Monocular Simultaneous Localization and Mapping Initialization Method Based on Point and Line Features

Qi Qian; Tingming Bai; Yunfeng Bi; Chengyu Qiao; Zhiyu Xiang

doi:10.3788/AOS202141.1215002

Acta Optica Sinica, Volume. 41, Issue 12, 1215002(2021)

Monocular Simultaneous Localization and Mapping Initialization Method Based on Point and Line Features

Qi Qian¹, Tingming Bai¹, Yunfeng Bi¹, Chengyu Qiao¹, and Zhiyu Xiang^1,2、*

Author Affiliations

¹College of Information Science and Electronic Engineering, Zhejiang University, Hangzhou, Zhejiang 310027, China

²Zhejiang Provincial Key Laboratory of Information Processing Communication and Networking, Hangzhou, Zhejiang 310027, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Epipolar geometry of a point

Download full size

Fig. 2. Epipolar geometry of a line segment

Download full size

Fig. 3. Overlap length of the matched line segment feature

Download full size

Fig. 4. Four cases of overlapping line segments

Download full size

Fig. 5. Common view relationship of keyframe, map points and lines

Download full size

Fig. 6. Schematic diagram of the reprojection error. (a) Point feature; (b) line feature

Download full size

Fig. 7. Initialization results of two methods on the TUM dataset. (a) Our method; (b) traditional method

Download full size

Fig. 8. Variation curve of ATE of the system with N

Download full size

Fig. 9. Trajectories of the two methods on the cafe2 sequence. (a) Our method; (b) traditional method

Download full size

Fig. 10. Initialization results of the two methods on the corridor1-1 sequence. (a) Our method; (b) traditional method

Download full size

Table 1. Positioning of the system before and after line feature optimizationunit: cm
View table
Table 1. Positioning of the system before and after line feature optimizationunit: cm
TUM sequence ORB-SLAM ORB-SLAM+Line init
f2_xyz 1.61 1.24
f2_desk 1.54 1.48
f2_desk_person 1.22 1.16

Table 2. ATE of different algorithms on TUM-RGBD sequencesunit: cm

View table

Table 2. ATE of different algorithms on TUM-RGBD sequencesunit: cm

TUM-RGBD sequence	LSD-SLAM	ORB-SLAM	Ours
f1_xyz	9.04	1.38	0.96
f2_xyz	2.21	0.98	0.93
f2_rpy	ambiguity	ambiguity	1.48
f2_desk_person	4.60	0.89	1.12
f3_long_office	38.12	1.42	1.38
f3_nstr_tex_near	6.50	1.34	1.83
f3_nstr_tex_far	18.32	ambiguity	3.65
f3_sitting	7.71	2.76	2.11
f3_walking	15.20	ambiguity	2.84

Table 3. ATE of different methods on the TUM dataset unit: cm

View table

Table 3. ATE of different methods on the TUM dataset unit: cm

TUM-RGBD	PL-SLAM	ORB-SLAM	Ours
f1_xyz	3.02	1.38	1.00
f2_xyz	1.13	0.98	0.97
f2_rpy	ambiguity	ambiguity	1.48
f2_desk_person	1.63	0.89	1.34
f3_nstr_tex_far	ambiguity	ambiguity	3.65
f3_sitting	3.00	2.76	2.65
f3_walking	ambiguity	ambiguity	2.45

Table 4. [in Chinese]

View table

Table 4. [in Chinese]

OpenLORISsequence	PL-SLAM	ORB-SLAM	Ours
office1	0.061	0.059	0.053
office2	0.146	0.154	0.137
office3	0.041	0.040	0.031
office4	0.141	0.137	0.134
office5	ambiguity	ambiguity	0.035
office6	0.063	0.058	0.061
office7	0.043	0.043	0.041
cafe1	0.163	0.167	0.120
cafe2	0.255	0.267	0.128
corridor1-1-part1	0.654	0.671	0.231
corridor1-1-part2	0.633	0.624	0.280
corridor1-1-part3	0.267	0.290	0.223
corridor1-1-part4	0.403	0.415	0.121

Tools

Get Citation

Copy Citation Text

Qi Qian, Tingming Bai, Yunfeng Bi, Chengyu Qiao, Zhiyu Xiang. Monocular Simultaneous Localization and Mapping Initialization Method Based on Point and Line Features[J]. Acta Optica Sinica, 2021, 41(12): 1215002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites