Dynamic SLAM Algorithm Based on Instance Segmentation and Optical Flow Feature Clustering

Heng Zhang; Xiaoqiang Zhang; Guanwu Jiang; Zhixin Zhang; Yang He; Xuliang Wang

doi:10.3788/LOP241797

Laser & Optoelectronics Progress, Volume. 62, Issue 8, 0815008(2025)

Dynamic SLAM Algorithm Based on Instance Segmentation and Optical Flow Feature Clustering

Heng Zhang¹, Xiaoqiang Zhang^1、*, Guanwu Jiang^1,2, Zhixin Zhang¹, Yang He¹, and Xuliang Wang²

Author Affiliations

¹School of Information Engineering, Southwest University of Science and Technology, Mianyang 621010, Sichuan , China

²Engineering Research Center for Integration and Application of Digital Learning Technology, Ministry of Education, Beijing 100039, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(23)

Fig. 1. Overall system framework

Download full size

Fig. 2. YOLACT network structure

Download full size

Fig. 3. Potential dynamic object detection and elimination

Download full size

Fig. 4. Comparison of optical flow feature in different scenes

Download full size

Fig. 5. Gaussian mixture model based on optical flow feature clustering

Download full size

Fig. 6. Static key-point filtering

Download full size

Fig. 7. BEBLID descriptor extraction process

Download full size

Fig. 8. Match result comparison

Download full size

Fig. 9. Comparison of trajectory and error between ORB-SLAM2 and proposed algorithm

Download full size

Fig. 10. Comparison of trajectory and error between DS-SLAM and proposed algorithm

Download full size

Table 1. Matching accuracy comparison
View table
Table 1. Matching accuracy comparison
Sequence BRIEF BEBLID
f1_xyz 71.7 84.7
f1_rpy 70.4 81.1
f1_desk 73.3 81.4

Table 2. ATE comparison of YOLACT and ORB-SLAM2
View table
Table 2. ATE comparison of YOLACT and ORB-SLAM2
Sequence ORB-SLAM2 /m YOLACT /m
RMSE S.D RMSE S.D
w_half 0.6995 0.2677 0.0219 0.0136
s_static 0.0085 0.0041 0.0064 0.0032

Table 3. RPE translation of YOLACT and ORB-SLAM2
View table
Table 3. RPE translation of YOLACT and ORB-SLAM2
Sequence ORB-SLAM2 /m YOLACT /m
RMSE S.D RMSE S.D
w_half 0.0594 0.0544 0.0132 0.0082
s_static 0.0053 0.0027 0.0048 0.0024

Table 4. RPE rotation of YOLACT and ORB-SLAM2
View table
Table 4. RPE rotation of YOLACT and ORB-SLAM2
Sequence ORB-SLAM2 /（°） YOLACT /（°）
RMSE S.D RMSE S.D
w_half 1.2843 1.1377 0.4209 0.2358
s_static 0.1633 0.0851 0.1638 0.0859

Table 5. ATE comparison of YOLACT+LK and ORB-SLAM2
View table
Table 5. ATE comparison of YOLACT+LK and ORB-SLAM2
Sequence ORB-SLAM2 /m YOLACT+LK /m
RMSE S.D RMSE S.D
w_half 0.6995 0.2677 0.0206 0.0116
s_static 0.0085 0.0041 0.0059 0.0032

Table 6. RPE translation of YOLACT+LK and ORB-SLAM2
View table
Table 6. RPE translation of YOLACT+LK and ORB-SLAM2
Sequence ORB-SLAM2/m YOLACT+LK /m
RMSE S.D RMSE S.D
w_half 0.0594 0.0544 0.0128 0.0077
s_static 0.0053 0.0027 0.0047 0.0025

Table 7. RPE rotation of YOLACT+LK and ORB-SLAM2
View table
Table 7. RPE rotation of YOLACT+LK and ORB-SLAM2
Sequence ORB-SLAM2 /m YOLACT+LK /m
RMSE S.D RMSE S.D
w_half 1.2843 1.1377 0.4105 0.2212
s_static 0.1633 0.0851 0.1588 0.0854

Table 8. Comparison of ATE between BEBLID and ORB-SLAM2
View table
Table 8. Comparison of ATE between BEBLID and ORB-SLAM2
Sequence ORB-SLAM2 /m BEBLID+BOW /m
RMSE S.D RMSE S.D
w_half 0.6995 0.2677 0.4509 0.2479
s_static 0.0085 0.0041 0.0080 0.0038

Table 9. RPE translation of BEBLID and ORB-SLAM2
View table
Table 9. RPE translation of BEBLID and ORB-SLAM2
Sequence ORB-SLAM2 /m BEBLID+BOW /m
RMSE S.D RMSE S.D
w_half 0.0594 0.0544 0.0234 0.0151
s_static 0.0053 0.0027 0.0043 0.0021

Table 10. RPE rotation of BEBLID and ORB-SLAM2
View table
Table 10. RPE rotation of BEBLID and ORB-SLAM2
Sequence ORB-SLAM2 /m BEBLID+BOW /m
RMSE S.D RMSE S.D
w_half 1.2843 1.1377 0.6043 0.3454
s_static 0.1633 0.0851 0.1566 0.0808

Table 11. ATE comparison results

View table

Table 11. ATE comparison results

Sequence	ORB-SLAM2 /m		DS-SLAM /m		DynaSLAM /m		Blitz-SLAM /m		Algorithm of reference ［12］ /m		Algorithm of reference ［15］ /m		Proposed algorithm /m
Sequence	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D
w_xyz	0.7991	0.4597	0.1982	0.0093	0.0184	0.0087	0.0153	0.0078	0.0191	0.01068	0.0133	0.0068	0.0149	0.0083
w_static	0.4157	0.1906	0.0069	0.0029	0.0064	0.0025	0.0102	0.0052	0.0073	0.0034	0.0071	0.0030	0.0083	0.0045
w_rpy	0.6935	0.3259	0.1574	0.0942	0.0425	0.0282	0.0356	0.0220	0.4077	0.2052	0.0275	0.0158	0.0298	0.0189
w_half	0.6995	0.2677	0.0328	0.0169	0.0250	0.0103	0.0256	0.0126	0.0288	0.0157	0.0241	0.0112	0.0175	0.0084
s_static	0.0085	0.0041	0.0064	0.0030	0.0064	0.0031			0.0061	0.0032	0.0064	0.0033	0.0055	0.0029

Table 12. RPE translation part comparison results

View table

Table 12. RPE translation part comparison results

Sequence	ORB-SLAM2 /m		DS-SLAM /m		DynaSLAM /m		Blitz-SLAM /m		Algorithm of reference ［12］ /m		Algorithm of reference ［15］ /m		Proposed algorithm /m
Sequence	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D
w_xyz	0.0270	0.0159	0.0151	0.0099	0.0196	0.0121	0.0197	0.0096	0.02808	0.0152	0.0179	0.0090	0.0129	0.0082
w_static	0.0156	0.0114	0.0071	0.0043	0.0074	0.0040	0.0129	0.0069	0.0105	0.0047	0.0092	0.0048	0.0058	0.0037
w_rpy	0.0285	0.0184	0.0252	0.0193	0.0433	0.0342	0.0473	0.0283	0.1389	0.1027	0.0381	0.0262	0.0210	0.0144
w_half	0.0594	0.0544	0.0148	0.0093	0.0179	0.0102	0.0253	0.0123	0.0403	0.0199	0.0251	0.0122	0.0122	0.0070
s_static	0.0053	0.0027	0.0063	0.0030	0.0138	0.0080			0.0091	0.0044	0.0076	0.0038	0.0048	0.0024

Table 13. RPE rotation part comparison results

View table

Table 13. RPE rotation part comparison results

Sequence	ORB-SLAM2 /（°）		DS-SLAM /（°）		DynaSLAM /（°）		Blitz-SLAM /（°）		Algorithm of reference ［12］ /（°）		Algorithm of reference ［15］ /（°）		Proposed algorithm /（°）
Sequence	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D	RMSE	S.D
w_xyz	0.6480	0.3949	0.8633	0.6026	0.5946	0.3494	0.6132	0.3348	0.7179	0.4562			0.4083	0.2883
w_static	0.3191	0.2036	0.1608	0.0821	0.2126	0.1217	0.3038	0.1437	0.2863	0.1244			0.1803	0.0948
w_rpy	0.6906	0.4126	0.5847	0.3812	1.0239	0.7642	1.0841	0.6668	2.7803	2.0133			0.5378	0.3684
w_half	1.2843	1.1377	0.4148	0.2330	0.5735	0.3285	0.7879	0.3751	0.9302	0.4632			0.3925	0.2063
s_static	0.1633	0.0851	0.4766	0.2246	0.4146	0.2213			0.3162	0.1410			0.1610	0.0838

Tools

Get Citation

Copy Citation Text

Heng Zhang, Xiaoqiang Zhang, Guanwu Jiang, Zhixin Zhang, Yang He, Xuliang Wang. Dynamic SLAM Algorithm Based on Instance Segmentation and Optical Flow Feature Clustering[J]. Laser & Optoelectronics Progress, 2025, 62(8): 0815008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites