Semi-global Matching for Stereo Processing

• Overview

Overview

Typical steps of stereo methods:

• Matching cost computation: compute pixelwise matching cost between pixels of subjective image and pixels of reference image
• Cost aggreation: often sum over costs over a fixed sized window with certain weighting strategy (like computing correlation)
• Disparity computation/optimization: select the disparity by winner takes all or global graph-based strategy; even consider occlusions, visibility and left/right or symmetric consistency
• Disparity refinement: refinement steps like removing peaks, checking consistency and subpixel interpolation