Loop Closure Detection of Visual SLAM Based on Improved LBD and Data-Dependent Measure

Shi Jiahao; Meng Qinghao; Dai Xuyang

doi:10.3788/LOP202158.0615001

Laser & Optoelectronics Progress, Volume. 58, Issue 6, 615001(2021)

Loop Closure Detection of Visual SLAM Based on Improved LBD and Data-Dependent Measure

Shi Jiahao^1,2, Meng Qinghao^1,2, and Dai Xuyang^1,2、*

Author Affiliations

¹School of Electrical and Information Engineering,Tianjin University, Tianjin 300072, China

²Institute of Robotics and Autonomous System, Tianjin University, Tianjin 300072, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Illustration of LBD and binary conversion

Download full size

Fig. 2. Distribution of band descriptors. (a) Line 1; (b) Line 2

Download full size

Fig. 3. Flowchart of loop closure detection

Download full size

Fig. 4. Results of line band descriptor matching. (a) Rotation and scale change; (b) viewpoint change; (c) illumination change

Download full size

Fig. 5. Performance of different methods on different datasets. (a) CC dataset; (b) NC dataset; (c) L6I dataset; (d) L6O dataset

Download full size

Fig. 6. Performance of loop closure detection algorithms with different descriptors. (a) CC dataset; (b) NC dataset; (c) L6I dataset; (d) L6O dataset

Download full size

Fig. 7. P-R curves on different datasets

Download full size

Table 1. Comparison of band descriptors
View table
Table 1. Comparison of band descriptors
Number of bits 1 2 3 4 5 6 7 8
Line 1 1 1 1 1 1 1 1 1
Line 2 1 1 1 1 1 1 1 1

Table 2. Datasets for loop closure detection
View table
Table 2. Datasets for loop closure detection
Dataset Environment Number of images Image size
Lip6 Indoor(L6I) Indoor 388 240×192
Lip6 Outdoor(L6O) Outdoor 531 240×192
City Center(CC) Outdoor 2146 640×480
New College(NC) Outdoor 2474 640×480

Table 3. Accuracy of line band descriptor matching

View table

Table 3. Accuracy of line band descriptor matching

Parameter		Rotation and scale	Viewpoint	Illumination
Number of all matches		173	257	159
LBD	Number of correct matches	146	168	144
LBD	Accuracy/%	84.39	65.37	90.57
Improved LBD	Number of correct matches	160	195	155
Improved LBD	Accuracy/%	92.49	75.88	97.48

Table 4. Maximum recall at different features and 100% accuracy unit: %
View table
Table 4. Maximum recall at different features and 100% accuracy unit: %
Dataset Point feature LBD Improved LBD Point and line feature
CC 50.62 59.89 67.55 69.88
NC 73.37 46.73 63.44 76.51
L6I 81.82 60.00 72.73 86.36
L6O 66.11 14.29 24.58 71.10

Table 5. Average execution time of loop closure detection unit: ms/frame
View table
Table 5. Average execution time of loop closure detection unit: ms/frame
Dataset ORB+LBD+L₁ norm method Proposed method
Feature processing Loop detection Feature processing Loop detection
CC 140.9 7.7 142.9 39.6
NC 153.9 6.9 157.5 34.5
L6O 23.7 1.2 24.3 3.6
L6I 12.2 0.3 12.4 0.7

Table 6. Maximum recall of different algorithms at 100% accuracy unit: %

View table

Table 6. Maximum recall of different algorithms at 100% accuracy unit: %

Dataset	Algorithm in Ref.[8]	Algorithm in Ref.[10]	Algorithm in Ref.[12]	Algorithm in Ref.[21]	Algorithm in Ref.[22]	Proposed algorithm
CC	37	43.03	67.59	52.36	74.01	69.88
NC	47		57.14		67.05	76.51
L6I	23.64		80.45	42.32		86.36
L6O			69.65	49.55		71.10

Tools

Get Citation

Copy Citation Text

Shi Jiahao, Meng Qinghao, Dai Xuyang. Loop Closure Detection of Visual SLAM Based on Improved LBD and Data-Dependent Measure[J]. Laser & Optoelectronics Progress, 2021, 58(6): 615001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites