On Detecting a Human and Its Body Direction from a Video

Abstract

Recently, car vision technologies have been paid much attention in the field of ITS (Intelligent Transport System). In particular, techniques for automatic detection of pedestrians or humans have been studied enthusiastically. A robust human detection algorithm has been reported recently. The HOG feature proposed by Dalal and Triggs 1) is a well known feature for representing and recognizing a human image. But, in its original algorithm, it includes an unnecessary part of the image such as the background. In this paper, we propose a method of detecting a human and its body direction employing the HOG feature based on a human model. The reason for detecting body direction is to predict his/her sudden step into a road. To the best of our knowledge, the HOG feature has been employed to date only for detecting a human and not for recognizing its body direction. The proposed method is described in the following and the performance of the method is shown by the experiments employing real outdoor scenes.