Problem regarding image processing in binary image

I'm currently working on image processing on an image on streets. Here are the codes I'm currently using

C=im2double(img); 
B=im2double(bgd);
maxchan=max(B,[],3);
threshmax = multithresh(maxchan,4);Qmax=imquantize(maxchan,threshmax);
bw=bwareafilt( Qmax==2,1);
bw=imclose(bw,strel('disk',10));
D=rgb2gray(bw.*(C-B));
thresh=multithresh(D,2);
result=bwareafilt( imquantize(D,thresh)>1, [10,inf]);
result2=imfill(result,'holes')
imshow(result2)

Then, I'll have a result as following

Are there any solutions to get a more complete car-shaped white blob and remove the lines on the road ?? This is because I'm attempting to count the number of the cars. There is another problem where if i take another picture with different lighting or different angle, the result of the image could be worse where i could not see any blobs. The images of the current image are attached as .m files. Thank you in advance. The other pictures where I could not extract anything are also attached.

Answer 1

[Image Analyst]

I'd use imclose() first to connect some separate blobs, like the car split into two or 3 parts due to a luggage rack on the roof. Then use bwareaopen() or bwareafilt() to get rid of blobs smaller than a certain size.

[Adrian Lim]

Thank you for the suggestion! Will try it out!

[Adrian Lim]

Do you have any idea that why is it that I couldn't get the similar results using the same code when I tried taking pictures of different angles and lighting?

[Image Analyst]

Your segmentation algorithm is not robust enough. What have you tried? Have you tried the tracking stuff int he Computer Vision System Toolbox? Have you tried using Gaussian Mixture Models? Or even just taking a recent mode image to get the background? Sorry but I can't help you develop a robust turnkey car counting system - that could take months.

Some code is here: https://www.mathworks.com/help/vision/examples/detecting-cars-using-gaussian-mixture-models.html?prodcode=VP&language=en

and here:

https://www.mathworks.com/help/vision/examples/tracking-cars-using-optical-flow.html

[Adrian Lim]

I've tried searching for informations and I've looked through those sites. The problem that I am facing is im using images but not videos. I'm using a webcam to snapshot from the model by putting cars on it and trying subtraction method to remove the background and count the cars.

Answer 2

[KALYAN ACHARJYA]

%% To remove roads

LB=2000;
seg=bwareaopen(result2,LB);
figure, imshow(seg)

[Adrian Lim]

Thanks for the suggestion!

[KALYAN ACHARJYA]

Welcome