A network shaped cascade classifier based on potential functions for pedestrian detection

This paper proposes a Network Shaped Cascade Classifier(NSCC) based on potential functions for pedestrian detection. Potential function is exploited to capture the nonlinear information in the training set based on the multiple sample centers. A flexible structure in NSCC is used to combine the base classifier and potential function into a nonlinear cascade classifier, and NSCC can well inherit the advantages of the base classifier. We test our classifier on INRIA dataset, and achieve a much better performance than support vector machine.