is the disparity obtained is sufficient for 3d recontruction?

1 次查看（过去 30 天）

neenu jose 2011-10-6

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/17588-is-the-disparity-obtained-is-sufficient-for-3d-recontruction

please help..

is my disparity obtained is correct?
is my progran sufficient for 3d reconstruction intended for machine vision?
how can i increase the accuracy?
how can i generate x,y,z coordinates and 3d reconstruction from this?

data = video.ImageDataTypeConverter;
conv= video.ColorSpaceConverter('Conversion','RGB to intensity');
lftsingle = step(data,imread('TSUL.png'))
lftgray= step(conv,lftsingle);
rgtsingle = step(data,imread('TSUR.png'));
rgtgray = step(conv,rgtsingle);
figure(1);
imshow(lftsingle), title('left image');
figure(2), clf;
imshow(rgtsingle), title('right image');
figure(3), clf;
imshow(rgtgray), title('gray Right image');
figure(4), clf;
imshow(lftgray), title('gray left image');
figure(5), clf;
imshow(cat(3,rgtgray,lftgray,lftgray)),axis image;
title('merged image');
 D = zeros(size(lftgray), 'single');
 d = zeros(size(lftgray), 'single');
 D2 = zeros(size(lftgray), 'single');
  d3 = zeros(size(lftgray), 'single');
  D4 = zeros(size(lftgray), 'single');
  d4 = zeros(size(lftgray), 'single');
  disparity = zeros(size(lftgray), 'single');
  Z= zeros(size(lftgray), 'single');
  N=size(lftgray,1)
  M=size(lftgray,2)
% finding disparity without shifting
for m=1:N
    for n=1:M
        d(m,n)=lftgray(m,n)-rgtgray(m,n);
        d(m,n)=abs(d(m,n));
    end
end
%shifting to right
for m=1:N
    for n=1:M
    D2(m,n)=lftgray(m,mod((n-1),N)+1);
    end
end
%finding disparity
for m=1:N
    for n=1:M
        d3(m,n)=D2(m,n)-rgtgray(m,n);
        d3(m,n)=abs(d3(m,n));
    end
end
%shifting to left
for m=1:N
    for n=1:M
    D4(m,n)=rgtgray(m,mod((n-1),N)+1);
    end
end
%finding disparity
for m=1:N
    for n=1:M
        d4(m,n)=D4(m,n)-lftgray(m,n);
        d4(m,n)=abs(d4(m,n));
    end
end
%finding minimum of disparity
for m=1:N
    for n=1:M
        disp(m,n)=min(min(d(m,n),d3(m,n)),d4(m,n));
    end
end
figure(6)
imshow(disp)
K = [409.4433 0 204.1225 ;0 416.0865 146.4133;0 0 1.0000];
%generating z coordinate Z=focal length*((1+stereobasline)/disparity))
for m=1:N
    for n=1:M
        Z(m,n)=(1+.567)/disp(m,n);
    end
end
%for m=1:N
  %  for n=1:M
    %    image=rgtgray(m,n,z(m,n));
   % end
%end
    %3dimage=image
    %figure(7)
    %a=meshgrid(5)
    %mesh(a)

15 个评论
显示 13更早的评论隐藏 13更早的评论

David Young 2011-10-18

Yes, there is a gradient-based family of algorithms. You have to divide the temporal gradient by the spatial gradient and adopt a way to address the aperture problem. It's not sensible to try to set out details here - the approach is in textbooks and Wikipedia (look up Horn-Schunck method for example), and there is code on the file exchange (e.g contribution 41349).