Skip to Main Content
This article describes a new method to measure 3D positions of elevator buttons from the mobile guard robot which we have developed in order for the robot to autonomously get on and off the conventional elevators of a building for patrolling different floors. ne measurement system uses a single camera on the robot and a 2D artificial landmark on the wall holding the elevator buttons, and it also takes advantage of the laser navigation system which enables the robot to stop facing the wall. Since the images captured by the camera mounted at the front body of the robot consisted of perspective patterns of the wall, we utilizes two neural schemes: one is the competitive neural net for pattern recognition invariant to projective transformations, and the other is the competitive learning scheme for vector quantization in order to obtain efficient template patterns. We also develop a method to calculate the 3D positions of the elevator buttons and examine the measurement errors for the robot to push the buttons.