Tracking motion of particles: using edge detection and blob analysis?

Hi,

I would like to track particles in a stack of 500 images (451 x 191 x 500).

The raw images look like this:

I would like to detect, select a few different ROIs (for example, one of the very bright particles, and some of the darker 'dots' in this picture) and track their motion through the 500 frames.

I have tried using the canny edge detector, followed by a blob analysis to track it. Not too sure if this is the best way to go.

So far, I have this script:

data = load(data);

for i = 1:size(data, 3)
    BWStack(:, :, i) = edge(data(:, :, i), 'Canny');
end

% Select ROI interactively
figure;
imshow(BWStack(:, :, 1));
h = drawpolygon();
roiMask = createMask(h);

% Crop the ROI from each frame
ROIStack = false(size(BWStack));
for i = 1:size(BWStack, 3)
    ROIStack(:, :, i) = BWStack(:, :, i) & roiMask;
end

% Convert the ROIStack to uint8
ROIStack_uint8 = uint8(ROIStack) * 255;

% Motion tracking using foreground detection
foregroundDetector = vision.ForegroundDetector('NumGaussians', 5, 'NumTrainingFrames', 150);

% Initialize video player
videoPlayer = vision.VideoPlayer('Position', [100, 100, 500, 400]);

% Initialize a blob analysis system object
blobAnalysis = vision.BlobAnalysis('BoundingBoxOutputPort', true, ...
                                   'AreaOutputPort', true, ...
                                   'CentroidOutputPort', true, ...
                                   'MinimumBlobArea', 1);

pauseDuration = 0.01; % Adjust this value to control playback speed

% Initialize 
pointTracker = vision.PointTracker('MaxBidirectionalError', 2);
trajectories = cell(size(ROIStack_uint8, 3), 1);
trackerInitialized = false;
outputTrajectories = {};

for i = 1:size(ROIStack_uint8, 3)
    % Detect foreground
    foreground = foregroundDetector(ROIStack_uint8(:, :, i));

    % Perform blob analysis to find connected components
    [objArea, centroids, bbox] = step(blobAnalysis, foreground);

    % Ensure centroids have two columns
    if ~isempty(centroids) && size(centroids, 2) >= 2
        if ~trackerInitialized
            % Convert centroids to double
            centroids = double(centroids);
            
            % Initialize the point tracker with the centroids in the first frame
            initialize(pointTracker, centroids, ROIStack_uint8(:, :, i));
            trackerInitialized = true;
        else
            % Track the points
            [points, validity] = step(pointTracker, ROIStack_uint8(:, :, i));

            % Store valid points
            trajectories{i} = points(validity, :);

            % Append valid points to output trajectories
            if ~isempty(points(validity, :))
                outputTrajectories = [outputTrajectories; num2cell([repmat(i, sum(validity), 1), points(validity, :)])];
            end

            % Draw the points on the frame if there are valid points
            if ~isempty(points(validity, :))
                out = insertMarker(ROIStack_uint8(:, :, i), points(validity, :), '+', 'Color', 'red');
            else
                out = ROIStack_uint8(:, :, i);
            end
        end
    else
        % If no centroids are found, use the previous frame's points
        if i > 1 && ~isempty(trajectories{i-1})
            points = trajectories{i-1};
            trajectories{i} = points;

            % Append previous points to output trajectories
            outputTrajectories = [outputTrajectories; num2cell([repmat(i, size(points, 1), 1), points])];


            out = insertMarker(ROIStack_uint8(:, :, i), points, '+', 'Color', 'red');
        else
            out = ROIStack_uint8(:, :, i);
        end
    end

    % Draw a box around the detected objects
    Ishape = insertShape(out, 'Rectangle', bbox);
    
    % Display the frame
    step(videoPlayer, Ishape);

    % Pause to control playback speed
    pause(pauseDuration);
end

release(videoPlayer);
release(pointTracker);

% Display the trajectories
figure;
hold on;
for i = 2:size(trajectories, 1)
    if ~isempty(trajectories{i})
        plot(trajectories{i}(:, 1), trajectories{i}(:, 2), 'r.-');
    end
end
title('Particle Trajectories');
xlabel('X Position');
ylabel('Y Position');
hold off;

% Write trajectories to CSV
cell2csv('trajectories.csv', outputTrajectories, {'Frame', 'X', 'Y'});

% Function to write cell array to CSV
function cell2csv(fileName, cellArray, headers)
    fid = fopen(fileName, 'w');
    if ~isempty(headers)
        fprintf(fid, '%s,', headers{1:end-1});
        fprintf(fid, '%s\n', headers{end});
    end
    for row = 1:size(cellArray, 1)
        fprintf(fid, '%d,%f,%f\n', cellArray{row,:});
    end
    fclose(fid);
end

It seems to track many different points on the ROI but not continuously.

Example dataset 'data.mat' available here.

Any help and advice much appreciated!

Answer 1

[Image Analyst]

The Computer VIsion Toolbox has some tracking functionality, though I haven't used it.