夜间驾驶场景下基于中心像素的车辆检测方法

需积分: 7 75 浏览量更新于2024-09-05 收藏 1.54MB PDF 举报

本文主要探讨了夜间驾驶场景下，利用单目车内摄像头进行车辆检测的一种创新方法。针对在夜间识别车辆形状具有挑战性的问题，作者提出了一种基于中心敏感性和支持向量机（SVM）的视觉车辆检测策略。传统的自动多级阈值法在处理夜间场景时易受环境光干扰，因为这种方法依赖于整个图像的亮度来确定阈值。论文的核心技术是利用高斯拉普拉斯算子（Laplacian of Gaussian, LoG），这是一种局部特征检测器，它能够更有效地分离车辆的亮点（头灯和尾灯）与周围环境的亮度差异。LoG能够提高对背景噪声和光照变化的鲁棒性，从而减少误报和漏报的可能性。然而，LoG计算响应的代价较高，这可能会影响实时性能。为了降低计算负担，作者引入了名为"中心敏感性"（Center Sensitivity, CS）的方法。中心敏感性通过优化处理流程，聚焦于像素簇（即明亮的blob）的核心区域，减少了对整个图像的计算需求，同时保持了对车辆识别的有效性。这种方法巧妙地平衡了算法的性能和效率，使得在复杂且光照条件多变的夜间驾驶环境中，车辆检测的准确性和实时性都得到了提升。这项研究对解决夜间行车环境中车辆检测问题提供了新的思路，展示了如何利用高级图像处理技术（如LoG和中心敏感性）提高算法的适应性和稳定性，这对于智能交通系统的发展以及自动驾驶技术的进步具有重要意义。IEEETRANSACTIONSONINTELLIGENTTRANSPORTATIONSYSTEMS期刊上的这一研究成果，无疑为相关领域的研究者们提供了一个值得借鉴和改进的基础。

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 16, NO. 5, OCTOBER 2015 2599

Vision-Based Nighttime Vehicle Detection

Using CenSurE and SVM

Naoya Kosaka and Gosuke Ohashi, Member, IEEE

Abstract—In this paper, we propose a method for detecting

vehicles from a nighttime driving scene taken by an in-vehicle

monocular camera. Since it is difﬁcult to recognize the shape of

the vehicles during nighttime, vehicle detection is based on the

headlights and the taillights, which are bright areas of pixels

called blobs. Many research studies using automatic multilevel

thresholding are being conducted, but these methods are prone to

get affected by the ambient light because it uses the luminance of

the whole image to derive the thresholds. Owing to such reasons,

we focused on the Laplacian of Gaussian operator, which derives

the response of luminance difference between the blob and its

surroundings. Compared with automatic multilevel thresholding,

Laplacian of Gaussian operator is more robust to the ambient

light. However, the computational cost to derive the response

of this operator is large. Therefore, we used a method called

Center Surround Extremas to detect the blobs in high speed. Since

the detected blobs include nuisance lights, we had to determine

whether the blob is a light of a vehicle or not. Thus, we classiﬁed

them according to the features of the blob using support vector

machines. Then, we detected vehicle trafﬁc lane and speciﬁed

the region where the vehicle may exist. Finally, we classiﬁed the

blobs based on the movements across the frames. We applied the

proposed method to nighttime driving sequences and conﬁrmed

the effectiveness of the classiﬁcation process used in this method

and that it could process within frame rate.

Index Terms—Intelligent transport systems, nighttime driving

scenes, vehicle detection, Center Surround Extremas.

I. INTRODUCTION

ECENTLY, many researches on Intelligent Transport

Systems (ITS) are being conducted. As part of ITS, the

approaches based on vehicle detection from images taken by

a camera mounted on a vehicle have been conducted [1], [2].

Researches using a vehicle-mounted camera for various on-

road applications are attracting attention, because the cost of

the camera is lower than other sensors, such as radar, and it can

identify the shape, color, and size of an object in high speed ow-

ing to th e develo pment of arithmetic capacity of the c omputer.

Based on computer vision, a system that measures the distance

to the object to warn drivers o f the dangers of possible collision

and a system th at prevents the vehicle fro m deviating from the

lane are being developed. From such a background, the aim of

this stud y is to detect vehicles from images at nighttime for the

Manuscript recei ved September 22, 2014; revised December 26, 2014 and

March 9, 2015; accepted March 12, 2015. Date of publication May 7, 2015;

date of current version September 25, 2015. The Associate Editor for this paper

was B. Morris.

The authors are with the Department of Electrical and Electronic Engineer-

ing, Shizuoka University, Hamamatsu 432-8561, Japan (e-mail: tegooha@ipc.

shizuoka.ac.jp).

Color versions of one or more of the ﬁgures in this paper are available online

at http://ieeexplore.ieee.org.

Digital Object Identiﬁer 10.1109/TITS.2015.2413971

application of automatic headlight controller, which switches

between low and high beams depending on the existence of

other vehicles. At nighttime, if there are no oncoming and

leading vehicles existing in front of one’s vehicle, headlights

are preferred to be controlled to high beam to ensure visibility.

On the other hand, if oncoming or leading vehicle appears in

front, in order not to dazzle the driver, the headlight must be

dipped to low beam. Automating this operation would reduce

the driver’s operation error, leading to safe driving.

In daytime, vehicle detection is based on the shape of the

vehicles [3]–[8]. However, because of its low visibility at night-

time, it is difﬁcult to recognize shape; hence, detection at night-

time is based on the headlights and taillights of the vehicles,

which have high luminance [9]–[17]. In previous studies, to

segment the blobs, they proposed methods of binarization using

a simple threshold [9] or an automatic multilevel thresholding

using multilevel thresholds to the images [10]–[18]. In these

methods, however, under various ambient light conditions, it

may be difﬁcult to correctly segment the blob. In addition,

a scene that contains vehicles close to and far from one’s

vehicle may become difﬁcult to detect because the luminance

corresponds to the distance from the camera. Moreover, since

the headlights of onco ming vehicle an d the taillights of lead-

ing vehicle have different luminances, the detection becomes

more difﬁcult. In these researches, after detecting blobs using

multilevel thresholding, the blobs are classiﬁed based on their

size and shape, but as the threshold changes in frames, the

feature of size and shape would be different and would affect

the classiﬁcation. The examples of segmented im age using

Otsu thresholding, which automatically selects the appropriate

threshold for segmenting the image into two classes, are shown

in Fig. 1. The two class-segmented results of the original

images shown in Fig. 1(a) and (c) are shown in Fig. 1(b) and (d),

respectively. There are scenes that could extract lights as shown

in Fig. 1(b), but scenes in which ambient light conditions

change like in Fig. 1(c), it is difﬁcult to extract the lights of

the vehicle.

We focused on the response of Laplacian of Gaussian (LoG)

operator, to extract the bright region when compared with

the surroundings (contrast) without simply extracting the high

luminance area. The LoG operator detects objects brighter than

the surrounding area; hence, it is more robust than multilevel

thresholding, which uses the luminance of the whole image to

determine the threshold. To correspond to the various sizes of

the light of the vehicle, it is necessary to derive response from

multiple scales of LoG operator, but the calculation time is large

for practical use. Therefore, we use Center Surround Extremas

(CenSurE) [19], which uses an integral image for deriving

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

下载后可阅读完整内容，剩余9页未读，立即下载

jihong193

粉丝: 0
资源: 6

夜间驾驶场景下基于中心像素的车辆检测方法

com.sun.pdfview.PDFRenderer

Spire.Pdf 去水印+超10页+用例，Spire.Pdf.dll、Spire.License.dll。

数据模型与决策-第14章.pdf

com.itextpdf.kernel.pdf.PdfReader 获取pdf 总页数

python实现将文件夹“无括号“下的指定pdf”XMGJYH.PDF“复制5份，分别重命名为”XMGJYH1.PDF“”XMGJYH2.PDF“”XMGJYH3.PDF“”XMGJYH4.PDF“”XMGJYH5.PDF“，并保存在原文件夹

python 给以.pdf和.PDF结尾的文件进行重命名

org.apache.poi.xwpf.converter.pdf.PdfConverter 的maven依赖

字符串的格式如下：测试.pdf；fieldname*=utf-8''测试.pdf 使用java代码实现获取测试.pdf

pip install Spire.PDF

.net aspose.pdf.license

最新资源