Multi-Sensor Fusion SLAM Algorithm Based on Line Feature Optical Flow

Fig. 4. Line features obtained by two strategies applied to the image. (a) Matching results obtained by LSD+LBD; (b) tracking results obtained by ELSED+LK

Download full size

Fig. 5. Trajectories of various methods on the hall_03 sequence. (a) VINS-MONO; (b) PL-VINS; (c) LINS; (d) FAST-LIO2; (e) LIO-SAM; (f) LVI-SAM

Download full size

Fig. 6. Trajectories and their zoom-in views of various methods on the street_08 sequence

Download full size

Table 1. Comparison of the number of line features extracted and the time consumed by two strategies
View table
Table 1. Comparison of the number of line features extracted and the time consumed by two strategies
Module LSD+LBD ELSED+LK
Number of extracted line features 28 20
Number of line feature pairs 7 20
Time consumed to extract features /ms 81.134 76.709
Time consumed in tracking features /ms 3.247 1.232

Table 2. Features of used squences

View table

Table 2. Features of used squences

Squence	Environment feature	Motion feature
Door_02	Day, outdoor to indoor, glass curtain wall	Short-term
Hall_03	Room, glass curtain wall	Randon walk
Room_02	Room, bright	Quick walk
Rotation_01	Night, outdoor	Quick rotation
Street_08	Night, outdoor	Loop back, zigzag, long-term

Table 3. Comparison of localization accuracy obtained by different algorithms on five sequences of M2DGR dataset

View table

Table 3. Comparison of localization accuracy obtained by different algorithms on five sequences of M2DGR dataset

Sequence	VINS-MONO		PL-VINS		LINS		FAST-LIO2	LIO-SAM		LVI-SAM		PL2I-SLAM （ours）
Sequence	（w/o loop）	（w/ loop）	（w/o loop）	（w/ loop）	（w/o loop）	（w/ loop）	（w/o loop）	（w/o loop）	（w/ loop）	（w/o loop）	（w/ loop）	（w/o loop）	（w/ loop）
Door_02	2.3589	1.9379	3.1291	3.0259	1.5507	0.2092	0.3452	0.1970	0.1974	0.1975	0.1960	0.1956	0.1961
Hall_03	2.1822	1.5542	2.8527	3.1949	2.5638	0.3341	0.5767	0.5370	0.5547	0.6310	0.4839	0.3423	0.3308
Room_02	0.3325	0.2641	0.2963	0.2958	2.9108	1.8515	0.3240	0.1287	0.1283	0.1282	0.1275	0.1280	0.1261
Rotation_01	1.8112	1.3635	3.3769	3.7484	6.8797	0.8522	1.3710	0.4223	0.6284	0.7217	0.5550	0.5732	0.5183
Street_08	8.8959	5.7682	Fail	Fail	14.0657	0.2023	0.2235	2.0550	1.5278	0.2788	0.2522	0.2109	0.2059

Table 4. Time consumed by two algorithms to process a single frame of data at different stages

View table

Table 4. Time consumed by two algorithms to process a single frame of data at different stages

Sub-system	Thread	Module	Time /ms
Sub-system	Thread	Module	LVI-SAM	PL2I-SLAM
VIS	1	Point extraction	5.948	4.487
	1	Point track	0.817	1.039
	2	Line extraction	—	3.936
	2	Line track	—	2.061
	3	Pose optimization	15.983	21.343
	4	Loop detection	31.048	32.923
	Total		53.796	65.249
LIS	1	Feature extraction	3.863	4.949
	2	Match and optimization	24.823	26.151
	3	Loop detection	53.421	53.563
	Total		82.107	84.663

Tools

Get Citation

Copy Citation Text

Yuanbin Chi, Xiangyin Meng, Shide Xiao, Xiujie Lu, Shouye Wu. Multi-Sensor Fusion SLAM Algorithm Based on Line Feature Optical Flow[J]. Acta Optica Sinica, 2025, 45(8): 0815002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Dec. 7, 2024

Accepted: Feb. 17, 2025

Published Online: Apr. 27, 2025

The Author Email: Xiangyin Meng (xymeng@swjtu.edu.cn)

DOI:10.3788/AOS241859

CSTR:32393.14.AOS241859

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of the number of line features extracted and the time consumed by two strategies

Table 1. Comparison of the number of line features extracted and the time consumed by two strategies

Table 2. Features of used squences

Table 2. Features of used squences

Table 3. Comparison of localization accuracy obtained by different algorithms on five sequences of M2DGR dataset

Table 3. Comparison of localization accuracy obtained by different algorithms on five sequences of M2DGR dataset

Table 4. Time consumed by two algorithms to process a single frame of data at different stages

Table 4. Time consumed by two algorithms to process a single frame of data at different stages