Multi-Scale Stereo Matching Based on Bayesian Reasoning

Cancan Zeng; Mingjun Ren; Gaobo Xiao; Yuehong Yin

doi:10.3788/AOS201737.1215004

Journals >Acta Optica Sinica >Volume 37 >Issue 12 >Page 1215004 > Article

Acta Optica Sinica
Vol. 37, Issue 12, 1215004 (2017)

Multi-Scale Stereo Matching Based on Bayesian Reasoning

Cancan Zeng, Mingjun Ren^*, Gaobo Xiao, and Yuehong Yin

Author Affiliations

School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China

show less

DOI: 10.3788/AOS201737.1215004 Cite this Article Set citation alerts

Cancan Zeng, Mingjun Ren, Gaobo Xiao, Yuehong Yin. Multi-Scale Stereo Matching Based on Bayesian Reasoning[J]. Acta Optica Sinica, 2017, 37(12): 1215004 Copy Citation Text

show less

Flow chart of multi-scale optimization algorithm. (a) Left and right images of binocular vision; (b) disparity maps with scale information; (c) disparity space constructed by disparity maps; (d) histogram of cumulative plausibility of one pixel; (e) final optimization results

Fig. 1. Flow chart of multi-scale optimization algorithm. (a) Left and right images of binocular vision; (b) disparity maps with scale information; (c) disparity space constructed by disparity maps; (d) histogram of cumulative plausibility of one pixel; (e) final optimization results

Download full size

(a) Reference images; disparity maps with different window sizes: (b) window diameter is 1/100 image width; (c) window diameter is 1/50 image width; (d) window diameter is 1/30 image width

Fig. 2. (a) Reference images; disparity maps with different window sizes: (b) window diameter is 1/100 image width; (c) window diameter is 1/50 image width; (d) window diameter is 1/30 image width

Download full size

Fig. 3. Comparison of experimental results. (a) Ground truth maps; (b) RealTime BP algorithm; (c) FastBilateral algorithm; (d) MSO algorithm; (e)-(g) corresponding mismatched regions

Download full size

Algorithm	Average	Tsukuba			Venus			Teddy			Cones
Algorithm	Average	nocc	all	disc	nocc	all	disc	nocc	all	disc	nocc	all	disc
Ref. [20]	9.82	2.05	4.22	10.60	1.92	2.98	20.30	7.23	14.40	17.60	6.41	13.70	16.50
Ref. [23]	9.34	3.19	5.05	9.85	0.57	2.13	5.30	10.60	19.60	22.10	5.52	15.90	12.30
Ref. [19]	7.69	1.49	3.40	7.87	0.77	1.90	9.00	8.72	13.20	17.20	4.61	11.60	12.40
Ref. [31]	7.65	1.71	2.22	6.74	0.55	0.87	2.88	9.90	15.00	19.50	6.66	12.30	13.40
Ref. [32]	7.31	2.38	2.80	10.40	0.34	0.92	4.55	9.83	15.30	20.30	3.10	9.31	8.59
MSO(BM)	5.96	2.31	2.67	9.63	0.32	0.54	2.43	7.87	10.50	16.62	2.87	7.78	8.03

Table 1. Performance comparison of proposed algorithm with other real-time algorithms

Algorithm	Average	Tsukuba			Venus			Teddy			Cones
Algorithm	Average	nocc	all	disc	nocc	all	disc	nocc	all	disc	nocc	all	disc
MSO(BM)	5.96	2.31	2.67	9.63	0.32	0.54	2.43	7.87	10.50	16.62	2.87	7.78	8.03
MSO(GF)	5.64	2.34	2.61	8.85	0.31	0.54	2.70	7.53	10.33	15.22	2.58	7.50	7.16

Table 2. Performance comparison of proposed algorithm by using BM algorithm and GF algorithm to create disparity space

Algorithm	Processor	Time /ms	MDE per second
Ref. [23]	2.5 GHz CPU	8000	0.27^*
Ref. [19]	3.0 GHz CPU, NVIDAI 7900 GTX	104.9^*	17.0
Ref. [32]	2.5 GHz CPU, NVIDAI GTX460	65.5	27.2^*
Ref. [20]	3.0 GHz CPU, ATI Radeon XL1800	33.6^*	52.7
Ref. [31]	2.66 GHz CPU, NVIDAI 8800 GTX	17.5	100.9
MSO(BM)	3.2 GHz CPU, NVIDAI GTX730	15.0	106.1

Table 3. Comparison of speed performance of proposed algorithm and other real-time algorithms

Cancan Zeng, Mingjun Ren, Gaobo Xiao, Yuehong Yin. Multi-Scale Stereo Matching Based on Bayesian Reasoning[J]. Acta Optica Sinica, 2017, 37(12): 1215004

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information