Pedestrian Detection Based On Histogram Of Oriented Gradient

Pedestrian Detection, as a branch of target detection and an important component ofseveral realms such as Video Surveillance, Intelligent Transportation System (ITS), andKinematic Analysis, has important applications in practice. It is of great value for study and ithas long been one of the hot topics concerned by numerous researchers. Currently, themainstream research methods are as these three steps: to extract features from the targetscenario, to establish the target models, and to convert to the issues about patternclassification of machine learning. However, during the procedure of extracting features, theaccuracy can be influenced consequently when some changes occurred to the target, such asgesture, direction, shelter, illumination, or noise.In this paper, we achieve the pedestrian detection by adopting target characteristic ofHistogram of Oriented Gradient (HOG), and using Support Vector Machine (SVM) sorter fortraining. The improvement mentioned in the paper is to optimize the features of Histogram ofOriented Gradient (HOG) by Local Binary Pattern (LBP) in use. Histogram of OrientedGradient (HOG)’s feature reflects changes of gradient on the edge of an image, while LocalBinary Pattern (LBP) has an excellent expressing effect towards the local texture of an image.With plentiful experiments, the feature extraction method introduced in this paper caneffectively characterize the valid contour information of human bodies, and the SVM-basedclassification method has a good detection performance.