Do you have a reference video with no pedestrians in it?
If yes, then it will be easy to find the size of pedestrians.
Try using this approach:
Get all the frames in real time and subtract each of them from their respective reference frame from reference video.
function frame = readFrame()
frame = step(obj.reader);
From the difference, You will get some blobs. Use regionprops() to detect the area of those blobs. It will give the idea of their size.
I hope it will be useful.