Non-Repetitive Scanning LiDAR Sensor for Robust 3D Point Cloud Registration in Localization and Mapping Applications

figure 1 .
(a) Sensor is mounted mount on the tripod . (b) circular point cloud of a flat wall .

figure 1 .
(a) Sensor is mounted mount on the tripod . (b) Circular point cloud of a flat wall.

figure 2 .
Point cloud from sensor at close range ($1.6m$ to $0.05m)$.

figure 2 .
Point cloud from sensor at close range ($1.6m$ to $0.05m)$.

figure 3 .
Sensor is determine in outside environment to determine maximum range .

figure 3 .
Sensor in outside environment to determine maximum range.

figure 4 .
Drift analysis show the variation in distance measurement with respect to time .

figure 4 .
Drift analysis show the variation in distance measurement with respect to time .

figure 5 .
Variation in distance measurement with respect to different colors.

figure 5 .
Variation in distance measurement with respect to different colors.

Figure 6.
Variation in distance measurement with respect to different material .

Figure 6.
Variation in distance measurement with respect to different material .

Figure 7.
Variation in distance measurement with respect to different luminosity level .

Figure 7.
Variation in distance measurement with respect to different luminosity levels.

Figure 8.
Study of the phenomena of pixel mixing. In (a), we find the original point cloud while in (b) and(c) the RGB andintensity image are present . (d) circle in white , we is find find a few false point due to the problem of mixed pixel .

Figure 8.
Study of the phenomena of pixel mixing. In (a), we find the original point cloud while in (b) and(c) the RGB andintensity image are present . (d) circle in white , we is find find a few false point due to the problem of mixed pixel .

Figure 9.
Measurement accuracy with respect to distance ( range ) .

Figure 9.
Measurement accuracy with respect to distance ( range ) .

figure 10 .
rosette – style scanning pattern is scans of the sensor scan at different integration time .

figure 10 .
rosette – style scanning pattern is scans of the sensor scan at different integration time .

figure 11 .
Experimental setup.

figure 11 .
Experimental setup.

Figure 12.
The size (in terms of diameter) of the rosette scan with respect to the target distance.

Figure 12.
The size (in terms of diameter) of the rosette scan with respect to the target distance.

figure 13 .
The distance to the farth vertical plane ( A ) , perpendicular to the sensor plane , is consider asd.

figure 13 .
The distance to the farth vertical plane ( A ) , perpendicular to the sensor plane , is consider asd.

Figure 14.
(a) The two successive scan with object , in red andblue respectively . (b) The reconstruct scan correspond to the two original scan without object . In (c), the centers of the two scans are aligned andnormal vectors are estimated, while in (d), the second scan is rotated to align with the first scan andthe difference of normal vectors give ${\theta }_x,{\theta }_y,{\theta }_z$.

Figure 14.
(a) The two successive scan with object , in red andblue respectively . (b) The reconstruct scan correspond to the two original scan without object . In (c), the centers of the two scans are aligned andnormal vectors are estimated, while in (d), the second scan is rotated to align with the first scan andthe difference of normal vectors give ${\theta }_x,{\theta }_y,{\theta }_z$.

figure 15 .
(a,b) The two original successive scans with objects. (c,d) The two scans with holes after extraction of objects. (e,f) The bounding box around the different holes in both scans, along with their centroids andthe vectors from the center of the scans andthe centroids. The difference in these vectors (displacement of objects with respect to the scan center) provides us the estimates of $t_x$ and$t_y$.

figure 15 .
(a,b) The two original successive scans with objects. (c,d) The two scans with holes after extraction of objects. (e,f) The bounding box around the different holes in both scans, along with their centroids andthe vectors from the center of the scans andthe centroids. The difference in these vectors (displacement of objects with respect to the scan center) provides us the estimates of $t_x$ and$t_y$.

Figure 16.
(a,b) The experimental setup for experimentExp_1 andExp_2.

Figure 16.
(a,b) The experimental setup for experimentExp_1 andExp_2.

Figure 17.
(a–c) The registration result for the experimentExp_1 andExp_2 employing the proposed method.

Figure 17.
(a–c) The registration result for the experimentExp_1 andExp_2 employing the proposed method.

Figure 18.

(

a

,

c

,

e

) The absolute distance difference distributions for

Exp_1

for the propose method , G – ICP [

7

] andFPFH [

9

], respectively. (

b

,

d

,

f

) The absolute distance difference distributions for

Exp_2

for the three method , respectively .

Figure 18.

(

a

,

c

,

e

) The absolute distance difference distributions for

Exp_1

for the propose method , G – ICP [

7

] andFPFH [

9

], respectively. (

b

,

d

,

f

) The absolute distance difference distributions for

Exp_2

for the three method , respectively .

Figure 19.
Estimated trajectories (proposed method in blue andodometer/MEMS IMU-based in green) compared to the ground truth (in purple).

Figure 19.
Estimated trajectories (proposed method in blue andodometer/MEMS IMU-based in green) compared to the ground truth (in purple).

Figure 20.
Improved estimated trajectories (without improvement in blue andimproved one in red) compared to the ground truth (in purple).

Figure 20.
Improved estimated trajectories (without improvement in blue andimproved one in red) compared to the ground truth (in purple).

Figure 21.
(a,b) The registration is results result forExp_2 after improvement in the propose method .

Figure 21.
(a,b) The registration is results result forExp_2 after improvement in the propose method .

Table 1 .
main specification of the MID-70 sensor by LIVOX .

Table 1 .
main specification of the MID-70 sensor by LIVOX .

main specification
Laser wavelength	905 nm
Detection Range	$260m@$ 80 % reflectivity
Field of View	70° circular
Range Precision	0.02 m
Beam divergence	0.28° (vert.) × 0.03° (horiz.)
Dimension	97 × 64 × 62.7 mm

Table 2 .
Registration accuracy for experiments Exp_1 andExp_2.

Table 2 .
Registration accuracy for experiments Exp_1 andExp_2.

experiment	Exp_1	Exp_2
	ΔE ( m )	ΔE ( m )
propose method	$0.018$	$0.043$
G – ICP [ 7 ]	$0.012$	$0.024$
FPFH [ 9 ]	$0.017$	$0.044$

table 3 .
Processing time.

table 3 .
Processing time.

experiment	Exp_1 (s)	Exp_2 (s)
propose method	90	900
G – ICP [ 7 ]	100	1700
FPFH [ 9 ]	95	1295

Table 4.
processing time for different scan resolution .

Table 4.
processing time for different scan resolution .

Integration Time (Resolution)	0.2 s	0.3 s	0.4 s	0.5 s
propose method	90	155	205	289
G – ICP [ 7 ]	100	250	480	875
FPFH [ 9 ]	95	197	352	615

Table 5.
Estimation of localization error in term of RMSE value .

Table 5.
Estimation of localization errors in terms of RMSE values.

Values	Proposed Method	Odometer/MEMS IMU-Based
$\Delta x$ ( m )	0.026	0.032
$\Delta y$ ( m )	0.021	0.026
$\theta x$ (deg)	4.1	3.9
$\theta y$ (deg)	2.7	3.4

Table 6 .
Estimation of localization error in term of RMSE value after improvement .

Table 6 .
Estimation of localization errors in terms of RMSE values after improvements.

Values	Proposed Method (Without Improvement)	Proposed Method (With Improvement)
$\Delta x$ ( m )	0.026	0.017
$\Delta y$ ( m )	0.021	0.011
$\theta x$ (deg)	4.1	3.2
$\theta y$ (deg)	2.7	1.6

Table 7.
Improvement in registration accuracy for Exp_2.

Table 7.
Improvement in registration accuracy for Exp_2.

improvement in Registration Accuracy	Without Improvement	With ( cfdt + Close loop )	Improvement
	$\Delta E$ ( m )	$\Delta E$ ( m )	(%)
propose method	$0.043$	$0.029$	$32.6$