An efficient local stereo matching method based on an adaptive exponentially weighted moving average filter in SLIC space

Abstract

Rapidly obtaining accurate dense disparity maps has been the focus of stereo matching research. At present, approaches that achieve superior disparity maps require a large amount of computation, which is not suitable for practical applications. To address this issue, this paper proposes an efficient local matching method based on an adaptive exponentially weighted moving average filter and simple linear iterative clustering segmentation algorithm. First, an effective matching cost is introduced to adaptively integrate absolute intensity difference with Census transform, which is robust against texture free and luminance variate. Following this, during the cost aggregation, the exponentially weighted moving average filter and the SLIC segmentation are combined to handle the problems of computing consumption and adaptive expansion of the cost aggregation window. Finally, the dense disparity map is obtained by a winner-takes-all approach and disparity refinement. To demonstrate its efficiency and validity, the method is quantitatively tested and compared to existing approaches on the Middlebury benchmark. The results show that it has a non-occlusion accuracy of 90.66% and an average runtime of 7.01 s on the 2014 Middlebury dataset. Compared with existing competitive methods, the proposed method achieves superior matching results with a lower time cost. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Robust Cost Volume Generation Method for Dense Stereo Matching in Endoscopic Scenarios

“…So we raise the proportion of the cost volume on the finer scale in the inhomogeneous areas. We adopt the method in [ 36 ] to determine the homogeneity of each area whose basic ideas are to convolve the pixels in the image with a Gaussian kernel sign-flipped in the horizontal or vertical direction and to use the sum of squares of the two directional convolution results as the indicator of homogeneity. The horizontal sign-flipped Gaussian kernel is calculated by where x and y represent the pixel position in the convolution window.…”