I think he uses thresholding, but not on the image itself, but on the difference of two images.
Probably this application uses a fixed camera and he grabbed a reference frame that contains only the background. The application would work like this (pseudocode):
capture(referenceframe) //get a frame which doesn't contain any foreground
do
capture(newframe) //capture a new frame
read(backgroundframe) //read (generate or capture) frame that will be the replaced background
diff=abs(newframe-referenceframe) //calculate the difference between the two frames
thresholdedDiff=threshold(diff,thresholdvalue); //threshold
denoisedDiff=open(thresholdDiff) //some denoising
result=denoisedDiff*newframe+(1-denoisedDiff)*backgroundframe;
// or: foreach pixel in denoisedDiff : if pixel==1 : resultpixel=newframepixel, else : resultpixel=backgroundpixel
imshow(newframe,backgroundframe,result);
while(!keypressed)
I hope you get the idea, it isn't difficult at all. However it has some limitations, it won't work if the camera moves or the illumination changes (you have to update the reference frame).