How to get the coordinates of each room in a floor plan image through object identification method?

I have a floor plan image which contains objects and walls. I removed all the objects in it so it now contains only walls. Now I want to get the coordinates of each room.

I am new to this platform. Can somebody please help me how to get the coordinates of each room? Or Can I get the end points of each line segment?

This is the code.

function [bbox_loc,sign_objects,index1,st,furniture]=classify_objects_test1(I1)
%  clear all;
%  clc;
%  close all;
I_initial=imread("C:\\Users\\User\\Desktop\\tkinter_codes\\floorplans\\ROBIN\\Dataset_3roomsmall\\Cat1_1.jpg");

%I_initial = imread('room_4.png');
I=I_initial;
imshow(I);
SE = strel('square',10);
I = imdilate(I,SE);
SE = strel('square',10);
I = imerode(I,SE);
wall_image=I;

wall = I;
I=I_initial;
wall=im2bw(wall);
imshow(wall);

I=im2bw(I);
I=imcomplement(I);
wall = imcomplement(wall);
I=I-wall;
% imshow(I);
I=imcomplement(I);
SE = strel('square',1);
IM2 = imdilate(I,SE);

IM2 = I;
IM2 = imerode(IM2,SE);

imwrite(IM2,'fl_0sym.tif');


se1=strel('square',8);
intm1=imerode(IM2,se1);
se2=strel('square',4);
intm2=imdilate(intm1,se2);
intm3=imdilate(intm2,se2);
imwrite(intm3,'fl_0sym1.tif');
I = imread('fl_0sym1.tif');
I = imcomplement(I);

J = imfill(I);
J = ~J;

imwrite(J,'blb_0_1.jpg');


figure=I_initial;
[m,n] = size(figure);
figure1 = figure(:,:,[1 1 1]);
figure = im2bw(figure);
[m,n] = size(figure);
image = imread('blb_0_1.jpg');
BW = im2bw(image);
BW = imresize(BW,[m,n]);
BW = ~BW;


%for 4,1,9,8
 SE = strel('square',10);
 BW = imdilate(BW,SE);

%% figure;imshow(BW);
st = regionprops(BW,'BoundingBox');
 shape.Inserter = vision.ShapeInserter('LineWidth',4,'BorderColor','Custom','CustomBorderColor',uint8([255 0 0]));
 
str_ele=strel('disk',1,0);
 for k = 1:length(st)
%  k=1;       
%         for k = 1:1
thisBB = st(k).BoundingBox;
%figure1 = step(shape.Inserter,figure1,bbox_loc{k});
thisBBarea=thisBB(3)*thisBB(4);
if thisBB(3)>20 &&  thisBB(4)>20 
rectangle = int32([thisBB(1),thisBB(2),thisBB(3),thisBB(4)]);


end
 if(thisBBarea >870)
bbox_loc{k}=rectangle;
 
 crop_I_new=imcrop(figure1,rectangle);
 crop_I_new_gray=rgb2gray(crop_I_new);
 crop_I=im2bw(crop_I_new_gray,0.7);
%   figure;imshow(crop_I);
signature{k}=signature_find(crop_I);
if isempty(signature{k})
   
    signature{k}=[99999,99999,99999];
end
    
  figure1 = step(shape.Inserter,figure1,bbox_loc{k});
 end
end     
signature=signature(~cellfun('isempty',signature)); 

save('signature.mat','signature');
signature=load('signature.mat');
sign_objects=load('sign_object2');
for i=1:length(signature.signature)
    for j=1:length(sign_objects.sign_object2)

diff{i,j}=abs(signature.signature{1,i}-sign_objects.sign_object2{j,1}.count);

    
    end
       

end
for(i=1:size(diff,1))
    for(j=1:size(diff,2))
        if(sign_objects.sign_object2{j,1}.count(3)>1000)
            temp1(i,j)=diff{i,j}(1)+diff{i,j}(2);
        else
            temp1(i,j)=diff{i,j}(1)+diff{i,j}(2)+diff{i,j}(3);
        end
    end
end
for(i=1:size(temp1,1))
[temp1_min(i),index1(i)]=min(temp1(i,:));  
disp(sign_objects.sign_object2{index1(i),2});
  furniture{i,1}=sign_objects.sign_object2{index1(i),2}
  figure;imshow(figure1);
bbox_loc=bbox_loc(~cellfun('isempty',bbox_loc)); 
end

for k = 1:length(bbox_loc)
     H= text( double(bbox_loc{k}(1)),double(bbox_loc{k}(2)),sign_objects.sign_object2{index1(k),2});
           set(gcf,'DefaultTextColor','blue')
end
% for i=1:size(diff,1)
%      for j=1:size(diff,2)
% 
%         temp1(i,j)=(diff{i,j}(1));
%         temp2(i,j)=(diff{i,j}(2));
%         temp3(i,j)=(diff{i,j}(3));
% 
% 
%      end
%   [temp1_min(i),index1(i)]=min(temp1(i,:));  
%   disp(sign_objects.sign_object1{index1(i),2});
%   furniture{i,1}=sign_objects.sign_object1{index1(i),2}
% end

I am attaching a zip file which contains files which will be used by this code while running.

This is the floorplan image with objects.

This is the image with only walls.

Answer 1

[Matt J]

One approach is to close all of the doors morphologically:

load('Walls')

e=ones(1,105);

A=imclose(imclose(~Image,e),e.');
B=imfill(A,'holes')&~A;

imshow(B)

Now, you can simply use regionprops to get all of the pixel coordinates in each room,

>> Rooms=regionprops(B,'PixelList')

Rooms = 

  3×1 struct array with fields:

    PixelList

[CHINTALA NIRMAL SRINIVAS]

Could you please tell me how can I get the bounding boxes around the rooms? I am unable to get.

[Matt J]

regionprops will give those as well.

[CHINTALA NIRMAL SRINIVAS]

No I mean when I try to get bounding boxes it is giving error like unrecognised variable 'rectangle' (that function which I use for getting bounding boxes).Could you please tell me what should I write?

[Matt J]

I don't know what you mean by "that function". Why not simply do,

regionprops(B,'BoundingBox')

[CHINTALA NIRMAL SRINIVAS]

When I tried to draw rectangles around it, it gave the error. I used regionprops to get the bounding boxes

[CHINTALA NIRMAL SRINIVAS]

For floorplan where rooms are not in the form of rectangles,(they are of the form of trapezium) , closing all the doors is not done appropriately.

Could you please suggest me a solution to this problem?

[Matt J]

I think you should start a new question for this more complex case. It looks like it might require a very different line of solution.

Answer 2

[Matt J]

The Computer Vision Toolbox has a few corner detection routines,

https://www.mathworks.com/help/vision/ref/detectharrisfeatures.html

https://www.mathworks.com/help/vision/ref/detectmineigenfeatures.html

[CHINTALA NIRMAL SRINIVAS]

Could you please tell me how can I extract the corner points to a file?

[Image Analyst]

But, just scanning your code, it looks like you get the bounding boxes, so that gives you the coordinates of each room, right?

[CHINTALA NIRMAL SRINIVAS]

Also since the size of the wall line is large, there are two corner points at one corner. I want to have one point at one corner. I tried to skeleton the image but the skeletonization was not done properly. Could you please tell me a solution?

[CHINTALA NIRMAL SRINIVAS]

I am getting the bounding boxes of the objects in the floorplan. But How can I get the coordintes of room?

I am unable to bound the rooms with bounding boxes.

[Image Analyst]

Call bwskel() to get the skeleton of the walls

skelImage = bwskel(~mask);

Then get the endpoints of that image with bwmorph. Call pdist2() to get all the distances between endpoints. If the endpoints are within a certain distance (whatever a door width is), AND either the rows are the same or the columns are the same (so we're only getting vertical or horizontal gaps), then draw a line connecting the room going across the doorway.

There are hundreds of algorithms for getting floor plans that have been published here:

http://www.visionbib.com/bibliography/cartog943i1.html

Look them over and code up the most promising one.

[CHINTALA NIRMAL SRINIVAS]

when I use bwskel some part of the wall is missing. What should I do?

[CHINTALA NIRMAL SRINIVAS]

Could you please tell me what should I do to get the bounding boxes around the rooms?