Change Detection

Synonyms

Normalcy modeling

Definition

A determination that there are significant differences between visual scenes

Background

Change detection is a key task for computer vision algorithms. The goal is to compare two or more visual scenes and report any significant differences between the scenes. As with many vision tasks, the meaning of significant is application dependent. The change detection task can be rendered somewhat more concrete by considering the types of changes that are not typically of interest. Examples of changes that are usually irrelevant are:

• Camera viewpoint

• Varying illumination

• Wind-based motion

• Weather, e.g., snow and rain

The implementation of algorithms that can detect interesting changes while ignoring trivial changes such as these is a very difficult problem, and only quite limited change detection capabilities have achieved to date. It is also the case that the change detection task, when viewed broadly, overlaps the scope of many other vision tasks such as...