没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
试读
15页
在对象检测中,检测具有100个像素的对象与检测具有10个像素的对象有很大不同。 许多物体检测算法都假设行人规模在检测过程中是固定的,例如DPM检测器。 但是,检测器在不同规模的情况下往往会产生不同的检测效果。 如果使用检测器以不同比例执行行人检测,则可以提高行人检测的准确性。 提出了一种多分辨率DPM行人检测算法。 在模型训练阶段,将分辨率因子添加到潜在SVM模型的一组隐藏变量中。 然后,在检测阶段,将标准DPM模型用于高分辨率对象,对于低分辨率对象则采用刚性模板。 在我们的实验中,我们发现在低分辨率对象的情况下,标准DPM模型的检测精度低于刚性模板的检测精度。 在加州理工学院,多分辨率DPM检测器的遗漏率为52%,每个图像有1个假阳性(1FPPI); 就标准DPM检测器而言,遗漏率上升到59%(1FPPI)。 在加州理工学院的大型样本集中,多分辨率和标准DPM检测器给出的遗漏率分别为18%(1FPPI)和26%(1FPPI)。
资源推荐
资源详情
资源评论
Journal of Computer and Communications, 2017, 5, 102-116
http://www.scirp.org/journal/jcc
ISSN Online: 2327-5227
ISSN Print: 2327-5219
DOI: 10.4236/jcc.2017.59007
July 17, 2017
An Improvement of Pedestrian Detection
Method with Multiple Resolutions
Guodong Zhang
1
, Peilin Jiang
2
, Kazuyuki Matsumoto
1
, Minoru Yoshida
1
, Kenji Kita
1
1
Faculy and School of Engineering, Tokushima University, Tokushima, Japan
2
Xi’an Jiaotong University, Xian, China
Abstract
In object detection, detecting an object with 100 pixels is substantially diffe
r-
ent from detecting
an object with 10 pixels. Many object detection algorithms
assume that the pedestrian scale is fixed during detection,
such as the DPM
detector. However, detectors often give rise to different detection effects u
n-
der the circumstance of different scales. If a detector is used to perform ped
e-
strian detection in different scales, the accuracy of pedestrian detection could
be improved. A multi-resolution DPM pedestrian detection algorithm is pr
o-
posed in this paper.
During the stage of model training, a resolution factor is
added to a set of hidden variables of
a latent SVM model. Then, in the stage of
detection, a standard DPM model is used for the
high resolution objects and a
rigid template is adopted in case of the low resolution objects. In our exper
i-
ments, we
find that in case of low resolution objects the detection accuracy of
a standard DPM model is lower than that of
a rigid template. In Caltech, the
omission ratio of a multi-resolution DPM
detector is 52% with 1 false positive
per image (1FPPI); and the omission ratio rises to 59% (1FPPI)
as far as a
standard DPM detector is concerned. In the large-scale sample set of
Caltech,
the omission ratios given by the multi-resolution and the standard DPM d
e-
tectors are 18% (1FPPI) and 26% (1FPPI), respectively.
Keywords
Deformable Part Model, Pedestrian Detection, Multi-Resolution; Latent SVM
1. Introduction
Pedestrian detection has been a hotspot in computer vision research [1]. The
corresponding detection algorithm has been developed towards high precision
and instantaneity [2] [3]. For a driverless automobile, the usage of which has
How to cite this paper:
Zhang, G.D.,
Jiang,
P.L.,
Matsumoto, K., Yoshida, M. and Kita
,
K.
(2017)
An Improvement of Pedestrian
Detection Method
with Multiple Resolu-
tions
.
Journal of Computer and Comm
u-
nications
,
5
, 102-116.
https://doi.org/10.4236/jcc.2017.59007
Received:
October 10, 2016
Accepted:
July 14, 2017
Published:
July 17, 2017
Copyright © 201
7 by authors and
Scientific
Research Publishing Inc.
This work is licensed under the Creative
Commons Attribution International
License (CC BY
4.0).
http://creativecommons.org/licenses/by/4.0/
Open Access
G. D. Zhang et al.
103
become popular nowadays, its intelligent system should be able to detect the lo-
cations and quantities of pedestrians ahead, to analyze the road conditions, and
to guarantee the safety of these pedestrians [4]. For such cases, the pedestrian
detection is an inevitable procedure. The pedestrian detection problem is diffi-
cult because that the target people often have various characteristics and the
surrounding environments also change frequently [5].
The pedestrian sizes in real world are different from each other. Besides the
height diversity of different people, many imaging differences are incurred by
the different distances between people and the camera.
Figure 1 shows a high
resolution corresponds to the large pedestrian scale and a low resolution corres-
ponds to the small pedestrian scale in the process of pedestrian detection.
Pedestrians contain rich information in the case of high resolution [6], and it
is more likely for them to be detected. Even if they are locally overlapped, many
algorithms have the capability to detect these targets [7]. However, in the case
low resolution, the pedestrians which contain a small amount of information
cannot be detected easily. Meanwhile, low resolution pedestrians are very vul-
nerable to the interferences of the surrounding environments. In most cases, a de-
tection algorithm has a much better detection result for the high resolution pede-
strians than that for the low resolution pedestrians. Dalal and Triggs [8] proposed
a HOG detector. If the detection window is fixed to
128 64×
pixels during
training and detection, this detector can generate good effects at the time of de-
tecting pedestrians with pixels greater than
128 64×
. However, when the target
pedestrians are smaller than
128 64×
, the detector almost fails to detect any pe-
destrian. Although the target can be increased to larger than
128 64×
pixels by
means of interpolation, the detection accuracy is still brought down. The DPM
pedestrian detector makes use of a root filter and several part filters to describe the
pedestrians. Information in the pedestrians of high resolution is sufficient.
Figure 2(a) and Figure 2(b) are results obtained by utilizing a standard DPM
Figure 1. The pedestrians with Multiple resolution in a sample picture.
G. D. Zhang et al.
104
(a) (b)
Figure 2. The detection result of standard DPM. (a) The part filter and root filter in DPM
(b) The detection result of standard DPM.
detector to detect pedestrians in
Figure 1. It is obvious that the small-scale pe-
destrians cannot be detected successfully. Therefore, the overall detection effect
can be improved if we can improve the detection effect for low resolution pede-
strians and prevent affecting the detection effect for high resolution pedestrians.
In this paper, we propose a multi-resolution DPM pedestrian detection algo-
rithm, which takes advantage of the standard DPM framework in training the
pedestrian with the resolution factor as a hidden variable. For the high resolu-
tion pedestrians, the response can be figured out in the first place. And its loca-
tion can be estimated with the combination of this high resolution response and
the response under a corresponding low resolution. However, for the low resolu-
tion pedestrians, the judgment over possible locations of these targets is carried
out by only calculating the responses under the low resolution. High resolution
and low resolution are only intuitive concepts in the common sense. In addition,
resolution is closely associated with the heights of pedestrian samples.
Structure of this paper is as follows. In section 2, we thoroughly illustrate the
DPM model for pedestrian detection, depicts the DPM learning algorithm, and
describe the parameter initialization and the training procedures. In section 3,
we illustrate the improved DPM algorithm in case of multi-resolution targets, by
analyzing the features of pedestrian detection under multi-resolution, and de-
scribing the improved multi-resolution DPM pedestrian detection algorithm in
detail. In Section 4, we apply this improved algorithm to a general dataset to
comparatively analyze the experimental results.
2. Overview of Related Theory
2.1. Deformable Part Model
The deformable part model (DPM) consists of a root filter and several part filters
to describe the pedestrians. Specifically, the root filter describes each pedestrian
as a whole, while each part filters describe a part of the pedestrian, such as the
head and hand [9]. In this way, the constructed model can effectively capture the
pedestrian information, and adapt well to the changes of body posture and
dressing of the pedestrian [10].
剩余14页未读,继续阅读
资源评论
weixin_38651165
- 粉丝: 4
- 资源: 901
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功