Use of Artifical Neural Networks for Fusion of Infrared and Vision Sensors in a Mobile Robot Navigation System

Abstract

This paper presents a mobile robot navigation system that uses artificial neural networks for sensor fusion. The control system is based on the fusion of data from three infrared sensors and one vision sensor (VGA camera). Data from VGA camera was used to detect edges of obstacles in the environment, while infrared sensors were used to measure the distance from obstacles. Canny edge detector and LPQ descriptor were used for image pre-processing, i.e. edge detection and extraction of features from images. A multilayer perceptron network trained by the backpropagation algorithm was used to classify the detected obstacles. The control algorithm was implemented in the MATLAB software package and tested on the Robotino 2 mobile robot. The experiment was done indoors with static obstacles that represent the real environment. This experiment showed that the developed algorithm gives very good results with an accuracy of 72.07%. Ultimately, navigation system performs the tasks of detecting and avoiding most of the obstacles on which it was tested.