Object detection by combined model based on cascaded adaboost

Single feature-based model always meets the difficulties of poor detection performance and slow detection speed for object with large variances in color, texture, and shape. A novel cascaded and additive model based on cascaded Adaboost classifier is proposed in this paper. This combined model consists of two cascaded Adaboost classifiers which are independently trained with edge-fragment feature and Haar feature to describe the whole object and one of its stable components, respectively. The final classification decision of the combined model is made according to the stage indexes by which a sample is rejected or accepted in the two cascaded classifiers. Experiments on several test databases show that the combined model can take advantages of the speed merit of Haar feature and the robustness of edge-fragment feature. Compared with single feature-based model, the detection performance of the combined model is greatly improved. © 2009 Acta Automatica Sinica. All rights reserved.