Proposal

Proposal Develop an algorithm to calculate the Dense correspondence between the pair of images.Estimate the disparity map.

Problem defined: Proposal

• Develop an algorithm to calculate the Dense correspondence between the pair of images
• Estimate the disparity map.

Standard Steps at hands:

• Camera Calibration
  • The assumption in our course work is that the camera position will be provided.
• Image rectification
  • The assumption in our course work is that rectified images are provided.
• Dense correspondence matching
  • Euclidean Distance (ED)
  • Or Normalized Correction (NC)
  • Or Semi Global Block Matching algorithm
• Depth Image generation

Algorithms used in this coursework described as below: Proposal

Image Rectification

The images for later process have to be rectified, in other words, the corresponding points from the two images are located on the same rows. In our coursework, the images provided are already taken by moving the camera horizontally, therefore assuring the images been rectified already. But just in case, we can still implement this algorithm in our course work to guarantee the result is decent.Proposal

Dense Corresponding Matching:

In Dense correspondence matching, dense correspondence is established at the pixel level in the two images. We first calculate the SIFT images by calculating a SIFT descriptor for each pixel. If scale maps are enabled, then the scale for the SIFT descriptor will be taken from the image’s scale map. Then we will make a 2-sided SIFT-flow for the selected image pairs. Apply geometric filtering. Using RANSAC on the matches of each image pair to find the corresponding Fundamental Matrix.

Disparity Map: Proposal代写

To calculate the disparity map between two stereo images, we first load the images and convert them into greyscale. Then, we can use stereo anaglyph, use red-scan stereo glasses to view images in 3D. Then compute the disparity map using the disparity function I found in MATLAB. The algorithm for calculating the disparity map chosen is ‘Semi-Global Block Matching’. The algorithm would compute the disparity by comparing the sum of absolute difference of each block of pixels in the images. Then, it would force the similar disparity on the neighboring blocks. As compared to another algorithm called BlockMatching, this algorithm will return a more complete result in disparity estimate. As for a better illustration of the disparity map result, we can display disparity range for the ‘imshow’ in MATLAB.Proposal

Proposal案例