Conventional approaches to moving objects detection
include frame difference (or temporal difference) algorithms,
background subtraction algorithms, optical flow algorithms,
and statistical learning algorithms. Optical flow and
statistical learning algorithms have much computational
complexity and are not suitable for video surveillance
applications. While background subtraction is very sensitive
to light changes. In contrast, frame difference algorithms
discern static objects (having null differences) from moving
objects (having no-null differences), and they are simple
and easy to be implemented. So they are considered to be
suitable for video surveillance applications.