RGB-D Visual Odometry Combined with Points and Lines

Junxin Lu; Zhijun Fang; Jieyu Chen; Yongbin Gao

doi:10.3788/AOS202141.0415001

Journals >Acta Optica Sinica >Volume 41 >Issue 4 >Page 0415001 > Article

Acta Optica Sinica
Vol. 41, Issue 4, 0415001 (2021)

RGB-D Visual Odometry Combined with Points and Lines

Junxin Lu, Zhijun Fang^*, Jieyu Chen, and Yongbin Gao

Author Affiliations

College of Electrical and Electronic Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

show less

DOI: 10.3788/AOS202141.0415001 Cite this Article Set citation alerts

Junxin Lu, Zhijun Fang, Jieyu Chen, Yongbin Gao. RGB-D Visual Odometry Combined with Points and Lines[J]. Acta Optica Sinica, 2021, 41(4): 0415001 Copy Citation Text

show less

Fig. 1. Framework of system

Download full size

Results of each line extraction algorithm in scene of low-texture. (a) Hough transform algorithm; (b) EDLines algorithm; (c) LSD algorithm; (d) CannyLines algorithm

Fig. 2. Results of each line extraction algorithm in scene of low-texture. (a) Hough transform algorithm; (b) EDLines algorithm; (c) LSD algorithm; (d) CannyLines algorithm

Download full size

Fig. 3. Local neighborhood of line midpoint

Download full size

Fig. 4. Dataset of ICL-NUIM

Download full size

Fig. 5. Results of proposed algorithm on ICL-NUIM dataset. (a) Input images; (b) point and line features extracted from the images; (c) estimated camera trajectories

Download full size

Fig. 6. Some challenging scenarios of visual odometry. (a) White walls; (b) ceilings; (c) corridors; (d) scenes of illumination change

Download full size

Fig. 7. Estimated camera trajectories of proposed algorithm on TUM dataset. (a) fr1/desk; (b) fr2/desk; (c) fr3/long_office; (d) fr3/cabinet; (e) fr3/str_ntex_near; (f) fr3/str_ntex_far

Download full size

Fig. 8. Intuitive diagram of relative pose error of proposed algorithm on TUM dataset. (a) fr1/desk; (b) fr3/long_office; (c) fr3/str_ntex_far

Download full size

Fig. 9. Estimated camera trajectories of proposed algorithm on CoRBS dataset. (a) D1 sequence; (b) E1 sequence; (c) E4 sequence; (d) H1 sequence

Download full size

Fig. 10. Location accuracy and convergence time of proposed algorithm using I and

ξ_{c - 1, k}^{n}

initial

ξ_{ck}^{0}

. (a) Comparison of ATE RMSE; (b) comparison of convergence time

Download full size

Fig. 11. Location accuracy ATE RMSE and convergence time of our algorithm using

ξ_{c - 1, k}^{n}

and

ξ_{c - 1, k}^{n}

􀱋ξ_c_-2,_c_-1 initial

ξ_{ck}^{0}

. (a) Comparison of ATE RMSE; (b) comparison of convergence time

Download full size

Sequence	lr-kt0	lr-kt1	lr-kt2	lr-kt3	of-kt0	of-kt1	of-kt2	of-kt3
ORB-SLAM2(VO)	0.0943	0.0612	0.0669	0.1777	0.1577	0.0669	0.0909	0.0710
REVO	0.1486	0.0501	0.0256	0.0509	0.0541	1.9124	0.0318	0.0314
LPVO	0.0150	0.0390	0.0340	0.0520	0.0610	0.0520	0.0390	0.0300
Proposed	0.0094	0.0110	0.0162	0.0212	0.0181	0.0106	0.0171	0.0123

Table 1. [in Chinese]

Sequence	REVO	PL-SVO	DSO	ORB-SLAM2(VO)	DLGO	Proposed
fr1/rpy	0.0373	-	-	0.0368	-	0.0154
fr1/xyz	0.0236	0.1089	0.0882	0.0173	0.0674	0.0317
fr2/xyz	0.0581	0.0209	0.0772	0.0038	0.0808	0.0088
fr2/desk	0.1747	0.0693	1.810	0.0289	1.6400	0.0067
fr3/long_office	0.0195	0.1660	1.512	0.0198	1.4640	0.0186
Average(rich-texture)	0.0626	0.0913	0.8718	0.0213	0.8131	0.0142
fr3/str_ntex_near	0.0167	-	-	×	-	0.0212
fr3/str_ntex_far	0.0169	-	1.052	×	0.9460	0.0176
fr3/cabinet	0.0703	-	1.560	0.0731	1.5700	0.0087
fr3/nstr_ntex_far	0.1445	-	0.876	×	0.7400	0.0442
Average(texture-less)	0.0621	-	1.1627	0.0731	1.0853	0.0229

Table 2. [in Chinese]

Sequence	REVO	PL-SVO	DSO	ORB-SLAM2(VO)	DLGO	Proposed
fr1/rpy	0.0504	-	-	0.0289	-	0.0132
fr1/xyz	0.0562	0.0167	0.0690	0.0107	0.0538	0.0222
fr2/xyz	0.2219	0.0503	0.0619	0.0070	0.0653	0.0073
fr2/desk	1.0098	0.0645	1.6500	0.4125	1.3300	0.0054
fr3/long_office	0.4980	0.1637	1.1800	0.4403	1.1680	0.0372
Average(rich-texture)	0.3673	0.0738	0.7402	0.1799	0.6543	0.0171
fr3/str_ntex_near	0.0134	-	-	×	-	0.0127
fr3/str_ntex_far	0.0110	-	0.7900	×	0.8650	0.0103
fr3/cabinet	0.2982	-	1.0800	×	1.0500	0.0061
fr3/nstr_ntex_far	0.2607	-	0.6770	×	0.5040	0.0304
Average(texture-less)	0.1458	-	0.8490	×	0.8063	0.0149

Table 3. [in Chinese]

Item	Sequence	D1	E1	E4	H1	Average
	DVO	0.0596	0.0335	0.0335	0.0616	0.0470
RPE RMSE /(m·s^-1)	ORB-SLAM2(VO)	0.0203	0.0288	0.0291	0.0233	0.0254
	REVO	0.0132	0.0477	0.0260	0.0116	0.0246
	Proposed	0.0097	0.0165	0.0051	0.0077	0.0096
	ORB-SLAM2(VO)	0.0387	0.0461	0.0528	0.0715	0.0523
ATE RMSE /(m·s^-1)	REVO	0.0234	0.1314	0.0346	0.0322	0.0554
	Proposed	0.0178	0.0337	0.0253	0.0269	0.0184