Selective motion detection by Genetic Programming

Abstract

Motion detection is a vital part of vision systems, either biological or computerized. Conventional motion detection methods in machine vision can differentiate moving objects from background, but cannot directly handle different types of motions. In this paper, we present Genetic Programming (GP) as a method which not only removes relatively stationary background, but also can be selective on what kind of motions to capture. Programs can be evolved to select a certain type of moving objects and ignore other motions. That is to select fast moving target and ignore slowing moving ones. Furthermore programs can be evolved to handle these tasks even when the camera itself is in relatively arbitrary motion. This general GP method does not require additional process to differentiate various types of motions.