Active learning based robust monocular vehicle detection for on-road safety systems

In this paper, the framework is presented for using active learning to train a robust monocular on-road vehicle detector for active safety, based on Adaboost classification and Haar-like rectangular image features. An initial vehicle detector was trained using Adaboost and Haar-like rectangular image features and was very susceptible to false positives. This detector was run on an independent highway dataset, storing true detections and false positives to obtain a selectively sampled training set for the active learning training iteration. Various configurations of the newly trained classifier were tested, experimenting with the trade-off between detection rate and false detection rate. Experimental results show that this method yields a vehicle classifier with a high detection rate and low false detection rate on real data, yielding a valuable addition to environmental awareness for intelligent active safety systems in vehicles.