Optimizing a Conspicuous Point Detector for Camera Trajectory Estimation with Brain Programming


The interaction between a visual system and its environment is an important research topic of purposive vision, seeking to establish a link between perception and action. When a robotic system implements vision as its main source of information from the environment, it must be selective with the perceived data. In order to fulfill the task at hand we must contrive a way of extracting data from the images that will help to achieve the system’s goal; this selective process is what we call a visual behavior. In this paper, we present an automatic process for synthesizing visual behaviors through genetic programming, resulting in specialized prominent point detection algorithms to estimate the trajectory of a camera with a simultaneous localization and map building system. We present a real working system; the experiments were done with a robotic manipulator in a hand-eye configuration. The main idea of our work is to evolve a conspicuous point detector based on the concept of an artificial dorsal stream. We experimentally show that it is in fact possible to find conspicuous points in an image through a visual attention process, and that it is also possible to purposefully generate them through an evolutionary algorithm, seeking to solve a specific task.