Code to count number of heads(people) present in a image

I want a matlab program which will count number of heads present in image and not other things present such as trees,other objects etc.please send me entire code.

[Walter Roberson]

Who will get credit for handing in the assignment ?

[MRahul]

create a folder, put the image in jpg format on which u want to run this code, save the program in the same folder. Run

instead of 15.jpg put your image file name.

%get the input image
I=imread('15.jpg');
imshow(I),title('Image:1');
%change the color space
cform = makecform('srgb2lab');
J=applycform(I,cform);
figure;imshow(J),title('Image:2');
%equalise brightness to get skin area
K=J(:,:,2);% 2nd page of 3-d vector j
figure;imshow(K),title('Image:3');
L=graythresh(J(:,:,2));% find appropriate gray thresh value
BW1=im2bw(J(:,:,2),L);% convert to binary image based on threshold
figure;imshow(BW1),title('Image:4');
bw2=imfill(BW1,'holes');% fill patches with holes
figure;imshow(bw2)
bw3 = bwareaopen(bw2,1890); %opens area greater than 1890
cc=bwconncomp(bw3)% connected comp for finding the density of people in
image
density=cc.NumObjects / (size(bw3,1) * size(bw3,2))
figure;imshow(bw3)
labeledImage = bwlabel(bw3, 8);%same as connected components
figure;imshow(labeledImage)
blobMeasurements = regionprops(labeledImage,'all');%measure all properties of the image
numberOfPeople = size(blobMeasurements, 1)% count the number of people
% draw bounding boxes
imagesc(I);
hold on;
title('Original with bounding boxes');
for k = 1 : numberOfPeople % Loop through all blobs.
% Find the mean of each blob.
% directly into regionprops.
thisBlobsBox = blobMeasurements(k).BoundingBox; 
% Get list of pixels in current blob.
x1 = thisBlobsBox(1);%1st side
y1 = thisBlobsBox(2);%2nd side
x2 = x1 + thisBlobsBox(3);%3rd side
y2 = y1 + thisBlobsBox(4);%4th side
x = [x1 x2 x2 x1 x1];
y = [y1 y1 y2 y2 y1];
%subplot(3,4,2);
plot(x, y, 'LineWidth', 2);
end

[Walter Roberson]

As I glance at that code, I do not see how it handles dark-skinned people.

Answer 1

[Image Analyst]

It should be pretty easy if you have the Computer Vision System Toolbox, as this demo from it shows counted heads: http://www.mathworks.com/products/computer-vision/description4.html

[javed patel]

thank you

Answer 2

[Image Analyst]

See crowd estimation articles here: http://www.visionbib.com/bibliography/motion-f735m1.html#Crowds,%20Tracking%20Multiple%20People,%20Multiple%20Pedestrian%20Tracking

[Iqra]

I need a Matlab code which can count only head in image, This code is not working properly,any demos will also be helpful. Thankyou

[Image Analyst]

Then start calling people who wrote the papers I gave the link to and see if they'll give or sell you their code. Or else start writing, or start looking for someone to hire to write one of the papers for you.

[babar ali]

I need a Matlab code which can count only head in image, This code is not working properly,any demos will also be helpful. Thankyou

[Image Analyst]

What code? Perhaps your image is not what the code was expecting. Did you see the link on crowd estimation that I gave above? If your code doesn't work, try one of the published algorithms.

[babar ali]

yes this code is not working correctly can you suggest me any algorithm for crowd estimation Thanks

[babar ali]

 i am expecting the code for calculating number of people in an image

[Image Analyst]

Again, WHAT "code is not working correctly"? You haven't posted your code or your images or even read this link yet.

And I DID suggest algorithms - Lots of them. Evidently you chose one and coded it up but you haven't said which one or even shared the code, so what can anyone do to help you????

I don't even know how many heads you have in your image. Three or four moving heads is a lot easier and more accurate than tens of thousands of still heads, like from an aerial view.

Keep in mind that this is no 5 minute project and we are not going to develop a turnkey system for you. You're either going to have to write the thing yourself, or hire someone to write it for you.

Answer 3

[Gogul]

%get the input image

I=imread('15.jpg');

imshow(I),title('Image:1');

%change the color space

cform = makecform('srgb2lab');

J=applycform(I,cform);

figure;imshow(J),title('Image:2');

%equalise brightness to get skin area

K=J(:,:,2);% 2nd page of 3-d vector j

figure;imshow(K),title('Image:3');

L=graythresh(J(:,:,2));% find appropriate gray thresh value

BW1=im2bw(J(:,:,2),L);% convert to binary image based on threshold

figure;imshow(BW1),title('Image:4');

bw2=imfill(BW1,'holes');% fill patches with holes

figure;imshow(bw2)

bw3 = bwareaopen(bw2,1890); %opens area greater than 1890

cc=bwconncomp(bw3)% connected comp for finding the density of people in

image

density=cc.NumObjects / (size(bw3,1) * size(bw3,2))

figure;imshow(bw3)

labeledImage = bwlabel(bw3, 8);%same as connected components

figure;imshow(labeledImage)

blobMeasurements = regionprops(labeledImage,'all');%measure all properties of the image

numberOfPeople = size(blobMeasurements, 1)% count the number of people

% draw bounding boxes

imagesc(I);

hold on;

title('Original with bounding boxes');

for k = 1 : numberOfPeople % Loop through all blobs.

% Find the mean of each blob.

% directly into regionprops.

thisBlobsBox = blobMeasurements(k).BoundingBox;

% Get list of pixels in current blob.

x1 = thisBlobsBox(1);%1st side

y1 = thisBlobsBox(2);%2nd side

x2 = x1 + thisBlobsBox(3);%3rd side

y2 = y1 + thisBlobsBox(4);%4th side

x = [x1 x2 x2 x1 x1];

y = [y1 y1 y2 y2 y1];

%subplot(3,4,2);

plot(x, y, 'LineWidth', 2);

end

[Image Analyst]

There is a function rgb2lab that makes the colorspace conversion easier than using makecform and applycform.

You say you want to "equalise brightness to get skin area" -- this is not necessary, plus you didn't even do it. You created K as the "A" channel image of the LAB image, and showed it as "Image3" but you never use it.

The code is intended for use in images where only the face is visible, with no overhead views and no arms, hands, or other reddish regions in the image (since you threshold on the A channel to get red things).

You forgot to attach '15.jpg' or even a screenshot to show how well your algorithm works. So let's see how well it works with a standard demo image:

% Get the input image - standard demo image that ships with MATLAB.
I=imread('visionteam.jpg');
imshow(I),title('Image:1');

%change the color space
cform = makecform('srgb2lab');
J=applycform(I,cform);
figure;imshow(J),title('Image:2 LAB image');

%equalise brightness to get skin area
K=J(:,:,2);% 2nd page of 3-d vector j
figure;imshow(K, []),title('Image:3 The A channel only');

L=graythresh(J(:,:,2));% find appropriate gray thresh value
BW1=im2bw(J(:,:,2),L);% convert to binary image based on threshold
figure;imshow(BW1),title('Image:4 thresholded A channel');

bw2=imfill(BW1,'holes');% fill patches with holes
figure;imshow(bw2)

% Change by Image Analyst to figure out minimum allowable size of red region.
% Existing value of 1890 did not work for this particular image.
% Get areas of blobs
props = regionprops(bw2, 'Area');
allAreas = sort([props.Area], 'descend')

allAreas = 1x78

3362 1830 1585 1321 1271 1162 661 376 299 292 282 279 264 244 242 199 123 119 89 87 80 78 71 59 56 55 43 36 36 35

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

% Let's use a minimum area of 1000 for this particular photo.
minAllowableArea = 1000;
bw3 = bwareaopen(bw2,minAllowableArea); %opens area greater than 1890
% End of change by Image Analyst to figure out minimum allowable size of red region.

cc=bwconncomp(bw3)% connected comp for finding the density of people in image

cc = struct with fields:
    Connectivity: 8
       ImageSize: [413 800]
      NumObjects: 6
    PixelIdxList: {[3362x1 double]  [1585x1 double]  [1162x1 double]  [1321x1 double]  [1271x1 double]  [1830x1 double]}

density=cc.NumObjects / (size(bw3,1) * size(bw3,2))

density = 1.8160e-05

figure;imshow(bw3)

labeledImage = bwlabel(bw3, 8);%same as connected components
figure;imshow(labeledImage, [])

blobMeasurements = regionprops(labeledImage,'all');%measure all properties of the image
numberOfPeople = size(blobMeasurements, 1)% count the number of people

numberOfPeople = 6

% draw bounding boxes
imagesc(I);
hold on;
title('Original with bounding boxes');
for k = 1 : numberOfPeople % Loop through all blobs.
    % Find the mean of each blob.
    % directly into regionprops.
    thisBlobsBox = blobMeasurements(k).BoundingBox;
    % Get list of pixels in current blob.
    x1 = thisBlobsBox(1);%1st side
    y1 = thisBlobsBox(2);%2nd side
    x2 = x1 + thisBlobsBox(3);%3rd side
    y2 = y1 + thisBlobsBox(4);%4th side
    x = [x1 x2 x2 x1 x1];
    y = [y1 y1 y2 y2 y1];
    %subplot(3,4,2);
    plot(x, y, 'LineWidth', 2);
end

It seemed to miss one person and count red sleeves of another person. A fair start but you might want to make it more robust by putting an upper threshold on your segmentation of the A channel to eliminate abnormally red regions and figure out how to handle beards and mustaches to capture the missing person without capturing hands.