A Method for Ground Filtering of Laser Point Cloud and Extraction of Tunnel Wall and Target Sphere

Yong Li; Yinzheng Luo; Qipeng Pu; Mingfei Han; Feng Shuang

doi:10.3788/CJL220845

Chinese Journal of Lasers, Volume. 50, Issue 13, 1310001(2023)

A Method for Ground Filtering of Laser Point Cloud and Extraction of Tunnel Wall and Target Sphere

Yong Li^1,3, Yinzheng Luo¹, Qipeng Pu¹, Mingfei Han^2、*, and Feng Shuang¹

¹Guangxi Key Laboratory of Intelligent Control and Maintenance of Power Equipment, School of Electrical Engineering, Guangxi University, Nanning 530004, Guangxi, China

²Department of Digital Information, Hebei Institute of International Business and Economics, Qinhuangdao 066311,Hebei, China

³Artificial Intelligence Key Laboratory of Sichuan Province, Yibin 644000, Sichuan, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Flowchart of the proposed algorithm. (a) Ground filtering processing; (b) tunnel wall extraction process; (c) target ball extraction process

Download full size

Fig. 2. Flowchart of ground filtering algorithm combining RANSAC plane fitting with pass-through filtering based on normal evaluation

Download full size

Fig. 3. Tunnel wall and target ball processing state at each stage. (a) DBSCAN coarse segmentation of region of interest; (b) area where target balls may be distributed; (c) tunnel wall; (d) spherical surface fitting; (e) visualization of fitting target balls; (f) distribution of fitted target balls in original tunnel

Download full size

Fig. 4. Original tunnel point cloud data. (a) Schematic of Scene 1; (b) schematic of Scene 2

Download full size

Fig. 5. Schematic of volume of intersecting part of two spheres

Download full size

Fig. 6. Poor filtering effect. (a) Ground owe-filtering during pass-through filtering; (b) ground over-filtering during pass-through filtering; (c) ground anomaly during RANSAC plane fitting

Download full size

Fig. 7. Filtering error accuracy and Kappa coefficient distribution. (a) Total error of nine filtering methods; (b) Kappa coefficient of nine filtering methods

Download full size

Fig. 8. Ground filtering effects for Scene 1 ( red point indicates tunnel ground point cloud and blue point indicates non-tunnel ground point cloud). (a) Ground truth; (b) method 1; (c) method 2; (d) method 3; (e) method 4; (f) method 5; (g) method 6; (h) method 7; (i) method 8; (j) our method

Download full size

Fig. 9. Ground filtering effect for Scene 2 ( red point indicates tunnel ground point cloud and blue point indicates non-tunnel ground point cloud). (a) Ground truth; (b) method 1; (c) method 2; (d) method 3; (e) method 4; (f) method 5; (g) method 6; (h) method 7; (i) method 8; (j) our method

Download full size

Table 1. Point statistics of experimental dataset
View table
Table 1. Point statistics of experimental dataset
Scene
Quantity of ground
points
Quantity of off-ground points
Scene 1 2139686 13360430
Scene 2 870063 3024774

Table 2. Construction modes of each ground filtering method

View table

Table 2. Construction modes of each ground filtering method

Method	Ground filter method
Method 1	Cloth simulation filter（CSF）^［19］
Method 2	Slope filter^［21］
Method 3	Region growing filter^［26］
Method 4	Grid elevation filter^［35］
Method 5	Pass-through filter
Method 6	RANSAC
Method 7	RANSAC+normal estimation
Method 8	RANSAC+normal estimation+pass-through filter
Ours	Voxel downsampling+RANSAC+normal estimation+pass-through filter

Table 3. Filtering accuracy evaluation results

View table

Table 3. Filtering accuracy evaluation results

Scene	Method	$a_{0}$	$b_{0}$	$c_{0}$	$d_{0}$	E_{Type I} /%	E_TypeⅡ /%	E_Total /%	$K_{K a p p a}$ /%	Time /s
Scene 1	Method 1	2117946	21740	220753	13139677	1.00	1.65	1.56	93.67	43.56
	Method 2	941344	1198342	5382	13355048	56.01	0.04	7.77	57.39	92.59
	Method 3	1114268	1025418	447867	12912563	47.92	3.35	9.50	54.95	9869.62
	Method 4	1864664	267431	808615	13092999	12.85	2.00	3.50	85.27	46.11
	Method 5	2139660	26	385993	12974437	0.01	2.89	2.49	90.27	4.337
	Method 6	1280132	859554	1828	13358602	40.17	0.01	5.56	71.92	354.45
	Method 7	1567680	572006	2978	13357452	26.73	0.02	3.71	82.45	360.44
	Method 8	2036749	102937	6954	13353476	4.81	0.05	0.71	96.96	357.36
	Ours	2124060	15626	68303	13292127	0.73	0.50	0.54	97.75	8.73
Scene 2	Method 1	831525	38538	168974	2855800	4.43	5.59	5.33	85.42	12.78
	Method 2	699708	170355	7154	3017620	19.57	0.24	4.56	85.92	22.89
	Method 3	264733	605330	510681	2514093	69.57	16.88	28.65	14.09	899.45
	Method 4	855390	14673	87157	2937617	1.69	2.88	2.61	92.68	11.60
	Method 5	870040	23	133780	2890994	0.03	4.42	3.44	90.61	1.18
	Method 6	708615	161448	2211	3022563	18.56	0.07	4.20	87.05	88.77
	Method 7	756647	113416	3674	3021100	13.04	0.12	3.01	90.93	94.38
	Method 8	869016	1047	63285	2961489	0.12	2.09	1.65	95.36	92.13
	Ours	868328	1735	43540	2981234	0.20	1.44	1.16	96.71	3.50
Average accuracy of our method for two scenes						0.47	0.97	0.85	97.23	6.12

Table 4. Fitting results of target sphere

View table

Table 4. Fitting results of target sphere

Sphere

Method

Radius /m

Center coordinate

E_{d i s t a n c e}

R_{f i t}

Sphere 01

$R M S$

$M a n u a l + R A N S A C$

$M a n u a l + N L S$

$D B R T S + N L S$

0.07035

0.07027

0.07167

0.07128

［3.83904，-16.09806，902.3149］

［3.83896，-16.09745，902.3160］

［3.83973，-16.09924，902.3151］

［3.83933，-16.09884，902.3151］

0.00129 0.00138

0.00136

98.4655 99.9975

99.8583

Sphere 02

$R M S$

$M a n u a l + R A N S A C$

$M a n u a l + N L S$

$D B R T S + N L S$

0.07106

0.07208

0.07153

0.07099

［0.31248，-12.10221，901.3812］

［0.31056，-12.10076，901.3817］

［0.31239，-12.10131，901.3812］

［0.31255，-12.10242，901.3810］

0.00160 0.00091

0.00029

99.1556 99.7815

99.5304

Sphere 03

$R M S$

$M a n u a l + R A N S A C$

$M a n u a l + N L S$

$D B R T S + N L S$

0.07080

0.07219

0.06996

0.06954

［13.3178，-1.87194，901.2911］

［13.3082，-1.87553，901. 2895］

［13.3148，-1.87130，901.28954］

［13.3143，-1.87129，901.28960］

0.01036

0.00339 0.00387

91.6317 94.4794

92.9250

Tools

Get Citation

Copy Citation Text

Yong Li, Yinzheng Luo, Qipeng Pu, Mingfei Han, Feng Shuang. A Method for Ground Filtering of Laser Point Cloud and Extraction of Tunnel Wall and Target Sphere[J]. Chinese Journal of Lasers, 2023, 50(13): 1310001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites