How to properly initialize Histogram-based tracker?

Question

Kang Hao Tan 2019-3-1

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/447622-how-to-properly-initialize-histogram-based-tracker

编辑： Kang Hao Tan 2019-3-1

Hi all, I am trying to track flying drones for a school project.

My strategy is to use the detectSURF features function to detect a SURF point on the object and then use this point to construct an search box around the drone to initialize the Histogram-based tracker in the computer vision toolbox.

I am able to detect the drone SURF feature but the Histogram tracker is not behaving as intended as the red bounding box keeps expanding towards all edges of the frame. The green box below shows the placement of the search box. The next picture with red lines at the frame edges shows the problem I am facing - the Histogram tracker is simply not tracking the target. I have followed the face-tracking example for HistogramBasedTacker in computer vision toolbox, given in the MATLAB documentation, but my code doesn't work and I can't figure out what's wrong.

I am posting my code to seek advice on how I can overcome this problem. Appreciate your help pls. There are mainly 2 FOR loops in the code.

The first FOR loop reads the file to detect SURF points. The frame is then processed into HSV color space to segment by hue values. Once detected, the Histogram tracker is initialized with a region of interest (i.e the search box) and the algorithm breaks out of the first FOR loop.

The second FOR loop mainly runs the Histogram tracker until end of video. (At least this was my intention, but the code isn't working )

%test of image processing
clear;
clc;
%read inputs and test play the video
dronevid = VideoReader('scott 2.mp4');
get(dronevid);
nframes = dronevid.NumberOfFrames;
%videoFileReader = vision.VideoFileReader('scott 2.mp4');
%videoPlayer = vision.VideoPlayer();
shapeInserter = vision.ShapeInserter('BorderColor','Custom', ...
    'CustomBorderColor',[1 0 0]);
init=1;
k=1;
for i=0:nframes
    if i==0
        continue
    end    
    
    frame=read(dronevid,i);
    image(frame);
    hold on;
    %img_tmp = double(frame); %load in the image and convert to double to allow for computations on the image
    img = frame(:,:,1);
    
    points = detectSURFFeatures(img);  
    y=points.selectStrongest(1);
    plot(points.selectStrongest(1));
    k=i+1;
    
    if (~isempty(points) & i > 80 )     %break out of FOR loop when SURF features is detected
        %initialize tracker 
        hsv = rgb2hsv(frame);
        hframe=hsv(:,:,1);
        sframe=hsv(:,:,2);
        initx=round(y.Location(1));
        inity=round(y.Location(2));
        
        %objectRegion=round(getPosition(imrect));  
        objectRegion = [initx-10, inity-30, 65, 46];
        rectangle('Position',objectRegion,'EdgeColor','g','LineWidth',2);
        pause(3);
        
        %objectImage = shapeInserter(frame, objectRegion);
        %imshow(objectImage);
        tracker = vision.HistogramBasedTracker;
        initializeObject(tracker, hframe , objectRegion);
        
        %bbox = tracker(frame(:,:,1));   % draw first bounding box
        %rectangle('Position',[bbox(1),bbox(2),bbox(3),bbox(4)],'EdgeColor','y','LineWidth',2);
        
        pause(3)
        init=0;  % tracker initialization is done    
        break
    end
    drawnow;
    
end
hold off;     
 
for j=i:nframes 
    frame1=read(dronevid,j);
    image(frame1);
    hold on;
    hsv = rgb2hsv(frame1);
    hframe=hsv(:,:,1);
    sframe=hsv(:,:,2);
    bb = tracker(hframe);
    rectangle('Position',[bb(1),bb(2),bb(3),bb(4)],'EdgeColor','r','LineWidth',2);
    pause(0.1);
    drawnow;  
end
   