Dynamic SLAM Algorithm Based on Object Detection and Point-Line Feature Association

Wenxuan Deng; Jianwu Dang; Jiu Yong

doi:10.3788/LOP242050

Laser & Optoelectronics Progress, Volume. 62, Issue 10, 1015007(2025)

Dynamic SLAM Algorithm Based on Object Detection and Point-Line Feature Association

Wenxuan Deng¹, Jianwu Dang^1,2、*, and Jiu Yong²

Author Affiliations

¹School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, Gansu , China

²National Virtual Simulation Experimental Teaching Center for Rail Transit Information and Control, Lanzhou 730070, Gansu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(17)

Fig. 1. Overall architecture of the LDF-SLAM algorithm

Download full size

Fig. 2. Overall structure of improved object detection network

Download full size

Fig. 3. Overall structure of ResAM module

Download full size

Fig. 4. Schematic diagrams of multi-view geometry method. (a) Key point q' belongs to a static object; (b) key point q' belongs to a dynamic object

Download full size

Fig. 5. Test results. (a) Line feature extraction map without setting optimization threshold; (b) line feature extraction map after setting optimization threshold

Download full size

Fig. 6. Some samples showing the detection results of SLAM algorithm fusion object detection on TUM-working highly dynamic sequence data set

Download full size

Fig. 7. Final distribution of dynamic feature points

Download full size

Fig. 8. Comparison of final feature extraction results. (a) Static feature points that have not been matched with line features; (b) successfully matched static feature points after adding line features

Download full size

Fig. 9. Comparison of LDF-SLAM and ORB-SLAM3 motion trajectory maps under different sequences of TUM data set

Download full size

Fig. 10. Sample images of different sequences in TUM data set

Download full size

Fig. 11. Dense point cloud maps after removing dynamic point clouds

Download full size

Table 1. RMSE comparison of the absolute trajectory error of LDF-SLAM and ORB-SLAM3 on TUM data set

View table

Table 1. RMSE comparison of the absolute trajectory error of LDF-SLAM and ORB-SLAM3 on TUM data set

Sequence	RMSE /m		Error reduction amplitude /%
Sequence	ORB-SLAM3 （RGB-D）	LDF-SLAM	Error reduction amplitude /%
sitting_static	0.009	0.005	44.5
sitting_xyz	0.044	0.010	77.3
sitting_halfsphere	0.047	0.009	80.6
walking_static	0.015	0.006	60.5
walking_xyz	0.270	0.011	95.9
walking_halfsphere	0.291	0.016	94.5

Table 2. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on TUM data set

View table

Table 2. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on TUM data set

Sequence	RMSE /m
Sequence	ORB-SLAM3	DynaSLAM	ReFusion	ACEFusion	OA-SLAM	LDF-SLAM
sitting_static	0.048	0.007	0.011	0.028	0.040	0.005
sitting_xyz	0.044	0.015	0.026	0.021	0.294	0.010
sitting_halfsphere	0.047	0.028	0.038	0.035	0.169	0.009
walking_static	0.015	0.007	0.017	0.011	0.092	0.006
walking_xyz	0.270	0.017	0.099	0.025	0.286	0.011
walking_halfsphere	0.291	0.026	0.104	0.035	0.363	0.016

Table 3. RMSE comparison of the relative pose error translation part of LDF-SLAM and other dynamic SLAM algorithms on TUM data set

View table

Table 3. RMSE comparison of the relative pose error translation part of LDF-SLAM and other dynamic SLAM algorithms on TUM data set

Sequence	RMSE /m
Sequence	ORB-SLAM3	DynaSLAM	ReFusion	ACEFusion	OA-SLAM	LDF-SLAM
sitting_static	0.015	0.009	0.014	0.031	0.043	0.007
sitting_xyz	0.048	0.018	0.029	0.026	0.295	0.011
sitting_halfsphere	0.049	0.033	0.043	0.039	0.173	0.010
walking_static	0.018	0.009	0.021	0.018	0.095	0.007
walking_xyz	0.289	0.022	0.101	0.031	0.289	0.011
walking_halfsphere	0.301	0.028	0.109	0.039	0.368	0.015

Table 4. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on Bonn data set

View table

Table 4. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on Bonn data set

Sequence	RMSE /m
Sequence	ORB-SLAM3	DynaSLAM	ReFusion	ACEFusion	OA-SLAM	LDF-SLAM
crowd	0.996	0.017	0.138	0.098	0.105	0.010
p_no_box	0.749	0.057	0.102	0.121	0.367	0.031
r_no_box	0.011	0.010	0.026	0.035	0.052	0.009
person_tracking	0.593	0.391	0.245	0.085	0.252	0.022

Table 5. Performance comparison of object detection algorithms
View table
Table 5. Performance comparison of object detection algorithms
Algorithm Parameter quantity /10⁶ Inference time /ms
YOLOv3 68.5 312.7
YOLOv5 11.6 30.8
YOLOv8 11.2 26.7
Ours 4.5 19.1

Table 6. RMSEcomparison of absolute trajectory error of each algorithm on TUM data set

View table

Table 6. RMSEcomparison of absolute trajectory error of each algorithm on TUM data set

Sequence	RMSE /m
Sequence	YOLO-SLAM	PL-SLAM	LDF-SLAM
sitting_static	0.015	0.009	0.005
sitting_xyz	0.021	0.014	0.010
sitting_halfsphere	0.032	0.021	0.009
walking_static	0.019	0.012	0.006
walking_xyz	0.024	0.018	0.011
walking_halfsphere	0.056	0.029	0.016

Tools

Get Citation

Copy Citation Text

Wenxuan Deng, Jianwu Dang, Jiu Yong. Dynamic SLAM Algorithm Based on Object Detection and Point-Line Feature Association[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1015007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Sep. 26, 2024

Accepted: Nov. 18, 2024

Published Online: Apr. 27, 2025

The Author Email: Jianwu Dang (dangjw@mail.lzjtu.cn)

DOI:10.3788/LOP242050

CSTR:32186.14.LOP242050

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. RMSE comparison of the absolute trajectory error of LDF-SLAM and ORB-SLAM3 on TUM data set

Table 1. RMSE comparison of the absolute trajectory error of LDF-SLAM and ORB-SLAM3 on TUM data set

Table 2. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on TUM data set

Table 2. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on TUM data set

Table 3. RMSE comparison of the relative pose error translation part of LDF-SLAM and other dynamic SLAM algorithms on TUM data set

Table 3. RMSE comparison of the relative pose error translation part of LDF-SLAM and other dynamic SLAM algorithms on TUM data set

Table 4. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on Bonn data set

Table 4. RMSE comparison of absolute trajectory error of LDF-SLAM with other dynamic SLAM algorithms on Bonn data set

Table 5. Performance comparison of object detection algorithms

Table 5. Performance comparison of object detection algorithms

Table 6. RMSEcomparison of absolute trajectory error of each algorithm on TUM data set

Table 6. RMSEcomparison of absolute trajectory error of each algorithm on TUM data set