Design of an accurate pedestrian detection system using modified HOG and LSVM

Abstract

This paper focuses on detecting a pedestrian in an image. This real time application aims for high detection accuracy as well as faster computation. For higher accuracy and detection rate Histogram of Oriented Gradients (HOG) algorithm is used. Further, Linear Support Vector Machine (LSVM) classification is used for faster and reliable classification. Since the HOG algorithm is compute expensive several modifications have been made in order to get the best results for real time application. We have used bilinear interpolation and L2-normalisation for more reliable output. Further since the data is linearly separable a LSVM is designed in Matlab. The proposed algorithm provides an accuracy of 93.27% with a high true positive rate of 92.27% and a minor false positive rate of 4%.