Accurate Modeling of Image Features Using Evolutionary Computing

This chapter describes two parametric models capable of modeling and location of complex corners, as well as retro-reflective targets, through a certain class of intensity variations of a depicted 3D scene. We present a parametric corner modeling based on a Unit Step Edge Function (USEF) that defines a straight-line edge. The simplicity of model definition provides the flexibility and generality useful in modeling complex corners. Thus, the proposed model can be scaled in complexity to create a multi-corner detector using simple arithmetic operations. Also, we provide a new parametric model useful in the accurate detection of retro-reflective targets. Both models are distribution functions that model the optical and physical characteristics found in digital imaging systems. Once a model is built, it is possible to retrieve through least squares the information that is useful in other machine vision tasks. The criteria for the high-accurate location of corners and targets are described and numerous examples of a real working system are presented for precision up to sub-pixel accuracy.