Stereo Correspondence Estimation by Two Dimensional Real Time Spiral Search Algorithm
Md. Abdul Mannan Mondal1, Mohammad Haider Ali2

1Md. Abdul Mannan Mondal*, Department of Computer Science and Engineering, University of Dhaka, Dhaka, Bangladesh. 
2Mohammad Haider Ali, Department of Computer Science and Engineering, University of Dhaka, Dhaka, Bangladesh.
Manuscript received on April 13, 2020. | Revised Manuscript received on April 20, 2020. | Manuscript published on June 30, 2020. | PP: 96-103 | Volume-9 Issue-5, June 2020. | Retrieval Number: D8592049420/2020©BEIESP | DOI: 10.35940/ijeat.D8592.069520
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: This paper presents a new searching algorithm titled “Two Dimensional Real Time Spiral Search Algorithm (2DRTSSA)” to compute the stereo correspondence or dense disparity map of two rectified images. The proposed algorithm can estimate the minimum stereo correspondence or disparity among all the window costs of a fixed axis from minimum to maximum range of that axis. It can also simultaneously calculate the dense disparity of another axis with the same range of axis. So the proposed method calculates stereo correspondence two dimensionally at a time and thus it increases the speed and accuracy over the existing state-of-the-arts methods of one dimensional and left-right searching strategy. The 2DRTSSA method calculates firstly the two window costs; one is along with the +x direction and another is along with –y direction .The minimum disparity of estimated two window costs and their distance parameters are remaining contribute in final selection. The rest of two window costs of –x direction and +y direction are also calculated using the same procedure. The minimum disparity of newly estimated two window costs and their distance are remaining contribute in final selection. The process is then repeated for the successive pixels of reference image along with the 2D scan lines from left to right of the whole image. The 2DRTSSA method is able to optimize the speed and accuracy of estimated dense disparity. Experimental results are compared in Section-IV (A), Section-IV (B) and Section-IV(C) with the current state-of-the-arts methods those are tested on Middlebury Standard stereo data set. The proposed 2DRTSSA method establishes the highest speed and accuracy with properly reconstructed 3D of dense disparity image.
Keywords: Stereo correspondence, window cost, spiral search, disparity, sum of square differences, normalized correlatio technique.