Dynamic ORB-SLAM3 Optimization Method for Adaptive Separation of Dynamic Targets

Fig. 3. Comparisons between improved and original algorithm for extracting feature points in TUM dataset. (a) ORB-SLAM3 algorithm; (b) GCNv2 algorithm

Download full size

Fig. 4. Result of target detection and segmentation

Download full size

Fig. 5. Process of adaptive threshold setting

Download full size

Fig. 6. Dynamic feature point elimination process

Download full size

Fig. 7. Comparison of translation error between proposed algorithm and ORB-SLAM3. ORB-SLAM3: (a) walking-rpy, (b) walking-halfsphere, (c) walking-static, (d) walking-xyz; proposed algorithm: (e) walking-rpy, (f) walking-halfsphere, (g) walking-static, (h) walking-xyz

Download full size

Fig. 8. Comparisons of trajectory errors between proposed algorithm and ORB-SLAM3. ORB-SLAM3: (a) walking-rpy, (b) walking-halfsphere, (c) walking-static, (d) walking-xyz; proposed algorithm: (e) walking-rpy, (f) walking-halfsphere, (g) walking-static, (h) walking-xyz

Download full size

Fig. 9. Dense 3D point cloud map construction. (a) ORB-SLAM3 global dense map construction; (b) improved dense global map construction; (c) before local filtering of dynamic feature points; (d) after partial filtering of dynamic feature points

Download full size

Table 1. Presents comparison of RMSE for ATE before and after adding adaptive threshold in TUM dataset
View table
Table 1. Presents comparison of RMSE for ATE before and after adding adaptive threshold in TUM dataset
Sequence name Before /m After /m Performance improvement/%
fr3_walking_xyz 0.0171 0.0132 23
fr3_walking_static 0.0103 0.0061 41
fr3_walking_rpy 0.1468 0.0576 61
fr3_walking_hs 0.0219 0.0172 21

Table 2. Experimental environment
View table
Table 2. Experimental environment
Configuration name Argument
Processor 11th Gen Intel Core i7-11800H
Dominant frequency 2.30 GHz
Internal memory 16 G
Graphics card NVDIA GeForce RTX 3070 LapTop GPU
Operating system Ubuntu 18.04

Table 3. Comparisons of ATE between ORB-SLAM3 and proposed algorithm

View table

Table 3. Comparisons of ATE between ORB-SLAM3 and proposed algorithm

Dataset name	ORB-SLAM3		Textual algorithm		Performance improvement
Dataset name	RMSE /m	SD /m	RMSE /m	SD /m	RMSE /%	SD /%
fr3_walking_xyz	0.2623	0.1371	0.0129	0.0067	95	95
fr3_walking_static	0.0214	0.0092	0.0080	0.0032	63	65
fr3_sitting_static	0.0075	0.0036	0.0059	0.0030	21	17
fr3_walking_rpy	0.1522	0.0663	0.0588	0.0461	61	30
fr3_walking_hs	0.2402	0.0850	0.0182	0.0085	92	89
fr3_sitting_hs	0.0229	0.0126	0.0167	0.0083	27	34

Table 4. Comparisons of RPE between ORB-SLAM3 and proposed algorithm

View table

Table 4. Comparisons of RPE between ORB-SLAM3 and proposed algorithm

Dataset name	ORB-SLAM3		Textual algorithm		Performance improvement
Dataset name	RMSE/m	SD/m	RMSE/m	SD/m	RMSE/%	SD/%
fr3_walking_xyz	0.0112	0.0062	0.0101	0.0056	10	10
fr3_walking_static	0.0052	0.0033	0.0014	0.0011	73	67
fr3_sitting_static	0.0055	0.0030	0.0044	0.0023	20	23
fr3_walking_rpy	0.0211	0.0152	0.0058	0.0036	73	76
fr3_walking_hs	0.0124	0.0068	0.0119	0.0063	4	7
fr3_sitting_hs	0.0131	0.0075	0.0076	0.0040	42	47

Table 5. Comparison of ATE of proposed algorithm and ORB-SLAM2 in KITTI dataset
View table
Table 5. Comparison of ATE of proposed algorithm and ORB-SLAM2 in KITTI dataset
Sequence name ORB-SLAM2 Proposed algorithm
KITTI02 2.8134 2.3056
KITTI03 0.3420 0.3011
KITTI05 0.4006 0.2965

Table 6. Comparison of RMSE for ATE of proposed algorithm with other dynamic SLAM algorithms in TUM dataset

View table

Table 6. Comparison of RMSE for ATE of proposed algorithm with other dynamic SLAM algorithms in TUM dataset

Sequence name	ORB-SLAM3	Dyna-SLAM	DS-SLAM	Textual algorithm
fr3_walking_xyz	0.2428	0.0130	0.1204	0.0135
fr3_walking_static	0.0217	0.0062	0.0065	0.0053
fr3_walking_rpy	0.1503	0.0260	0.1373	0.0527
fr3_walking_hs	0.2240	0.0171	0.0206	0.0166

Tools

Get Citation

Copy Citation Text

Tao Song, Daiheng Yue, Yichen Yang, Ting Chen, Yuan Gong. Dynamic ORB-SLAM3 Optimization Method for Adaptive Separation of Dynamic Targets[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1015003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites