An artificial vision system for identify and classify objects J.N. Rodríguez, L. Acosta, A. Hamilton, J.A. Mendez, L. Moreno Department of Physics, Electronics & Systems University of La Laguna. Spain Abstract : - In this paper a vision system for a highly specialized autonomous robot is presented. The task of the robot is the setting and clearing of tables in a restaurant. The environment in which the robot will operate will be controlled and well known. The first objective of the robot is to navigate around the room to find collectable sites, that is, tables. Specifically, this paper is focused on the task that has to be performed once the robot is located in a table. This consists of analyzing the scene and extracting all the necessary information to collect the distinct objects located on it. The application is designed to identify and collect the following objects: dishes, bottles, glasses, forks, spoons, and knives. Key-Words: - Artifical vision, Autonomous robotics, Hough transformation. 1 Introduction This work presents a vision system for scene analysis integrated in a service robot application (SERVIROB). SERVIROB is made up of a mobile base with a manipulator situated on top of it. This application consists of the design of an autonomous robot for setting and clearing tables in a restaurant. To accomplish this task the robot is autonomously navigating in the room and has to fulfill various objectives. The first one is to identify where the tables are. Once at the table the robot should be able to identify and classify the objects on it. In this study we have considered the following objects: -White dishes of two different sizes -Metal forks, spoons and knives -Clear glasses -Clear bottles These objects must be collected and deposited in individual recipients. A specially designed manipulator is mounted on the robot to accomplish this task. The vision module of the complete application analyses the scene environment when SERVIROB is in a position to collect objects. The objective of this module is to provide the necessary information to the main module of SERVIROB so that it can perform pick and place tasks in the table. The solution proposed is to equip the robot with image analysis software and artificial vision system composed of two cameras and a ring-shaped lighting mechanism. This lighting mechanism fulfills the need to stabilize/regulate the light conditions in all possible work environments where the robot can be found. We can analyze the scene by placing the cameras in different perspectives: one vertical and the other zenithal. This distribution allows relevant information can be retrieved to perform the collection of the object. A tool has been developed for the general purpose of capturing and processing images which provide a series of operators which are used when constructing classification strategies. These operators incorporate techniques which allow for improvements in the pre- processing of images, generation of contour maps, the calculation of generalised Hough transforms and moments. This tool has been developed using C++ with the graphics library GTK-- for Linux. Several steps were used in solving the problem. Two of them are common in all kinds of vision problems: pre-processing and contour detection. The last step in the classification process applies different techniques to obtain different diagnostics of the same scene. Solving the problem with different approximations provides greater robustness and confidence to the system. The techniques used are based on the description of distinct elements that can be found on top of tables using geometric characteristics. The most important tool used in the vision module is the Hough transformation. The idea is to describe different elements which can be found given these characteristics. These characteristics must be general enough to allow the system a margin of error but at the same time be sufficiently defined so that the object can