Forward vehicle detection using cluster-based AdaBoost

Abstract

A camera-based forward vehicle detection method with range estimation for forward collision warning system (FCWS) is presented. Previous vehicle detection methods that use conventional classifiers are not robust in a real driving environment because they lack the effectiveness of classifying vehicle samples with high intraclass variation and noise. Therefore, an improved AdaBoost, named cluster-based AdaBoost (C-AdaBoost), for classifying noisy samples along with a forward vehicle detection method are presented in this manuscript. The experiments performed consist of two parts: performance evaluations of C-AdaBoost and forward vehicle detection. The proposed C-AdaBoost shows better performance than conventional classification algorithms on the synthetic as well as various real-world datasets. In particular, when the dataset has more noisy samples, C-AdaBoost outperforms conventional classification algorithms. The proposed method is also tested with an experimental vehicle on a proving ground and on public roads, similar to 62 km in length. The proposed method shows a 97% average detection rate and requires only 9.7 ms per frame. The results show the reliability of the proposed method FCWS in terms of both detection rate and processing time.