Video Labeler App and exporting the data

Question

Martine Banville 2020-6-16

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/549543-video-labeler-app-and-exporting-the-data

编辑： Martine Banville 2020-6-19

Hi there,

I have been using the App Video Labeler to track objects. So I created ROI for various features but I am having a hard time exporting this data so that I can use it outside the App. Attached is an example of a short video where I have tracked 2 things (using rectangle ROI) and I also created an horizontal line (using a ROI line) but I now want to export that data to further calculate the trajectory, maybe the angle of the line with respect to the bottom of the photo, etc... but once I export the groundTruth data using the Export Labels + to File, I don't understand how I can use the groundTruth exported labels to use elsewhere. Can I get the x,y coordinate for each object for each frame? If so how do I access that? Can I get the line equation?

Anyway I have been searching the help but I find it very basic and can not make sense of all this. Does anyone have concret examples to share? Does anyone have any useful tutorial? I also tried with pixel labeler using Video Labeler and I was able to generate a movie of the pixel label but unable to extract separately each object from the movie.

Attached is an example of a movie with the exported data from the Video Labeler App. At this point anything would help! Thank you.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Harsha Priya Daggubati 2020-6-19

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/549543-video-labeler-app-and-exporting-the-data#answer_454093

在 MATLAB Online 中打开

Hi,

I can answer a part of your question, to my understanding you can get the 'position' of the object labelled by bounding boxes from 'gTruth' in MATLAB Workspace as follows:

You might know, that ground truth exported to MATLAB Workspace from Video Labeler App, is a structure with 3 fields: DataSource, Label Definitions, Label Data. You can get the position of every bounding box using Label Data field. You can work as follows:

labelData = gTruth.labelData;
head(labelData)
gTruthInterval(1,:).Car{1}.Position % [x y width height], in pixels

The code above helps you get x, y coordinates of an object.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Martine Banville 2020-6-19

编辑：Martine Banville 2020-6-19

在 MATLAB Online 中打开

Thank you I was finally able to get the gTruth data that I required using the following commands:

gTruth = selectLabels(gTruth,'Roof');  
% This creates another gTruth but with the isolated variable 'Roof' in this case.

or

for i=1:NumberofFrames
    C = gTruth.LabelData.Roof{i,1};
    ...
end
% This allow me to get access to the data from Label 'Roof' in this case for each frame.

However now I am having difficulty reading the pixel label data. I am using this code to retrieve all the label for each frame and later generating a video with only the label data:

Img = zeros(VideoHeight,VideoWidth,3,NumberofFrames); % Create a matrix of Black Frames 
labelIDs = [1,2]; 
labelcats = ["Aircraft" "FOD"];
for i=1:NumberofFrames %Looping though each Pixel Label Frame
    formatSpec = 'Label_%d.png';
    str = sprintf(formatSpec,i);
    LabelImageFile = fullfile(LabelDir,str);
    
    % Here since the Pixel Label Data might not be available for a particular 
    % frame due to the fact that some frame have no data (nothing was tagged in
    % certain frame) I decided to generate a black image when this occurs.
    if exist(LabelImageFile,'File') == 0 % If Label file does not exist generate a black image
        Img(:,:,1,i) = zeros(VideoHeight,VideoWidth); 
        Img(:,:,2,i) = zeros(VideoHeight,VideoWidth);  
        Img(:,:,3,i) = zeros(VideoHeight,VideoWidth); 
    else
        Label = imread(LabelImageFile); % If Label exist use the generated Label 
        LabelCats = categorical(Label,labelIDs,labelcats);  % Construct a categorical matrix 
                                                            % using the image and the definitions.
        Img(:,:,:,i) = labeloverlay(Img(:,:,:,i),LabelCats);    % This overlays all the pixel data 
                                                                % onto the one frame   
    end
end

So the above code works ok. It is probably not very efficient but this is the way I came up with for now. So my problem is that I don't have a good understanding of how to separate each label. What if I don't want to overlay all the label in one frame. Now I am using the 'labeloverlay' function but if I would want to only select one label how can I do so?

Is there a command to change the color of a particular label to a given color i.e. pure Red or pure Blue? I was unable to change the colors in the Video Label app.

If I could isolate one particular label per frame I could then turn it into a bw image and do some use some morphological processing to later use the function 'regionprops' to get some statistics of what I am tracking.

Thank you

请先登录，再进行评论。