Dynamic SLAM Modeling Algorithm for RGB-D Images Fused with 3D Gaussian Splatting

Lanxin Zhang; De'er Liu

doi:10.3788/LOP242480

Laser & Optoelectronics Progress, Volume. 62, Issue 16, 1612002(2025)

Dynamic SLAM Modeling Algorithm for RGB-D Images Fused with 3D Gaussian Splatting

Lanxin Zhang^1,2 and De'er Liu^1,2、*

¹Jiangxi Provincial Key Laboratory of Water Ecological Conservation in Headwater Regions, Jiangxi University of Science and Technology, Ganzhou 341000, Jiangxi , China

²School of Civil Engineering and Surveying & Mapping Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, Jiangxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Proposed method flow chart

Download full size

Fig. 2. Moving object detection flow chart

Download full size

Fig. 3. Visualization of moving object segmentation

Download full size

Fig. 4. Preprocessing effect

Download full size

Fig. 5. Different datasets used in the study. (a) TUM RGB-D dataset; (b) Davis2017 dataset; (c) RealSense self-building dataset; (d) Replica dataset

Download full size

Fig. 6. Image reconstruction results from different datasets. (a) Static scene reconstruction results in TUM RGB-D dataset; (b) partial static scene reconstruction results of dynamic data in TUM RGB-D dataset; (c)(d) reconstruction results in Realsense self-building dataset

Download full size

Fig. 7. Precision of the 3D model verifies the point cloud effect of the experiment. (a) Point cloud collected by the proposed method on the RealSense self-building dataset; (b) collected results of 3D laser scanner; (c) point cloud collected by the proposed method on the Replica dataset; (d) original model in the Replica dataset

Download full size

Fig. 8. Reconstruction results of dynamic scenes

Download full size

Table 1. Comparison of the improved results
View table
Table 1. Comparison of the improved results
Method Flow prompt Davis17（IoU↑）
SAM 2 × 69.9
Proposed √ 86.1

Table 2. Comparison of ATE

View table

Table 2. Comparison of ATE

Sequence	ORB-SLAM3		RDS-SLAM		DS-SLAM		Proposed
Sequence	RMSE /m	SD /m	RMSE /m	SD /m	RMSE /m	SD /m	RMSE /m	SD /m
f3walkxyz	0.274	0.124	0.057	0.023	0.024	0.016	0.014	0.007
f3walkhalf	0.274	0.111	0.080	0.043	0.031	0.014	0.028	0.015
f3walkrpy	0.661	0.291	0.160	0.087	0.444	0.235	0.032	0.018
fr3sitstatic	0.009	0.004	0.008	0.004	0.006	0.003	0.006	0.003

Table 3. Comparison of novel view synthesis precision and real-time performance

View table

Table 3. Comparison of novel view synthesis precision and real-time performance

Metric	Method	Room 0	Room 1	Room 2	Office 0	Office 1	Office 2	Office 3	Office 4	Average
PSNR /dB↑	Orbeez-SLAM	12.13	15.28	15.87	17.59	19.26	10.30	11.55	12.65	14.33
	GS-SLAM	31.56	32.86	32.59	38.70	41.17	32.36	32.03	32.92	34.27
	SplaTAM	32.60	33.55	34.83	38.09	39.02	31.95	29.53	31.55	33.89
	Gaussian-SLAM^［26］	38.88	41.80	42.44	46.40	45.29	40.10	39.06	42.65	42.08
	Proposed	32.20	35.36	34.42	40.31	40.75	33.85	34.08	36.47	35.93
SSIM↑	Orbeez-SLAM	0.699	0.787	0.808	0.795	0.672	0.783	0.760	0.838	0.768
	GS-SLAM	0.968	0.973	0.971	0.986	0.993	0.978	0.970	0.968	0.975
	SplaTAM	0.975	0.969	0.982	0.982	0.982	0.966	0.949	0.951	0.970
	Gaussian-SLAM^［26］	0.980	0.970	0.980	0.980	0.980	0.970	0.950	0.950	0.970
	Proposed	0.940	0.960	0.957	0.978	0.977	0.962	0.953	0.963	0.962
LPIPS↓	Orbeez-SLAM	0.612	0.477	0.509	0.515	0.413	0.554	0.540	0.528	0.529
	GS-SLAM	0.094	0.075	0.093	0.050	0.033	0.094	0.110	0.112	0.082
	SplaTAM	0.070	0.097	0.074	0.088	0.093	0.098	0.119	0.150	0.099
	Gaussian-SLAM^［26］	0.070	0.100	0.080	0.090	0.090	0.100	0.120	0.150	0.100
	Proposed	0.081	0.067	0.083	0.045	0.051	0.069	0.067	0.065	0.066
MappingFPS /（frame/s）↑	Orbeez-SLAM	24.05	24.78	24.00	19.22	26.74	23.88	24.45	26.11	24.15
	GS-SLAM	8.34	—	—	—	—	—	—	—	8.34
	SplaTAM	0.24	0.19	0.19	0.20	0.22	0.27	0.26	0.24	0.23
	Gaussian-SLAM^［26］	1.04	1.06	1.07	1.09	1.11	1.14	1.16	1.19	1.10
	Proposed	100.98	84.92	103.38	99.10	107.06	95.60	97.20	96.66	98.11
TrackingFPS /（frame/s）↑	Orbeez-SLAM	1.22	1.24	1.18	1.26	1.19	1.21	1.23	1.21	1.21
	GS-SLAM	1.20	1.16	1.25	1.19	1.23	1.27	1.18	1.21	1.21
	SplaTAM	1.16	1.23	1.20	1.18	1.25	1.17	1.24	1.20	1.20
	Gaussian-SLAM^［26］	1.15	1.22	1.24	1.18	1.26	1.19	1.21	1.23	1.21
	Proposed	135	136	138	142	143	145	147	148	142
RenderingFPS /（frame/s）↑	Orbeez-SLAM	1803	1780	1798	1812	1778	1806	1820	1766	1795
	GS-SLAM	2130	2140	2150	2165	2175	2180	2190	2198	2166
	SplaTAM	2133	2147	2151	2164	2173	2184	2187	2191	2167
	Gaussian-SLAM^［26］	2131	2143	2154	2167	2183	2185	2194	2176	2167
	Proposed	2126	2141	2156	2171	2182	2193	2121	2168	2157

Tools

Get Citation

Copy Citation Text

Lanxin Zhang, De'er Liu. Dynamic SLAM Modeling Algorithm for RGB-D Images Fused with 3D Gaussian Splatting[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1612002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites