moving shadow removal in video

4 views (last 30 days)
SHOBA MOHAN
SHOBA MOHAN on 24 Dec 2017
I am trying to remove moving shadows in video using the 'stationary wavelet transform technique' as mentioned in the reference paper. I coded in MATLAB but not getting expected output. I am begineer in MATLAB so Can anyone please review my code to check whether it followed all the steps as mentioned in the paper.
Recap of the algorithm followed in the reference paper:
(i) Convert the rgb video frame to hsv (ii)split the hsv into individual component (iii)Find the absolute difference between h,s and v component of current and background frame (iv)Apply the SWT transformation on difference 's' and 'v' (V)Compute the skewness value for the swt output (vi)For shadow detection: If the swt output of 'v' greater than its skewness, assign those pixel value to '1' else'0' (vii) For shadow removal: thresholding operation is applied based on swt output of's'
function shadowremoval()
obj = setupSystemObjects();
while ~isDone(obj.reader)
frame = readFrame();
mask1 = shadow(frame);
displayTrackingResults();
end
%%Create System Objects
function obj = setupSystemObjects()
% Create a video file reader.
obj.reader = vision.VideoFileReader('visiontraffic.avi');
% Create two video players, one to display the video,
% and one to display the foreground mask.
obj.videoPlayer = vision.VideoPlayer('Position', [10, 250, 700, 400]);
obj.maskPlayer = vision.VideoPlayer('Position', [720, 250, 700, 400]);
obj.detector = vision.ForegroundDetector('NumGaussians', 3, ...
'NumTrainingFrames', 40, 'MinimumBackgroundRatio', 0.7);
end
%%Read a Video Frame
% Read the next video frame from the video file.
function frame = readFrame()
frame = obj.reader.step();
end
%%Perform the operation to remove shadows
function mask1 = shadow(frame)
Background=0.0;
% Detect foreground.
mask1 = obj.detector.step(frame);
mask1 = uint8(repmat(mask1, [1, 1, 3])) .* 255;
% Apply morphological operations to remove noise and fill in holes.
% mask1 = imerode(mask1, strel('rectangle', [3,3]));
% mask1 = imclose(mask1, strel('rectangle', [15, 15]));
mask1 = imopen(mask1, strel('rectangle', [15,15]));
mask1 = imfill(mask1, 'holes');
% Now let's do the differencing
alpha = 0.5;
if frame == 1
Background = frame;
else
% Change background slightly at each frame
% Background(t+1)=(1-alpha)*I+alpha*Background
Background = (1-alpha)* frame + alpha * Background;
end
% Do color conversion from rgb to hsv
x=rgb2hsv(mask1);
y=rgb2hsv(Background);
% Split the hsv component to h,s,v value
Hx = x(:,:,1);
Sx = x(:,:,2);
Vx = x(:,:,3);
Hy = y(:,:,1);
Sy = y(:,:,2);
Vy = y(:,:,3);
% Calculate a difference between this frame and the background.
dh=(abs(double(Hx) - double(Hy)));
ds1=(abs(double(Sx) - double(Sy)));
dv1=(abs(double(Vx) - double(Vy)));
% Perform the 'swt'
[as,hs,vs,ds] = swt2(ds1,1,'haar');
[av,hv,vv,dv] = swt2(dv1,1,'haar');
%Compute the skewness value of 'swt of v'
sav=skewness(av(:));
shv=skewness(hv(:));
svv=skewness(vv(:));
sdv=skewness(dv(:));
%Compute the skewness value of 'swt of s'
sas=skewness(as(:));
shs=skewness(hs(:));
svs=skewness(vs(:));
sds=skewness(ds(:));
%Perform the thresholding operation
b=(av>=sav);
c=(hv>=shv);
d=(vv>=svv);
e=(dv>=sdv);
f=(as>=sas);
g=(hs>=shs);
h=(vs>=svs);
i=(ds>=sds);
j=(b&f);
k=(c&g);
l=(d&h);
m=(e&i);
%Perform the inverse 'swt'operation
recv = iswt2(b,c,d,e,'haar');
recs= iswt2(j,k,l,m,'haar');
de_shadow=cat(3,dh,recs,recv);
mask1=hsv2rgb(de_shadow);
mask1=rgb2gray(mask1);
end
function displayTrackingResults()
% Convert the frame and the mask to uint8 RGB.
frame = im2uint8(frame);
mask1 = uint8(repmat(mask1, [1, 1, 3])) .* 255;
% Display the mask and the frame.
obj.maskPlayer.step(mask1);
obj.videoPlayer.step(frame);
end
end
  2 Comments
Karbala'a Unvi. Science
Karbala'a Unvi. Science on 9 Oct 2020
Dear Sir
I whish you help me in a code to detect the shadow using HSV color system..
Looking forward for your help
Thank you
Zeyad

Sign in to comment.

Answers (1)

Bernhard Suhm
Bernhard Suhm on 7 Jan 2018
This forum is meant to help you overcome a problem with a specific MATLAB function, or general guidance for a narrowly defined problem - not debug your implementation of a whole algorithm. Another option you could explore is model the background and foreground pixels, you can do this using the vision.ForegroundDetector capability in CVST.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by