A new neural network approach for visual autonomous road following

Abstract

This paper presents an original approach for visual identification of road direction in autonomous vehicle navigation using a neural network classifier called Concurrent Self-Organizing Maps (CSOM). For comparison, we also evaluate the performances of other two neural classifiers ( Multilayer Perceptron (MLP) and supervised Self-Organizing Map (SOM) ) as well as those of the well-known statistical classifier of Nearest Mean (K-Means). The proposed model has two main processing stages: (a) feature selection, using either a standard edge detection algorithm or the Hough transform; (b) classification, using one of the above mentioned classifiers. The path to be identified has been quantized in three output directions. We present the experimental results obtained by computer simulation, when for training and testing the neural model we used a data set of 210 road images from the CMU VASC Image Database. A real time neural path follower implemented on a mobile robot is also experimented.