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

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