Improving Pedestrian Detection Using Support Vector Regression

Abstract

Pedestrian detection has been always a challenging problem in computer vision. Numerous approaches based on features extraction and classification have been proposed over the years. In this paper, we present a novel pedestrian detection approach based on supervised classification. We propose here the use of basic statistical operators to adapt support vector regression (SVR) to binary classification. The classification chain adopted in this work is presented as follows: First, we use Haar wavelet decomposition and Histograms of Oriented Gradients (HOG) for features extraction. For the classification task, we use our proposed method and compare it with KNN and SVM classifiers. Experiments have been done on a public pedestrian data set. The obtained results prove the high performance of our proposed classification approach.