transform Points Forward with cell

Hi, I need a help to apply a transformation to a cell array that contains the bboxPolygon values..here's the code:

%% Loop through video while ~isDone(videoReader) % get the next frame videoFrame = step(videoReader);

    % Use pointTracker to track feature points
    for i = 1:N
        [punti{i},validity] = step(tracker{i},videoFrame);

        % Keep only the valid points and discard the rest
        visiblePoints{i} = punti{i}(validity,:);
        oldInliers{i} = oldPoints{i}(validity,:);

            [xform, oldInliers{i}, visiblePoints{i}] = estimateGeometricTransform(...
                oldInliers{i}, visiblePoints{i}, 'similarity', 'MaxDistance', 4);
    end
           % Apply the transformation to the bounding box

for i = 1:P

    [bbcell{i}(1:2:end),bbcell{i}(2:2:end)]=transformPointsForward(xform, bbcell{i}(1:2:end)...
        ,bbcell{i}(2:2:end));

     end

    for i = 1:N
        videoFrame = insertShape(videoFrame, 'Polygon',bbcell{i});
        videoFrame = insertMarker(videoFrame, visiblePoints{i}, '+', ...
            'Color', 'white');

        %         % Save the new state of the point tracker
        %         oldPoints{i} = visiblePoints{i};
        %         setPoints(tracker{i}, oldPoints{i});

        % Display output
        step(videoPlayer, videoFrame);

    end
end

______________________________________

but when I start this code the shapes run away from the video clip or move themselves in wrong direction! Thanks a lot for helping me!

Answer 1

[Dima Lisin]

I would try capturing the output of transformPointsForward() into a temporary variable, and then assigning that to bbcell{i}.

On a different note, it looks like you are trying to track multiple objects. A better way to do that is to use a single vision.PointTracker, rather than multiple ones. The reason is that each of your point tracker objects will pre-process the frame exactly the same way. Using one point tracker will only pre-prosses the frame once, eliminating this wasted computation. However, that means you would need to concatenate all your points into a single Mx2 matrix, and you would need to maintain an Mx1 array of indices to keep track of which point belongs to which object.

Answer 2

[antonio]

Yes but I'm not able to do this with the code in matlab...can you help me? And I don't understand what you mean about to use a single vision.PointTracke. Thanks a lot for the helps!

[Dima Lisin]

Can you try this:

[x, y] = transformPointsForward(xform, bbcell{i}(1:2:end),...
     bbcell{i}(2:2:end));
bbcell{i}(1:2:end) = x;
bbcell{i}(2:2:end) = y;

As to your second question, you seem to have a cell array of point trackers, one for each object, correct? Instead, you should only use one point tracker for tracking all the points from all the objects.

Answer 3

[antonio]

I tried to use this solution but the rotation of bounding box is casual...I'm going crazy!! ahahah

Answer 4

[antonio]

I'm sending you my full code so you can find some errors...when I'm using one pointTracker, I get this:

"Index exceeds matrix dimensions.

Error in demo (line 77) oldInliers{i} = oldPoints{i}(isFound, :);" ......

[Dima Lisin]

Check out this example.