New Symmetry Peak Processing and GA-based SVM Algorithm for Pedestrian Detection

Abstract

In this paper, we consider the problem of pedestrian detection using a two-stage visionbased approach. The first stage is hypothesis generation (HG), in which potential pedestrian are hypothesized. We proposed a method called “colors difference” to determine the leg position of pedestrian. Then a new symmetry peaks processing is performed to define how many pedestrians are covered in one potential candidate region. The second stage is hypothesis verification (HV). In this stage, all hypotheses are verified by the combination between Decision Tree and a Modified Adaptive Genetic Algorithm to find the best features subset based on Haarlike feature and an appropriate parameters set of Support Vector Machine for classific ation. Our methods have been tested on different real road images and show very good performance