Fast multi-feature pedestrian detection algorithm based on histogram of oriented gradient using discrete wavelet transformMultimed Tools Appl

About

Authors
Gwang-Soo Hong, Byung-Gyu Kim, Young-Sup Hwang, Kee-Koo Kwon
Year
2015
DOI
10.1007/s11042-015-2455-2
Subject
Media Technology / Computer Networks and Communications / Hardware and Architecture / Software

Similar

Pedestrian detection using wavelet templates

Authors:
M. Oren, C. Papageorgiou, P. Sinha, E. Osuna, T. Poggio
1997

A novel algorithm of multi-sensor image fusion based on wavelet packet transform

Authors:
Yinglei Cheng, Rongchun Zhao, Fuyuan Hu, Ying Li
2006

An overview of fast pedestrian detection: Feature selection and cascade framework of boosted features

Authors:
Jian Zhang, Sakrapee Paisitkriangkrai, Chunhua Shen
2009

Text

Multimed Tools Appl

DOI 10.1007/s11042-015-2455-2

Fast multi-feature pedestrian detection algorithm based on histogram of oriented gradient using discrete wavelet transform

Gwang-Soo Hong · Byung-Gyu Kim ·

Young-Sup Hwang · Kee-Koo Kwon

Received: 2 April 2014 / Revised: 11 October 2014 / Accepted: 7 January 2015 © Springer Science+Business Media New York 2015

Abstract A convergence between a natural user interface (NUI) and advanced driver assistance system is considered as a next generation technology. This kind of interfacing system technology becomes more popular in driver assistance system of automobile. Especially, pedestrian detection is an important cue for intelligent vehicles and interactive driver assistance system. In this paper, we propose a pedestrian detection feature and technique by combining histogram of the oriented gradient (HOG) and discrete wavelet transform (DWT). In the method, the magnitude of motion is used to set region of interest (ROI) for improving detection speed. Then, we employ multi-feature for a pedestrian detection based on the HOG and DWT. In last stage, to classify whether a candidate window contains a pedestrian or not, the designed multi-feature is learned by using the training data with the support vector machine (SVM) mechanism. Experimental results show that the proposed algorithm increases the speed-up factor of 27.21 % by comparing to the existing method using the original HOG feature.

Keywords Natural user interface (NUI) · Region of interest (ROI) · Discrete wavelet transform (DWT) · Histogram of oriented gradient (HOG)

G.-S. Hong · B.-G. Kim () · Y.-S. Hwang

Department of Computer Engineering, SunMoon University, Asan, Korea e-mail: bg.kim@ieee.org

G.-S. Hong e-mail: honzolv@mpcl.sunmoon.ac.kr

Y.-S. Hwang e-mail: young@sunmoon.ac.kr

K.-K. Kwon

Automotive IT Platform Research Team, ETRI, Daegu, Republic of Korea e-mail: kwonkk@etri.re.kr

Multimed Tools Appl

Fig. 1 Timeline of pedestrian protection measures 1 Introduction

A natural user interface (NUI) is the next generation of computer interfaces that has started to take off in the past few years. A NUIs are a set of methods and mechanisms for more intuitively connecting people with technology. Instead of intermediating hardware for controlling and displaying information, a natural user interfaces enable driver to interface with natural behaviors or driving environment. In terms of this, NUIs can be widely utilized in interactive driver assistance system.

The driver assistance system technologies will be reflected in coordinated and streamlined displays, and control panels. The driver can achieve various kinds of warning messages or useful information from intelligent interactive assistance system. Such new information and control technologies that make vehicles smarter are arriving on the market as an optional equipment or specialty after-market components [1].

Driver assistance system can reduce the accidents caused by pedestrian negligence.

Pedestrian detection is valuable for intelligence vehicles and driver assistance. A traffic accident causes death and an injury around world. The pedestrian in potential danger can be marked and then the driver will be warned by following some measures. Some measures will be taken to protect the pedestrian by accurately detecting pedestrian [9, 10].

Pedestrian safety can be improved at several stages in Fig. 1. Pedestrian protection system can detect the pedestrians and prevent accidents by warning the driver or triggering autonomous braking through NUIs. The objective of pedestrian warning algorithms is to accurately detect pedestrians and provide the driver with informative warnings. In the eyes of the driver, the end product of good system provides a timely warning and additional information. Although generic image processing algorithm have been addressing similar goals for many years, there are several problems that are unique to image processing in automotive applications. To provide accurate warning and additional information to driver, problems to detect pedestrian accurately still to can be sorted out.

Conventionally, the computational complexity is one major issue of the pedestrian detection from video has two steps such as hypothesis generation and hypothesis verification. In the hypothesis generation step, the rectangular region of candidate pedestrian is selected,

Multimed Tools Appl

Fig. 2 Flow chart of the proposed algorithm

Read freame

Pre-Processing

Pedestrian candidates search

Compute magnitude of motion

Normalized magnitude of motion > threshold

Feature extraction and classification

Set pedestiran candidates

Multi-feature extraction

SVM (support vector machine) which is called region of interest (ROI). And then, in the hypothesis verification step, intensive algorithms area performed to recognize whether it is pedestrian in the ROIs [6–8].

The proposed algorithm makes two main contributions. The first main contribution is to speed up the process for pedestrian detection using detecting the pedestrian candidate regions in the image. If the number of ROIs generated in hypothesis generation step is large, whole system speed would be decreased significantly. Hence, most of the methods focused on the hypothesis generation step.

The second main contribution is that we focus on the hypothesis generation step as well as hypothesis verification step to reduce the computational complexity. In hypothesis verification step, to reduce the computational complexity, multi-feature which is combined between DWT and HOG feature is extracted. In hypothesis generation step, the magnitude of the motion information is used to distinguish the pedestrian candidates in whole image.

The structure of the paper is as follows. In Section 2, we will review the related works on the pedestrian detection problems. In Section 3, we will give details of the feature extraction and pedestrian candidate regions for fast detection. The experimental results and some discussions are presented in Section 4. We will make a conclusion in Section 5. 2 Related works