Performance evaluation of evolutionary algorithms for road detection

In this paper we present the first comparative study of evolutionary classifiers for the problem of road detection. We use seven evolutionary algorithms (GAssist-ADI, XCS, UCS, cAnt, EvRBF,Fuzzy-AB and FuzzySLAVE) for this purpose and to develop better understanding we also compare their performance with two well-known non-evolutionary classifiers (kNN, C4.5). Further we identify vision based features that enable a single classifier to learn to successfully classify a variety of regions in various roads as opposed to training a new classifier for each type of road. For this we collect a real-world dataset of road images of various roads taken at different times of the day. Then, using Information Gain (I.G) and CfsSubsetMerit values we evaluate the efficacy of our features in facilitating the detection. Our results indicate that intelligent features coupled with right evolutionary technique provides a promising solution for the domain of road detection.