This chapter will present the details of the attention operator, or landmark detector. The purpose of the landmark detector is to locate candidate landmarks in an image. These candidates are later provided to the tracker for the purposes of building a set of tracked landmarks. We will present here a brief overview of edge detection, followed by a motivation for avoiding high-level semantic feature extraction when stability and robustness are of importance. We will define a candidate landmark as a local maximum of the edge element distribution in an image, and provide some examples which will demonstrate the behaviour of the operator. Our approach will be motivated by the goals of robustness and domain-independence.