Sistema de localização topológica utilizando visão computacional

Abstract

This paper presents an innovative methodology that uses computer vision techniques to perform the topological localization of an autonomous vehicle. The great advantage of this technique is that it eliminates the use of GPS or any continuous time position sensor, which can significantly increase safety in regions where location sensors are limited or absent. The methodology consists of building a topological map of the region of interest, where the points of interest are defined. To do this, several images of each coordinate are collected and go through filters and processing to form a georeferenced image bank. From there, the system receives as input a video, where the images are compared with the images in the image bank, using the SURF algorithm, to define if there is a correspondence with the coordinates of interest. If a match is identified, the algorithm defines the location of the vehicle on the topological map. The results of the experiments performed show a 91.5% accuracy rate in detecting the points of interest within the topological map, indicating that this methodology can complement the navigation system of an autonomous vehicle efficiently and accurately.