and Fast Point Fea-

ture Histograms FPFH methods (Rusu and Cousins,

2011).

5.1 Registration Accuracy

The registration results for both experiments are pre-

sented in Figure 7. A quantitative analysis is pre-

sented in Table 2 for the two experiments and com-

pared with the state-of-the-art methods. ∆E presents

the registration error calculated in terms of average

absolute difference in dimensions (width and height)

of all the walls as compared to the ground truth (sim-

ilar to Aijazi et al. (2013)).

Table 2: Registration Accuracy for Experiment 1 and Ex-

periment 2.

Experiment 1 Experiment 2

∆E (m) ∆E (m)

Proposed

method

0.018 0.043

G-ICP

Aijazi et al. (2019)

0.012 0.024

FPFH

Rusu and Cousins (2011)

0.017 0.044

The results show that the proposed method is com-

parable to the other state-of-the-art methods in terms

of accuracy. The registration accuracy of the proposed

method was better in Experiment 1 and closer to G-

ICP as the room size was small and there were less

number of scans. In case of Experiment 2, G-ICP out-

performs the other methods. This is because, in the

G-ICP algorithm, there is a bundle adjustment step at

the end to improve the overall scan registration. So

the number of scans (or size of acquisition) has less

impact on the overall results, whereas in the proposed

method, the registration error usually tends to accu-

mulate more with larger number of scans (acquisition

size) as seen in the results.

5.2 Processing Time

The proposed method along with the other state-of-

the-art methods were run on the same laptop (proces-

sor Intel Core i9-10885H, CPU @ 2.40 GHz, mem-

ory 32 GB) under Windows 10 operating system. The

results of the registration time (in seconds) are pre-

sented in Table 3.

From the results, it can be seen that there is not

a signiﬁcant difference in the processing time, espe-

cially in case of Experiment 1, at this lower scan res-

olution (number of points per scan).

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

684

(a) (b)

(c)

Figure 7: (a), (b) and (c) show the registration results for Experiment 1 and Experiment 2 employing the proposed method.

Table 3: Processing time.

Experiment 1 (s) Experiment 2 (s)

Proposed

method

90 900

G-ICP

Aijazi et al. (2019)

100 1700

FPFH

Rusu and Cousins (2011)

95 1295

As the results presented in Table 3 were inconclu-

sive, the Experiment 1 was repeated at different scan

resolutions by varying the data integration time, i.e.

from 200ms (low resolution) to 500 ms (high resolu-

tion). The results are presented in Table 4.

Table 4: Processing time for different scan resolutions.

200ms 300ms 400ms 500ms

Proposed

method

90s 155s 205s 289s

G-ICP

Aijazi et al. (2019)

100s 250s 480s 875s

FPFH

Rusu and Cousins (2011)

95s 197s 352s 615s

The results demonstrate the strength of the pro-

posed method in terms of computational time as the

number of points per scan increases. For methods like

G-ICP, the processing time increases manifold with

the increase of number of points as the number of cor-

respondences between points increases. However, the

proposed method remains relatively unchanged. This

makes it more suitable for real time applications.

6 CONCLUSION

In this paper, a novel 3D registration method is pre-

sented that exploits the unique scanning pattern of the

sensor to register successive 3D scans. The method

was evaluated on real data and compared with other

state-of-the-art methods that show the prowess of the

proposed method. The results show that the method is

comparable with other methods in terms of accuracy

but surpasses them in performance in terms of pro-

cessing speed. In the proposed method, slight drift in

error is observed in case of larger distance and num-

ber of scans. However, in terms of processing time,

it is shown to be invariant to increase in number of

3D points per scan. For future, we would like to im-

prove the accuracy of the method by incorporating

some constraints like loop closure, whereas it would

also be considered to use an ICP method to further

improve or ﬁne register the resulting point cloud.

Robust 3D Point Cloud Registration Exploiting Unique LiDAR Scanning Pattern

685

ACKNOWLEDGEMENTS

This work is supported by the International Research

Center “Innovation Transportation and Production

Systems” of the I-SITE CAP 20−25. Financial sup-

port was also received from the Auvergne-Rh

one-

Alpes region through the ACCROBOT project (AC-

Costage haute pr

ecision ROBOTis

e − Chantier Tran-

sitique du laboratoire FACTOLAB) as part of the Pack

Ambition Recherche 2020.

REFERENCES

Aijazi, A. K., Checchin, P., and Trassoudaine, L. (2013).

Automatic removal of imperfections and change de-

tection for accurate 3d urban cartography by classi-

ﬁcation and incremental updating. Remote Sensing,

5(8):3701–3728.

Aijazi, A. K., Malaterre, L., Trassoudaine, L., Chateau, T.,

and Checchin, P. (2019). Automatic Detection and

Modeling of Underground Pipes Using a Portable 3D

LiDAR System. Sensors, 19(24).

Besl, P. and McKay, N. (1992). A Method for Registration

of 3-D Shapes. IEEE Trans. on PAMI, 14(2):239–256.

Brazeal, R. G., Wilkinson, B. E., and Hochmair, H. H.

(2021). A Rigorous Observation Model for the Ris-

ley Prism-Based Livox Mid-40 Lidar Sensor. Sensors,

21(14).

Choy, C., Dong, W., and Koltun, V. (2020). Deep Global

Registration. In Proceedings of the IEEE Computer

Society Conference on Computer Vision and Pattern

Recognition, pages 2511–2520.

He, Y., Ma, L., Jiang, Z., Tang, Y., and Xing, G. (2021).

VI-Eye: Semantic-Based 3D Point Cloud Registra-

tion for Infrastructure-Assisted Autonomous Driving.

In Proceedings of the 27th Annual International Con-

ference on Mobile Computing and Networking, Mobi-

Com ’21, page 573–586, New York, NY, USA.

Klasing, K., Althoff, D., Wollherr, D., and Buss, M. (2009).

Comparison of surface normal estimation methods for

range sensing applications. In 2009 IEEE Interna-

tional Conference on Robotics and Automation, pages

3206–3211.

Li, H., Zhao, H., Ye, B., and Zhang, Y. (2023). 3D semantic

map construction based on point cloud and image fu-

sion. IET Cyber-Systems and Robotics, 5(1):e12078.

Pais, G. D., Ramalingam, S., Govindu, V. M., Nascimento,

J. C., Chellappa, R., and Miraldo, P. (2019). 3DReg-

Net: A Deep Neural Network for 3D Point Registra-

tion. In Proceedings of the IEEE Computer Society

Conference on Computer Vision and Pattern Recogni-

tion, pages 7191–7201.

Rusu, R. B., Blodow, N., and Beetz, M. (2009). Fast Point

Feature Histograms (FPFH) for 3D registration. In

IEEE International Conference on Robotics and Au-

tomation, pages 3212–3217.

Rusu, R. B. and Cousins, S. (2011). 3D is here: Point Cloud

Library (PCL). In IEEE International Conference on

Robotics and Automation (ICRA), pages 1 – 4.

Segal, A., Haehnel, D., and Thrun, S. (2009). Generalized-

ICP. In Robotics: Science and Systems, volume 5,

pages 168–176.

Seraﬁn, J. and Grisetti, G. (2015). NICP: Dense normal

based point cloud registration. In IEEE International

Conference on Intelligent Robots and Systems, vol-

ume 2015-Decem, pages 742–749.

Tazir, M. L., Gokhool, T., Checchin, P., Malaterre, L., and

Trassoudaine, L. (2018). CICP: Cluster Iterative Clos-

est Point for sparse-dense point cloud registration.

Robotics Auton. Syst., 108:66–86.

Wang, Y., Bu, S., Chen, L., Dong, Y., Li, K., Cao, X.,

and Li, K. (2023). HybridFusion: LiDAR and Vision

Cross-Source Point Cloud Fusion.

Wang, Z., Wang, X., Fang, B., Yu, K., and Ma, J. (2021).

Vehicle detection based on point cloud intensity and

distance clustering. Journal of Physics: Conference

Series, 1748(4):042053.

Xie, Z., Liang, P., Tao, J., Zeng, L., Zhao, Z., Cheng, X.,

Zhang, J., and Zhang, C. (2022). An Improved Super-

voxel Clustering Algorithm of 3D Point Clouds for the

Localization of Industrial Robots. Electronics, 11(10).

Xue, G., Wei, J., Li, R., and Cheng, J. (2022). LeGO-

LOAM-SC: An Improved Simultaneous Localization

and Mapping Method Fusing LeGO-LOAM and Scan

Context for Underground Coalmine. Sensors, 22(2).

Zou, Z., Lang, H., Lou, Y., and Lu, J. (2023). Plane-based

global registration for pavement 3D reconstruction us-

ing hybrid solid-state LiDAR point cloud. Automation

in Construction, 152:104907.

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

686