Multi-Factor Segmentation of Point Cloud Based on Improved Multi-Rule Region Growing

Wenqi Wang; Zongchun Li; Yongjian Fu; Hua He; Feng Xiong

doi:10.3788/AOS202141.0528001

Journals >Acta Optica Sinica >Volume 41 >Issue 5 >Page 0528001 > Article

Acta Optica Sinica
Vol. 41, Issue 5, 0528001 (2021)

Multi-Factor Segmentation of Point Cloud Based on Improved Multi-Rule Region Growing

Wenqi Wang, Zongchun Li^*, Yongjian Fu, Hua He, and Feng Xiong

Author Affiliations

Institute of Geospatial Information, PLA Strategic Support Force Information Engineering University, Zhengzhou, Henan 450001, China

show less

DOI: 10.3788/AOS202141.0528001 Cite this Article Set citation alerts

Wenqi Wang, Zongchun Li, Yongjian Fu, Hua He, Feng Xiong. Multi-Factor Segmentation of Point Cloud Based on Improved Multi-Rule Region Growing[J]. Acta Optica Sinica, 2021, 41(5): 0528001 Copy Citation Text

show less

Fig. 1. An example of point cloud multi-factor segmentation based on traditional RG algorithm

Download full size

Fig. 2. An example of point cloud multi-factor segmentation based on MRG algorithm

Download full size

Fig. 3. An example of normal vector. (a) Point cloud; (b) normal vector of point cloud

Download full size

Fig. 4. An example of linear and planar point set segmentation. (a) Linear point set; (b) planar point set

Download full size

Fig. 5. Schematic diagram of threshold v_t setting

Download full size

Fig. 6. An example of volume change based on correct merging (ΔV=2.13). (a) Unmerged segments; (b) convex hulls of unmerged segments; (c) merged segment; (d) convex hull of merged segments

Download full size

Fig. 7. An example of volume change based on incorrect merging (ΔV=17214.04). (a) Unmerged segments; (b) convex hulls of unmerged segments; (c) merged segment; (d) convex hull of merged segments

Download full size

Fig. 8. Pseudo-code of merge strategy

Download full size

Fig. 9. Point cloud data set. (a) Airborne point cloud (scene I); (b) terrestrial point cloud (scene II); (c) vehicle-borne point cloud (scene III)

Download full size

Fig. 10. Reference data of each scene. (a) Scene I; (b) scene II; (c) scene III

Download full size

Fig. 11. Segmentation results of planar point set in Scene II and Scene III (N_t=15°). (a) MRG in Scene II; (b) IMRG in Scene II; (c) MRG in Scene III; (d) IMRG in Scene III

Download full size

Fig. 12. Incorrect segmentation. (a) RG+mean shift; (b) CSF+DBSCAN; (c) GMIC

Download full size

Method	T_P	F_P
RG+mean shift	351708	29669
CSF+DBSCAN	364594	34771
IMRG	340502	38946

Table 1. 1 Segmentation results of ground in Scene I

Method	N_t /(°)	P /%			R /%			F₁ /%
Method	N_t /(°)	Scene I	Scene II	Scene III	Scene I	Scene II	Scene III	Scene I	Scene II	Scene III
	5	97.38	99.23	99.40	84.96	99.06	99.14	90.75	99.15	99.27
MRG	10	97.32	99.40	99.34	77.80	97.41	75.80	86.47	98.40	85.99
	15	93.80	72.38	99.41	68.09	99.90	34.56	78.91	83.94	51.29
	5	97.56	99.33	99.45	87.68	99.00	99.25	92.35	99.16	99.35
IMRG	10	97.50	99.41	99.52	90.94	98.96	88.12	94.11	99.19	93.48
	15	97.23	99.43	99.38	89.91	96.91	64.50	93.43	98.16	78.23

Table 1. Segmentation evaluation of building plane in each scene

Method	F₁ /%							Time /s
Method	Building plan	Ground		Vegetation		Car		Time /s
Method 1	98.56		98.19		98.41		93.89	954.01
Method 2	97.52		98.69		97.01		92.72	1750.89
Method 3	93.08		96.18		89.58		82.65	1218.09
Method 4	99.01		98.71		98.90		95.39	987.68
Method 5	91.92		96.10		89.47		79.11	5922.04

Table 1. 0 F₁ score of segmentation in Scene III

Method	N_t /(°)	P /%			R /%			F₁ /%
Method	N_t /(°)	Scene I	Scene II	Scene III	Scene I	Scene II	Scene III	Scene I	Scene II	Scene III
	5	88.41	97.64	98.61	87.87	98.09	99.03	88.14	97.87	98.82
MRG	10	85.78	93.77	84.42	90.65	98.55	99.31	88.15	96.10	91.26
	15	83.09	93.65	67.05	90.55	3.66	99.54	86.66	7.05	80.13
	5	90.11	97.51	98.75	86.23	98.34	99.07	88.13	97.92	98.91
IMRG	10	89.85	97.39	91.50	89.81	98.56	99.31	89.83	97.97	95.25
	15	89.27	92.66	78.99	90.58	98.63	99.45	89.92	95.55	88.05

Table 2. Segmentation evaluation of ground in each scene

Scene	d_t /m	s_t /m	V_t
Scene I	1.54	0.40	18.71
Scene II	0.25	0.05	16.33
Scene III	0.13	0.03	14.61

Table 3. Merge threshold parameters

Type of ground object		Building plan	Ground	Vegetation
	P /%	89.70	87.53	91.77
Strategy① (M=1150)	R /%	80.50	87.92	92.14
	F₁ /%	84.85	87.72	91.96
	P /%	96.44	89.74	92.99
Strategy② (M=1515)	R /%	90.43	89.19	94.14
	F₁ /%	93.34	89.46	93.56

Table 4. Merge results in Scene I

Type of ground object		Building plan	Ground	Vegetation	Power line
	P /%	97.67	91.15	99.78	95.72
Strategy①(M=1550)	R /%	96.22	97.25	96.47	60.97
	F₁ /%	96.94	94.10	98.10	74.49
	P /%	99.38	97.42	99.78	99.40
Strategy②(M=2730)	R /%	98.97	98.45	99.79	98.36
	F₁ /%	99.17	97.94	99.79	98.88

Table 5. Merge results in Scene II

Type of ground object		Building plan	Ground	Vegetation	Car
	P /%	97.09	98.30	98.75	94.67
Strategy①(M=1673)	R /%	98.65	98.18	96.40	94.14
	F₁ /%	97.86	98.24	97.56	94.41
	P /%	98.96	98.72	99.01	95.35
Strategy②(M=2386)	R /%	99.07	98.71	98.79	95.44
	F₁ /%	99.01	98.71	98.90	95.39

Table 6. Merge results in Scene III

Scene	CSF		DBSCAN
Scene	l_c /m	h_c /m	r_d /m	m_d
Scene I	0.8	0.5	2	5
Scene II	2	0.4	0.5	10
Scene III	2	0.5	0.5	30

Table 7. Parameters for Method 4

Method	F₁ /%			Time /s
Method	Buildingplan	Ground	Vegetation	Time /s
Method 1	76.84	85.03	90.62	294.03
Method 2	92.99	92.17	94.71	4697.64
Method 3	83.70	93.35	94.42	3230.18
Method 4	93.34	89.46	93.56	372.53

Table 8. F₁ score of segmentation in Scene I

Method	F₁ /%							Time /s
Method	Building plan	Ground		Vegetation		Power line		Time /s
Method 1	97.00		94.56		97.53		54.95	499.06
Method 2	98.94		97.57		98.74		93.61	943.72
Method 3	98.64		97.51		99.19		79.06	1523.06
Method 4	99.17		97.94		99.79		98.88	830.65
Method 5	97.90		95.71		96.36		78.59	4627.44

Table 9. F₁ score of segmentation in Scene II

Wenqi Wang, Zongchun Li, Yongjian Fu, Hua He, Feng Xiong. Multi-Factor Segmentation of Point Cloud Based on Improved Multi-Rule Region Growing[J]. Acta Optica Sinica, 2021, 41(5): 0528001

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information