Output of number of people from a thermal image

How is it possible to get number of people as output from a theral image?I've attached a thermal image.

Answer 1

[Image Analyst]

Yes. First of all use the actual temperature image, not this pseudocolored RGB image. Then threshold at some temperature to get a binary image. Then call bwareafilt() to get rid of blobs that are too big or too small to be a person. Then call bwlabel() to count the number of blobs.

See my Image Segmentation Tutorial in my File Exchange:

Image Analyst's File Exchange

[Neha Tom Merin]

Can you suggest a program that can output number of people?

[Image Analyst]

@yanqi liu gave code below that does it on the RGB image. If you have the actual temperature image, then I suggest you use that instead. If all you have is the pseudocolored image then you might be using the cheapest FLIR camera where all it does is supply that. I recommend you spend just a little bit more and get the next better model which can give you the actual temperature image. Otherwise (less convenient), you can turn your pseudocolor image into a thermal image by extracting and examining the color bar and using that to change colors into the actual temperature it represents. I'm attaching a demo for that.

[Neha Tom Merin]

The code doesn't detect properly on any other images. I've attached another clear one.

[Image Analyst]

Did you try to adapt my code to reflect the location of the colorbar and image in your specific image? Obviously your colorbar is not in the same location so that will need to be adjusted.

[Neha Tom Merin]

Unfortunately, it's showing error.

[Image Analyst]

OK, so you didn't try to change the location. I've done it for you. See attached script.

Now that you have the temperature image you can threshold it and count the blobs.

mask = temperatureImage >= 27;
mask = bwareafilt(mask, [500, inf]); % Get rid of blobs smaller than 500 pixels.
props = regionprops(mask, temperatureImage, 'MeanIntensity');
allIntensities = [props.MeanIntensity] % Mean temperature of each blob.
numBlobs = length(props); 

[Neha Tom Merin]

What changes should be made if I dont have a colorbar for my image?

[Image Analyst]

Then you can't convert to temperature. Or you could just assume some colormap. Otherwise you'll have to just work on the pseudocolor image (like yanqi did), but conversion from RGB to gray scale is not linear at all, so that method is not reliable. For example, the gray scale of one color may be the same as that of a completely different color. You'd be best off trying to construct some kind of colormap to convert RGB into temperature.

Answer 2

[yanqi liu]

im = imread('https://ww2.mathworks.cn/matlabcentral/answers/uploaded_files/938099/rgb.png');
jm = rgb2lab(im);
bw = im2bw(mat2gray(jm(:,:,2)), 0.8);
bw = bwareaopen(bw, 500);
stats = regionprops(bw);
num = length(stats);
figure; imshow(im, []);
for i = 1 : num
    hold on; rectangle('position', stats(i).BoundingBox, 'EdgeColor', 'g', 'LineWidth', 2)
end
title(sprintf('共%d个候选目标', num));

[Neha Tom Merin]

I tried for other images. But, it doesn't come out right. Attaching another one here.

[yanqi liu]

yes，sir，it is one simple method for target image，so if we change image，may be should change some process step，such as

im = imread('https://ww2.mathworks.cn/matlabcentral/answers/uploaded_files/938954/gui.jpeg');
jm = rgb2lab(im);
bw = im2bw(mat2gray(jm(:,:,3)), 0.5);
bw = bwareaopen(imclearborder(bw), 1000);
stats = regionprops(bw);
num = length(stats);
figure; imshow(im, []);
for i = 1 : num
    hold on; rectangle('position', stats(i).BoundingBox, 'EdgeColor', 'g', 'LineWidth', 2)
end
title(sprintf('共%d个候选目标', num));