Small Target Objects and 3D Scene Reconstruction Based on Center Alignment and Improved FPFH

Yige Zhao; Kai Wang; Xiaoke Zhang; Chen Yang; Hui Chen

doi:10.3788/LOP241888

Laser & Optoelectronics Progress, Volume. 62, Issue 8, 0815011(2025)

Small Target Objects and 3D Scene Reconstruction Based on Center Alignment and Improved FPFH

Yige Zhao¹, Kai Wang², Xiaoke Zhang³, Chen Yang¹, and Hui Chen^1、*

Author Affiliations

¹College of Automation Engineering, Shanghai University of Electric Power, Shanghai 200090, China

²Anhui Energy Hefei Power Generation Co., Ltd., Hefei 230011, Anhui , China

³State Grid Henan Electric Power Company Electric Power Science Research Institute, Zhengzhou 450052, Henan , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(19)

Fig. 1. 3D scene reconstruction process based on center point scale transformation and improved FPFH

Download full size

Fig. 2. Small target object point cloud acquisition

Download full size

Fig. 3. Scene point cloud acquisition

Download full size

Fig. 4. Scale transformation schematic. (a) Vertex and center of; (b) calculation of scaling factor

Download full size

Fig. 5. Schematic of point cloud coarse registration process

Download full size

Fig. 6. Schematic of local point cloud in the scene. (a) Local observation point cloud; (b) local coordinate system

Download full size

Fig. 7. Distance from the point to the corresponding target point tangent plane

Download full size

Fig. 8. Scene RGB image

Download full size

Fig. 9. Public dataset information. (a) 3DCSR dataset; (b) Stanford 3D Scene Data

Download full size

Fig. 10. Scale transform visualization. (a) Initial pose; (b) scale transformation method based on centroid; (c) scale transformation method based on center point

Download full size

Fig. 11. Comparison of registration results. (a) Scale-Transformed; (b) FGR+ICP; (c) FPFH+ICP; (d) proposed method

Download full size

Fig. 12. Comparison of alignment results on public dataset 3DCSR dataset: (a) Original data; (b) scale-Transformed; (c) proposed method

Download full size

Fig. 13. Visual comparison of reconstruction results. (a) Original scene reconstruction; (b) proposed method

Download full size

Fig. 14. Visual comparison of 3DGS reconstruction results

Download full size

Table 1. Detailed information on small target objects and scene point clouds
View table
Table 1. Detailed information on small target objects and scene point clouds
Dataset Point cloud Dataset Point cloud
Object1 4096 Scene1 1739155
Object2 4096 Scene2 312304
Object3 4096 Scene3 698163
Object4 4096 Scene4 441770
Object5 4096 Scene5 1548097
Object6 4096 Scene6 181995

Table 2. Scale transformation factor size for different methods

View table

Table 2. Scale transformation factor size for different methods

Dataset	$s_{p r}^{c e n t r o i d}$	$s_{p r}^{c e n t e r}$	$s_{g t}$	$E_{s}^{c e n t r o i d}$	$E_{s}^{c e n t e r}$
Scene1	6.7334	5.4687	5.3465	1.3869	0.1222↓
Scene2	5.5334	4.0629	4.0282	1.5052	0.0347↓
Scene3	1.0781	1.0701	1.0751	0.0030↓	0.0050
Scene4	1.7505	1.4357	1.4250	0.3255	0.0107↓
Scene5	2.2863	2.5260	2.4860	0.1997	0.0400↓
Scene6	1.9079	1.4844	1.4580	0.4499	0.0264↓

Table 3. Comparison of coarse/fine registration errors of point clouds in different scenarios

View table

Table 3. Comparison of coarse/fine registration errors of point clouds in different scenarios

Dataset	Registration method	Coarse registration		Fine registration
Dataset	Registration method	$E_{R}$ /（°）	$E_{t}$ /mm	$E_{R}$ /（°）	$E_{t}$ /mm
Scene1	FGR+ICP	20.35	19.82	3.51	2.35
	FPFH+ICP	26.58	27.35	5.56	3.22
	Proposed	9.47	9.86	1.24	0.47
Scene2	FGR+ICP	8.24	5.38	2.05	0.58
	FPFH+ICP	48.58	12.11	10.35	6.38
	Proposed	5.76	2.34	3.12	0.69
Scene3	FGR+ICP	9.86	8.56	2.89	3.56
	FPFH+ICP	25.30	22.35	5.59	6.37
	Proposed	3.54	1.65	0.23	0.84
Scene4	FGR+ICP	22.56	25.68	6.89	8.75
	FPFH+ICP	156.38	126.70	32.56	25.49
	Proposed	5.34	6.37	0.22	0.57
Scene5	FGR+ICP	65.45	62.58	15.64	18.67
	FPFH+ICP	48.66	45.85	21.56	25.63
	Proposed	12.54	10.67	2.56	1.35
Scene6	FGR+ICP	84.69	88.77	19.68	14.67
	FPFH+ICP	58.94	66.78	9.57	11.25
	Proposed	34.56	34.75	1.58	2.75

Table 4. Registration results on public 3DCSR dataset
View table
Table 4. Registration results on public 3DCSR dataset
Dataset Scene1 Scene2 Scene3 Scene4 Scene5
RMSE 1.8954 1.7619 2.0583 2.0736 2.1772

Table 5. Comparison results of proposed method and 3DGS on public dataset Stanford 3D Scene Data
View table
Table 5. Comparison results of proposed method and 3DGS on public dataset Stanford 3D Scene Data
Dataset Registration method Small target object RMSE Overall scene RMSE Completion time /s
Scene1 3DGS 5.4424 6.5423 990
Proposed 3.1835 3.8832 901
Scene2 3DGS 8.4108 8.9312 1106
Proposed 5.2605 4.0472 862

Tools

Get Citation

Copy Citation Text

Yige Zhao, Kai Wang, Xiaoke Zhang, Chen Yang, Hui Chen. Small Target Objects and 3D Scene Reconstruction Based on Center Alignment and Improved FPFH[J]. Laser & Optoelectronics Progress, 2025, 62(8): 0815011

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites