Color Tracking in MATLAB - RGB

Hi,

I wrote code to track colors in MATLAB. The code plots three circles (Red, Blue, and Green) that follow the motion of colors moving on in front of the web cam (R, B, G). Red and blue work perfectly, but green does not. Any ideas? Here is my code after I set up the video acquisition:

 % Get snapshot of current frame
    snap = getsnapshot(vid);

    % Extract red by subtracting red from grayscale
    snap_red = imsubtract(snap(:,:,1), rgb2gray(snap));
    snap_blue = imsubtract(snap(:,:,3), rgb2gray(snap));
    snap_green = imsubtract(snap(:,:,2), rgb2gray(snap));

    % Filter out noise
    snap_red = medfilt2(snap_red, [3 3]);
    snap_blue = medfilt2(snap_blue, [3 3]);
    snap_green = medfilt2(snap_green, [3 3]);

    % Convert new snapshot to binary
    snap_red = im2bw(snap_red,0.18);
    snap_blue = im2bw(snap_blue,0.18);
    snap_green = im2bw(snap_green,0.18);

    % Remove pixles less than 300px
    snap_red = bwlabel(snap_red, 8);
    snap_blue = bwlabel(snap_blue, 8);
    snap_green = bwlabel(snap_green, 8);

    % Label all connected components
    bw_red = bwlabel(snap_red,8);
    bw_blue = bwlabel(snap_blue,8);
    bw_green = bwlabel(snap_green,8);

    % Properties for each labeled region
    stats_red = regionprops(bw_red, 'BoundingBox','Centroid');
    stats_blue = regionprops(bw_blue, 'BoundingBox','Centroid');
    stats_green = regionprops(bw_green, 'BoundingBox','Centroid');

    % Display new image
    drawnow

    % Puts red objects in rectangular box
    for object_red = 1:length(stats_red)
        bb_red = stats_red(object_red).BoundingBox;
        bc_red = stats_red(object_red).Centroid;
        plot(bc_red(1),-bc_red(2),'or')
        axis([0 300 -300 0])
    end
    hold on
    % Puts blue objects in rectangular box
    for object_blue = 1:length(stats_blue)
        bb_blue = stats_blue(object_blue).BoundingBox;
        bc_blue = stats_blue(object_blue).Centroid;
        plot(bc_blue(1),-bc_blue(2),'ob')
        axis([0 300 -300 0])
    end
    % Puts blue objects in rectangular box
    for object_green = 1:length(stats_green)
        bb_green = stats_green(object_green).BoundingBox;
        bc_green = stats_green(object_green).Centroid;
        plot(bc_green(1),-bc_green(2),'og')
        axis([0 300 -300 0])
    end
    hold off

[Vikash Varma]

can anyone explain why we have to remove all those pixels less than 300 from a binary image which contains only either 0 or 1 ?

[Image Analyst]

Do you actually mean "remove all those blobs less than 300 pixels"? If so, do this

binaryImage = bwareaopen(binaryImage, 300);

Answer 1

[Image Analyst]

Subtracting from the grayscale version of the image is a bad way to find the green, or any color for that matter. You need to look at the difference between the color you want and the average of the other two. For example this might give a better green signal:

greenDifference = abs(greenChannel - (redChannel + blueChannel)/2);

Then threshold on that. Cast them all to double first so you don't get any clipping.

[Sivakumaran Chandrasekaran]

can you explain that why we should take average of two channels here

[Image Analyst]

If you want to know the "greenness" of an image you want to know how much greener each pixel is than something. Well, what should that something be? Maybe you want to know how much more the green intensity is than the red intensity? Well that sounds reasonable. Maybe you want to know how much more the green intensity is than the blue intensity? Well that also sounds reasonable. Which one should you pick? How about seeing how much more intensity it has in the green than the average of the red and blue? Maybe that would be better than just picking one or the other.

[Marco]

@Image Analyst: where could i find some reference about your solution ?

[Image Analyst]

You can use me as a reference. This kind of analysis has been going on for decades and it's kind of ad hoc and intuitive so I don't know where to find textbooks or papers on it.

[KDN]

How can I detect white color and and it's brightness.

[Image Analyst]

See my demo http://www.mathworks.com/matlabcentral/fileexchange/28512-simple-color-detection-by-hue It would be something like this

hsvImage = rgb2hsv(double(rgbImage));
h = hsvImage(:,:,1);
s = hsvImage(:,:,2);
v = hsvImage(:,:,3);
whitePixels = s < 0.2 & v > 0.8;

or something like that.

whitePixels is a binary image, or "mask". Feel free to modify the parameters.

Answer 2

[Andrea]

Hello, to detect green should change a parameter in this condition:

%%snap_green = im2bw(snap_green,0.18);%%

0.18 is to detect objects with a light green colored and sometimes do not want to detect. I advise you to try with a strong green and put it to 0.09 parameter. I hope you still serve. Andrea

Answer 3

[daher]

Hi, i have a question.. assuming that we have to track the first identified blue object that appears in a video (O1) Next, another object (O2) with the same color (using HSV presentation) appears

how to distinguish between them and to keep tracking O1 no matter what will be identified later in the video without referring to geometry features of each object.. Any suggestions?