An adaptive weighted boosting algorithm for road detection

Road detection is one of most important branches of intelligent vehicle. Previously we developed a road detection system based on boosting using feature combination. In this paper, features are selected and weighted according to their classification ability, which come from boosting classifiers with each single feature. Moreover, they are weighted according to the boosting training process. Shadows are always high error rate in road detection. To improve performance of shadow, special features are induced in the feature set. The experiment result shows that the performance of this method is better than raw boosting.