Stereo

Abstract

Our goal is to compute a dense disparity map of stereo pairs by using Fundamental matrix. The method can reduce the search space and search time because it reduces search on plane to search on line.

Algorithm Design

• Find correspondences:
  extract the Speed-Up Robust Features(SURF) and match them.
• Estimate Fundamental Matrix:
  use RANSAC to eliminate outliers, and estimate the Fundamental Matrix by using inlier correspondences.
• Search epipolar line:
  for each point in image2, I compute the corresponding epipolar line in image 1, and use NCC to find the most fitted point.
• Compute a dense disparity map:
  compute the horizontal and vertical disparity respectively, and use following equations to scale the value: 255*(^val-min⁄_max-min)

Analysis

Since the two picture only moves in horizontal direction. The vertical disparity map is all black, which means no motion.