Low cost automatic object segmentation by detecting a signature motion within an optical flow signal

Abstract

Induced motion provides a cheap and reliable way of segmenting an object from the background within an image. As such it has potential applications to a wide variety of machine vision systems. This paper presents a system that segments an object from the background by inducing a sinusoidal motion and then searching for the signal in the resulting optical flow. The effectiveness of the system is evaluated using ROC curves and shown to be reliable and robust in the presence of distractor motions.