Included is the homework #3 files. In the email body, the analysis is 
included:

The bar graph shows a moderate amount of outlier pixels at the beginning, 
but these are simply noise pixels. Although there is a small further 
increase when the subject walks into the image, the difference drops to 
noise level as the subject continues to enter. The differences do not become 
significantly different from noise until the subject walks directly in front 
of the camera. This makes sense because when the subject is far from the 
camera, his disruption of the image is about the same as other noise, but 
when he gets close, he is disrupting many more pixels. A security system 
like this would present two problems. Either the system is triggered by 
noise, causing many false alarms, or the system might not detect potential 
threats, especially if the threat has the sense of mind not to walk directly 
in front of the camera.

The bar graph for v2 does not show a significant increase in pixels above 
noise level until frame 11 or so, when the subject had just begun to enter 
the image. After that, the number of pixels stays above noise level until 
the subject goes off-camera. Hence, the system does work better with the 
second image, although the resulting system is still not good enough for 
security, especially if the subject were to be seen in only a small portion 
of the camera's image, and any decent intruder would take the time to case 
out the camera locations as to minimize detection. Hence, the system could 
still be easily fooled by professionals. The difference in the background 
images were mainly in the positions that the person was in, as well as 
certain objects on the desk which may have had their lighting significantly 
changed during the entry.