Finding Transformation for Image registration with stereo Images

Hello I am running tests right now trying to take a pair of stereo images and compute a transformation matrix that transforms one of the images directly on top of the other image. My problem is when I run my code below the matrix that it spits out when applied to one of the images sends back a black image of all 0's. I need to be able to use the canny edge detection match points and then create a transformation from those matched points. Please I need help and suggestions, look below to see what i have tried. function [f] = imginterp( img1, img2 )

tic %%%%% Canny Edge Detection

color1=img1; %Save color image

gray1= im2double(rgb2gray(img1)); % make image a grayscale image Canny1=vision.EdgeDetector('Method','Canny','GaussianFilterStandardDeviation',2,... 'NonEdgePixelsPercentage',94); % create the canny matrix

binaryimg1= step(Canny1,gray1);% apply canny matrix to image

[x1 y1]=find(binaryimg1==1); % find coordinates of 1 values

img1coor(:,1)=x1; %store x values in variable first column

img1coor(:,2)=y1; %store y values in variable second column [features1, valid_points1] = extractFeatures(gray1, img1coor);% Stores features I am going to match

%Repeat for second image

color2=img2; %Save color image

gray2= im2double(rgb2gray(img2)); % make image a grayscale image Canny2=vision.EdgeDetector('Method','Canny','GaussianFilterStandardDeviation',2,... 'NonEdgePixelsPercentage',94); % create the canny matrix binaryimg2= step(Canny2,gray2);% apply canny matrix to image

[x2 y2]=find(binaryimg2==1);% find coordinates of 1 values

img2coor(:,1)=x2; %store x values in variable first column

img2coor(:,2)=y2; %store y values in variable second column

[features2, valid_points2] = extractFeatures(gray2, img2coor);% Stores features I am going to match

index_pairs = matchFeatures(features1, features2); % Indexed Matched features

matched_points1 = valid_points1(index_pairs(:, 1), :); % Select the features that were matched

matched_points2 = valid_points2(index_pairs(:, 2), :);% Select the features that were matched

[F Inliers_Index] = estimateFundamentalMatrix(matched_points1,matched_points2, 'Method','RANSAC','NumTrials',1000,... 'DistanceType', 'Sampson','DistanceThreshold', .00001); % Use Ransac to get rid of outliers and then estimate a transformation % between left over points.

end

Answer 1

[Matt J]

Did you examine the intermediate quantities produced by the code? Does the edge map look reasonable? Do the matched_points look reasonable?

Use the debugger to step through the code and examine these things.

Answer 2

[Ryan]

Yea I have tried that, I mean the edge map does look reasonable, There are still outliers here and there. I can't seem to get rid of all the outliers though, no matter how I manipulate the estimateFundamentalMatrix Function. So i guess I take back my first statement. I just looked over everything again and I have 600+ matches and about 100+ are still outliers even after manipulating the estimateFundamentalMatrix function to the full extent to get rid of outliers. I've tried all the methods such as Ransac and LTS along with w.e other methods are available.