Study of GP representations for motion detection with unstable background

Detecting moving objects is a significant component in many machine vision systems. One of the challenges in real world motion detection is the unstability of the background. An ideal method is expected to reliably detect interesting movements from videos while ignoring background/uninteresting movements. In this paper, Genetic Programming (GP) based motion detection method is used to tackle this issue, as it is a powerful learning method and has been successfully applied on various image analysis tasks. The investigation here focuses on the various representations of GP for motion detection and the suitability of these approaches. The unstable environments in this study include ripples on river, rainy background and moving cameras. It can be shown from the results that with a suitable frame representation and function set, reliable GP programs can be evolved to handle complex unstable background.