Lucas-Kanade Method Example 2

This example uses Lucas-Kanade method on two images and calculate the optical flow vector for moving objects in the image.

Contents

Load images

clear all;
load traffic_frames

figure();
subplot 211
imshow(fr1);
im1 = im2double(rgb2gray(fr1));
subplot 212
imshow(fr2);
im2 = im2double(rgb2gray(fr2));

Find corners

Define the window size for Lucas-Kanade method

ww = 40;
w = round(ww/2);

% Reduce the size of the image
sc = 2;
im2c = imresize(im2, 1/sc);
C1 = corner(im2c);
C1 = C1*sc;

% Discard coners near the margin of the image
k = 1;
for i = 1:size(C1,1)
    x_i = C1(i, 2);
    y_i = C1(i, 1);
    if x_i-w>=1 && y_i-w>=1 && x_i+w<=size(im1,1)-1 && y_i+w<=size(im1,2)-1
      C(k,:) = C1(i,:);
      k = k+1;
    end
end
% Plot corners on the image
figure();
imshow(fr2);
hold on
plot(C(:,1), C(:,2), 'r*');

Implementing Lucas Kanade Method

for each point, calculate I_x, I_y, I_t

Ix_m = conv2(im1,[-1 1; -1 1], 'valid'); % partial on x
Iy_m = conv2(im1, [-1 -1; 1 1], 'valid'); % partial on y
It_m = conv2(im1, ones(2), 'valid') + conv2(im2, -ones(2), 'valid'); % partial on t
u = zeros(length(C),1);
v = zeros(length(C),1);

% within window ww * ww
for k = 1:length(C(:,2))
    i = C(k,2);
    j = C(k,1);
      Ix = Ix_m(i-w:i+w, j-w:j+w);
      Iy = Iy_m(i-w:i+w, j-w:j+w);
      It = It_m(i-w:i+w, j-w:j+w);

      Ix = Ix(:);
      Iy = Iy(:);
      b = -It(:); % get b here

      A = [Ix Iy]; % get A here
      nu = pinv(A)*b;

      u(k)=nu(1);
      v(k)=nu(2);
end;

Draw the optical flow vectors

figure();
imshow(fr2);
hold on;
quiver(C(:,1), C(:,2), u,v, 1,'r')

Published with MATLAB® R2014a