Adaptive Support-Weight Stereo-Matching Approach with Two Disparity Refinement Steps

ABSTRACT

Machine vision is used to reconstruct three-dimensional information from images. To perceive depth information of target objects, many techniques in machine vision such as camera calibration, epipolar lines rectification, stereo matching, etc. are considered. In the stereo-matching field which is regarded as a critical step, local and global stereo-matching methods are the two key methods to get the disparity image which is used to calculate the depth information. A local method – Adaptive Support-Weight (ASW) algorithm which has high efficiency and simple structure can obtain comparable accuracy of results with the global method. But in the final disparity image obtained by ASW, there are still some significant mismatching areas. In this paper, ASW and two disparity refinement steps are proposed to refine the accuracy of the disparity image. The mismatching areas in the disparity image are corrected by the two disparity refinement steps which mainly include the variable-cross region, disparity inheritance, and fixed window voting methods. Then, the performance of the proposed algorithm under different parameters is analyzed based on the Middlebury benchmark. Finally, the experimental results show that the accuracy of the disparity image is improved after the disparity refinement steps and the final refined result is comparable to some stereo-matching algorithms.

KEYWORDS:

• Adaptive-support weight
• Binocular vision
• Disparity refinement
• Local stereo-matching
• Variable-cross region

ACKNOWLEDGMENTS

This research is supported by the National Natural Science Foundation of China (Grant Nos. 51475343 and 51675389), the International Science & Technology Cooperation Program, Hubei Technological Innovation Special Fund (Grant No. 2016AHB005), and Engineering and Physical Sciences Research Council (EPSRC), UK (Grant No. EP/N018524/1).

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Additional information

Funding

National Natural Science Foundation of China [grant number 51475343], [grant number 51675389]; International Science & Technology Cooperation Program, Hubei Technological Innovation Special Fund [grant number 2016AHB005]; Engineering and Physical Sciences Research Council [grant number EP/N018524/1].

Notes on contributors

Jiayi Liu

Jiayi Liu is a PhD student at the School of Information Engineering, Wuhan University of Technology. He received his Bachelor's degree in 2012 from Wuhan University of Technology. His research interests include machine vision, intelligent optimization, etc.

Zude Zhou

Zude Zhou is a professor at the School of Information Engineering, Wuhan University of Technology. His major research interests are digital manufacturing, intelligent manufacturing, etc.

E-mail: [email protected]

Wenjun Xu

Wenjun Xu is a professor at the School of Information Engineering, Wuhan University of Technology. His research interests include, sustainable manufacturing, energy-efficient manufacturing, manufacturing service, and manufacturing intelligence, etc.

E-mail: [email protected]

Jiwei Hu

Jiwei Hu is an associate professor at the School of Information Engineering, Wuhan University of Technology. His major researches include machine vision, image processing, etc.

E-mail: [email protected]