How to perform object detection correctly?

Question

Manisha Jaiswal 2018-4-18

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/395874-how-to-perform-object-detection-correctly

编辑： Manisha Jaiswal 2018-4-18

I have a code in which background subtraction is used to detect the number of cars in parking lot. But, the output is not desirable i.e. it does not detect the number of cars correctly. N = 1; % num of frames to use to make averaged background tps=10; status = [0 0 0 0 0 0 0 0 0 0]; img = zeros(960,540,N); for i = 1:N img_tmp = imread(f_list(i).name); img(:,:,i) = img_tmp(:,:,1); img(:,:,i) = rgb2gray(img_tmp); end bck_img = (mean(img,3)); %average background template %subplot(121); imagesc(bck_img) %subplot(122); imagesc(img(:,:,1)) clear img; pause(5) %gaussian filter hsize = 80; sigma = 20; gaus_filt = fspecial('gaussian',hsize , sigma); %gaus_filt = fspecial('log',hsize , sigma); %subplot(121); imagesc(gaus_filt) %subplot(122); mesh(gaus_filt) colormap(gray)

SE = strel('diamond', 0); %this makes a matrice object for passing into imdilate())

%% Operations on frames cd('C:\Users\USER\Desktop\final year project\FFF\car'); f_list= dir('*jpeg'); for i = N:1:length(f_list)

    img_tmp = double(imread(f_list(i).name)); %load image and convert to double for computation 
    img = img_tmp(:,:,1); %reduce to just the first dimension
   % img = rgb2gray(img_tmp)
    %subplot(221);
    imagesc(img);
    title('Raw');
    %subtract background from the image
    sub_img = (img - bck_img);
   % subplot(222);
    imagesc(sub_img);
    title('background subtracted');
    %gaussian blurr the image
    gaus_img = filter2(gaus_filt,sub_img,'same');     
   % subplot(223);
    imagesc(gaus_img);
    title('gaussian smoothed');
    %threshold the image
   % subplot(224);
   hist(gaus_img(:));
    thres_img = (gaus_img < 20);
    thres_img = ~thres_img;
    thres_img = bwareaopen(thres_img,20);
    se2 = strel('disk',1);        
    thres_img = imerode(thres_img,se2);
    %counting no of blobs
    [L, num] = bwlabel(thres_img);
    stats = regionprops(L, 'Centroid');
    thres_img = ~thres_img;
   % subplot(224);
   imagesc(thres_img);
    %subplot(224);imagesc(gaus_img);
    title('thresholded');
    clc
    fprintf('Total parking spaces  :  %i\n',tps)
    fprintf('filled parking spaces :  %i\n',num)
    fprintf('vacant parking spaces :  %i\n',tps-num)
Can you help me with the code? Also, sometimes greeen and blue cars are not detected.