978-1-5090-0654-0/16/$31.00 ©2016 IEEE 615 ICALIP 2016
RECOGNITION
OF INDIVIDUAL OBJECT IN FOCUS PEOPLE GROUP
BASED ON DEEP LEARNING
Liu Hui-bin
，
Wu Fei
，
Chen Qiang
，
Pan Yong
School of Electronic and Electrical Engineering, Shanghai University of Engineering Science,
Shanghai 201620, China
ABSTRACT
Deep leaning has become a hot research topic with the
rapid development of big data technology. As an
important branch of deep learning, convolutional neural
network has been widely used in image recognition, and
has achieved great success. Convolutional architecture for
fast feature embedding (Caffe) with features like speed,
extendibility and openness is currently top popular tool of
deep learning. In this paper, the authors use Caffe to
realize the recognition of individual object in a focus
people group. The training images can be obtained from
the video recorded by the camera through the method of
normalized cross-correlation histogram. The experimental
results show that the individual object can be matched
accurately by using pre training model. It can be used in
practical work like attendance system, criminal
investigation field etc.
Index Terms—deep learning, CNN, focus group,
object recognition, Caffe
1. INTRODUCTION
In the traditional machine learning, select a good feature
is the key to the success of the pattern recognition. When
processing unprocessed data, the ability of the traditional
machine learning is very limited, so it is called shallow
learning. Neural network algorithm simulating biological
neural network is a kind of pattern matching algorithm,
usually used for solving classification and regression
problems. It is a huge branch of machine learning, having
hundreds of different algorithms, which deep learning
belongs to. Deep learning is a kind of machine learning
based on unsupervised feature learning and hierarchical
structure, by building models with multiple hidden layers
and training massive data to learn more useful features,
for the purpose of improving the accuracy of recognition
and prediction. In recent years, the rapid development of
technology in the big data field makes deep learning to
become a hot research topic. As an important branch of
deep learning, convolutional neural network has been
widely used in image recognition, image classification,
and has achieved great success.
In this paper, the author capture videos of every
individual object in a focus people group then get the
images whose difference between adjacent frames
exceeds a threshold using the method of normalized cross-
correlation of histograms. Parts of them are used as
training images in convolutional neural network, and
others not been trained are used as matching images to
match individual object by being given the similarity of
every people by the pre training model. Pre training and
object matching are implemented in Caffe which is the
fast feature embedded convolutional neural network
framework. At the end of the paper, the experimental
results and the social value of this research are given.
2. DEEP LEARNING TOOL
2.1. Neural network and Deep learning
Neural network is a mathematical model of distributed
parallel information processing by imitating biological
neural network, which is used to solve the problem of
recognition and regression [1]. A neural network consists
of many connected neurons. Overall function of neural
network depends not only on the characteristics of single
neurons, also on interactions and connections between
neurons. Neurons can be representation of different
objects, such as features, letters, concepts, and so on.
Neural network processing unit can be divided into three
categories: input layer, output layer and hidden layer.
Input layer is connected with the external signal and data,
and output layer achieve the output of the system
processing results. Hidden layer is between input layer
and output layer, which can not be observed by the
external system. The connection weights between neurons
control the strength of the connections between the units.
The information processing ability of the whole system is
embodied in the connection of each processing unit in the
neural network. Figure 1 shows a neural network with one
single hidden layer. Neural network is a huge branch of
machine learning, there are hundreds of different
algorithms, and deep learning is one of the algorithms.
In 2006, Professor Geoffrey Hinton of the University
of Toronto published a paper in "science" [2], which
expresses two main points of deep learning: a neural
network with multi hidden layers has excellent ability of
learning features, and the learning features of the data are
more essential, so as to facilitate the visualization and
recognition; the difficulty of training the deeper neural
network can be solved by initializing the layers one by
one through unsupervised learning. In essence, deep
learning is a kind of machine learning based on
unsupervised feature learning and hierarchical structure,
资源评论
