Fast implementation of a robust pedestrian recognition system

In this contribution a real time capable pedestrian recognition system is presented. The AdaBoost based cascade approach is applied and enhanced. A special implementation of the feature extraction algorithms yields to a remarkable increasing of the computation efficiency. Furthermore the AdaBoost classifier uses decision trees as basic classifiers (weak learner). Therefore, the computational costs are reduced in addition without penalizing the classification performance. The details of the implementation, the computational costs as well as the classification results of real scenarios are presented. The presented work is part of the WATCH-OVER project.