Real-Time Stereo Matching Algorithm with Hierarchical Refinement

Yufeng Wang; Hongwei Wang; Yu Liu; Mingquan Yang; Jicheng Quan

doi:10.3788/AOS202040.0915002

Journals >Acta Optica Sinica >Volume 40 >Issue 9 >Page 0915002 > Article

Acta Optica Sinica
Vol. 40, Issue 9, 0915002 (2020)

Real-Time Stereo Matching Algorithm with Hierarchical Refinement

Yufeng Wang^1、2, Hongwei Wang^2、3、**, Yu Liu², Mingquan Yang², and Jicheng Quan^1、2、*

Author Affiliations

¹University of Naval Aviation, Yantai, Shandong 264001, China

²Aviation University of Air Force, Changchun, Jilin 130022, China

³Information Engineering University, Zhengzhou, Henan 450001, China

show less

DOI: 10.3788/AOS202040.0915002 Cite this Article Set citation alerts

Yufeng Wang, Hongwei Wang, Yu Liu, Mingquan Yang, Jicheng Quan. Real-Time Stereo Matching Algorithm with Hierarchical Refinement[J]. Acta Optica Sinica, 2020, 40(9): 0915002 Copy Citation Text

show less

Fig. 1. Whole architecture of proposed algorithm

Download full size

Fig. 2. Structure of FEM

Download full size

Fig. 3. Three DRM prototypes. (a) Prototype 1; (b) prototype 2; (c) prototype 3

Download full size

Fig. 4. Structure of MBF

Download full size

Fig. 5. Hierarchical results of proposed method. (a) Left image; (b)

{\hat{d}}_{5}

; (c)

{\hat{d}}_{4 <}

Download full size

Fig. 6. Visual results of proposed method. (a) Left image; (b) α_F=0.000; (c) α_F=0.001; (d) α_F=0.010; (e) α_F=0.100; (f) α_F=1.000

Download full size

Fig. 7. Visual results of proposed method. (a) Left image; (b) disparity map; (c) error map; (d) local details

Download full size

b	L=2	L=3	L=4	L=5	L=6	L=7
1	1.6	8.5	20.8	26.1	24.0	23.7
2	1.6	8.5	20.8	24.2	24.4	22.0
3	1.6	8.3	20.7	22.7	23.0	19.9
4	1.6	8.1	19.7	21.8	21.2	20.5
5	1.6	8.1	19.5	21.0	20.6	20.5
6	1.6	8.0	18.5	20.4	20.6	19.5

Table 1. Operating rates of proposed method under different L and bframe·s^-1

Parameter	L=4	L=5	L=6	L=7
${E_{ep}}_{}$ /pixel	1.108	1.106	1.117	1.180
${E_{D 1}}_{}$ /%	4.922	4.826	4.772	4.999

Table 2. Performance evaluation of proposed method under different L(b=3)

Parameter	b=1	b=2	b=3	b=4	b=5	b=6
${E_{ep}}_{}$ /pixel	1.133	1.115	1.106	1.097	1.077	1.089
${E_{D 1}}_{}$ /%	4.870	4.859	4.826	4.668	4.636	4.724

Table 3. Performance evaluation of proposed method under different b (L=5)

Prototype of DRM	FEM		${E_{ep}}_{}$ /pixel	${E_{D 1}}_{}$ /%	${f_{run}}_{}$ /(frame·s^-1)
Prototype of DRM	Fusion	SPP	${E_{ep}}_{}$ /pixel	${E_{D 1}}_{}$ /%	${f_{run}}_{}$ /(frame·s^-1)
Prototype 1	√	√	2.580	17.028	21.5
Prototype 2	√	√	1.517	6.753	19.1
Prototype 3	√	√	1.077	4.636	21.0
Prototype 3	√		1.105	4.707	21.9
Prototype 3		√	1.141	4.924	24.3

Table 4. Performance evaluation of models with different versions (L=5,b=5)

α_F	E_ep/pixel	E_D1/%
0.000	0.776	2.926
0.001	0.773	2.879
0.010	0.785	2.940
0.100	0.758	2.822
1.000	0.766	2.853

Table 5. Performance evaluation of proposed method under different α_F(L=5,b=5)

Method	All			Noc			Runtime /s
Method	${E_{D 1 ⁃ bg}}_{}$ /%	${E_{D 1 ⁃ fg}}_{}$ /%	${E_{D 1 ⁃ all}}_{}$ /%	${E_{D 1 ⁃ bg}}_{}$ /%	${E_{D 1 ⁃ fg}}_{}$ /%	${E_{D 1 ⁃ all}}_{}$ /%	Runtime /s
M2S_CSPN^[25]	1.51	2.88	1.74	1.40	2.67	1.61	0.50
GANet-deep^[26]	1.48	3.46	1.81	1.34	3.11	1.63	1.80
WSMCnetEB^[23]	1.72	4.19	2.13	1.51	3.57	1.85	0.39
DeepPruner(best)^[22]	1.87	3.56	2.15	1.71	3.18	1.95	0.18
PSMNet^[18]	1.86	4.62	2.32	1.71	4.31	2.14	0.41
iResNet-i2^[15]	2.25	3.40	2.44	4.11	3.72	4.05	0.12
CRL^[14]	2.48	3.59	2.67	2.32	3.12	2.45	0.47
GC-net^[17]	2.21	6.16	2.87	2.02	5.58	2.61	0.90
PBCP^[10]	2.58	8.74	3.61	2.27	7.71	3.17	68.00
SGM-Net^[11]	2.66	8.64	3.66	2.23	7.44	3.09	67.00
MC-CNN-arct^[4]	2.89	8.88	3.89	2.48	7.64	3.33	67.00
DeepPruner(fast)^[22]	2.32	3.91	2.59	2.13	3.43	2.35	0.06
DispNetC^[13]	4.32	4.41	4.34	4.11	3.72	4.05	0.06
MADnet^[28]	3.75	9.20	4.66	3.45	8.41	4.27	0.02
Proposed	2.59	4.80	2.96	2.22	4.14	2.54	0.05

Table 6. Performance evaluation of different methods on KITTI2015 test dataset

Yufeng Wang, Hongwei Wang, Yu Liu, Mingquan Yang, Jicheng Quan. Real-Time Stereo Matching Algorithm with Hierarchical Refinement[J]. Acta Optica Sinica, 2020, 40(9): 0915002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information