A distributed video management cloud platform using hadoop

SPECIAL SECTION ON BIG DATA SERVICES AND COMPUTATIONAL INTELLIGENCE
FOR INDUSTRIAL SYSTEMS
Received November 15, 2015, accepted December 3, 2015, date of publication December 11, 2015,
date of current version December 22, 2015.
Digital Object Identifier 10.1109/ACCESS.2015.2507788
A Distributed Video Management
Cloud Platform Using Hadoop
XIN LIU1, DEHAI ZHAO2, LIANG XU2, WEISHAN ZHANG2, JIJUN YIN1, AND XIUFENG CHEN1
1Hisense TransTech, Ltd., Qingdao 266071, China
2Department of Software Engineering, China University of Petroleum, Qingdao 266580, China
Corresponding author: W. Zhang (zhangws@upc.edu.cn)
This work was supported by the International Science and Technology Cooperation Program of China under Grant 2013DFA10980.
ABSTRACT Due to complexities of big video data management, such as massive processing of large amount
of video data to do a video summary, it is challenging to effectively and efficiently store and process these
video data in a user friendly way. Based on the parallel processing and flexible storage capabilities of cloud
computing, in this paper, we propose a practical massive video management platform using Hadoop, which
can achieve a fast video processing (such as video summary, encoding, and decoding) using MapReduce,
with good usability, performance, and availability. Red5 streaming media server is used to get video stream
from Hadoop distributed file system, and Flex is used to play video in browsers. A user-friendly interface is
designed for managing the whole platform in a browser–server style using J2EE. In addition, we show our
experiences on how to fine-tune the Hadoop to get optimized performance for different video processing
tasks. The evaluations show that the proposed platform can satisfy the requirements of massive video data
management.
INDEX TERMS Hadoop, J2EE, video processing, MapReduce, video management.
I. INTRODUCTION
With the development of Internet of things (IOT), streaming
media applications such as security and traffic video surveil-
lance are deployed rapidly. Streaming media data contains
considerable useful information. These massive audio and
video data are unstructured data that traditional approaches
can’t handle storage and processing efficiently. The majority
of existing approaches use centralized storage,1 such as disk
array, which is not scalable for efficient massive video data
processing.
In order to alleviate these problems, researchers are begin-
ning to make use of the relatively new cloud comput-
ing technologies to explore video storage and processing.
For example, work on using Hadoop2 [1], [2] to accelerate
video processing time utilizing MapReduce programming
style [3], [4], and similar work called Split & Merge archi-
tecture was proposed in [5] for high performance video com-
pression. However, these existing work lack details on how
to design a versatile video management platform in a user
friendly way, and how to effectively use Hadoop to fine tune
performance of video processing in order to build a practical
and usable video cloud platform.
1http://nffinc.com/sites/default/files/Unified%20VSS%20white%
20paper.pdf
2http://hadoop.apache.org/
In this paper, we propose a solution that seamlessly
integrates J2EE, and Flex with Hadoop to provide a user
friendly way for video management, and also we integrate an
open source media server to make our solution economical.
H.264 video files stored on HDFS [6], [7] are transcoded to
common video format using FFmpeg.3 Users can download
the transcoded files and play it with a video player directly.
Furthermore, Gaussian mixture model (GMM) is used to
implement video background subtraction. A web based inter-
face is designed using J2EE to provide users an easy way to
manage the platform, where Struts2, Spring and Hibernate are
integrated.
The contributions of this paper include:
• We propose an universal video data management archi-
tecture that integrates J2EE technologies with Red5
media server, and Flex technology, running on top of
Hadoop platform.
• We share our experiences on implementing this archi-
tecture, and on fine tuning Hadoop in order to achieve
good video processing performance.
• We evaluate the platform for its concurrent IO perfor-
mance, file uploading and downloading, to show the
effectiveness of the proposed platform.
3https://www.FFmpeg.org/
VOLUME 3, 2015
2169-3536
2015 IEEE. Translations and content mining are permitted for academic research only.
Personal use is also permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
2637
www.redpel.com +917620593389

X. Liu et al.: Distributed Video Management Cloud Platform Using Hadoop
FIGURE 1. Deployment diagram of the platform.
The outline of the paper is as follows: section 2 presents
the architecture of the distributed cloud platform, with imple-
mentation details discussed in Section 3. Section 4 evaluates
the proposed video cloud platform. And section 5 presents
some related work. Conclusions and future work end the
paper.
II. ARCHITECTURE AND DESIGN OF THE
VIDEO MANAGEMENT PLATFORM
As discussed in the introduction, the video management
platform should have a browser server style which can
improve usability for normal users. Also it will make use of
Hadoop platform to make use of the parallel and distributed
computing and storage capabilities. The architecture of the
video platform described with a UML deployment diagram
is shown in Figure 1.
The distributed cloud platform mainly consists of
four parts: Hadoop clusters, Red54 streaming server clusters,
Web servers and database servers. Video stream from video
capture devices (e.g. cameras) is stored on HDFS via Web
server [8]. Large amount of continuous video data (e.g. in
smart surveillance applications) need a huge storage space,
but it’s not necessary to store all the video from cameras,
so some processing is done on the DataNode. For example,
video summary can extract specified video that we care about
and remove useless video that contain little information.
4http://www.Red5server.org/
The background subtraction can separate background and
foreground effectively, which makes it easier to implement
video summary. Meanwhile, video encoding and decoding is
an important method to ensure efficient video transmission.
Red5 streaming media server uses RTMP, RTMPT, and
RTMPE to play audio and video when a users’ request is
sent to a Web server. There are two important components
with a Web server. The first one is FFmpeg, which is a
transcoding tool to transcode RTSP stream from cameras to
RTMP stream. Another one is Flex, which can get real time
stream from a Red5 server. All user information is stored on
MySQl database servers. Finally, a user-friendly graphical
interface is designed for an easier operation in a browser.
The package diagram of the platform is shown in Figure 2.
We can see that the video cloud platform is using a typi-
cal three-tier architecture style. The Data package is used
for database accessing. The Server package contains service
logic components and there are two sub-packages: stream
server package and Web server package. The Processing
package is used for various video processing features which
depend on a OpenCV5 package. The User package sends
requests to services and the responses are displayed by GUIs
in a browser.
The class diagram of the platform is shown in Figure 3.
Cameras form a group of video capture devices. The View
class implements GUIs including displaying user information,
playing video, etc. The Management class is the core class
5http://opencv.org/
2638 VOLUME 3, 2015

FIGURE 2. Package diagram of the platform.
which plays a role of resource scheduling, user requests
response, video stream transcoding, log generating, etc. User
class is a set of user information and operation. Similarly,
the video class contains video information and operation.
Four main video processing classes of the platform are video
summary, background subtraction, encoding and decoding.
And each genetic class extends the video processing class
as its parent class. Lastly, the Database class stores various
data and information, like user login data, user group, video
time, etc.
III. IMPLEMENTING THE VIDEO CLOUD PLATFORM
There are three important modules in Hadoop: HDFS, YARN
and MapReduce. HDFS is the Hadoop distributed file system
which can provide high reliability, high scalability and high
throughput data services. YARN is responsible for unify man-
agement and scheduling of cluster resources. And MapRe-
duce is a distributed computing framework that provides
efficient,fault-tolerant, and scalable parallel processing capa-
bilities.
A. DISTRIBUTED STORAGE
HDFS uses a master-slave configuration which includes a
NameNode and a number of DataNodes. The NameNode
(master) is a central server, and it manages the file system
name space and clients’ access to files. The DataNode (slave)
is the storage space that stores various data. Additionally,
Zookeeper6 cluster is used to manage the active NameNode
and the standby NameNode to ensure reliable hot-swap
between two NameNode when the active NameNode is
6http://zookeeper.apache.org/
FIGURE 3. Class diagram of the platform.
VOLUME 3, 2015 2639

not working. HDFS has a strong system scalability which
supports adding or removing nodes dynamically. Whenever
a larger storage space is needed, new nodes can be added
online like nothing happens and data can be redistributed to
old and new nodes automatically, i.e., the storage space of
Hadoop cluster can be infinite large theoretically. Meanwhile,
HDFS stores files on blocks with limited size, which can be
adjusted according to your own configuration. So the storage
strategy of HDFS is to cut those large files to little pieces
and scatter them over the storage space in the cluster. HDFS
stores files on partition with automatic block replication to
implement high fault tolerance and its backup number can
also be configured.
To access video files stored in HDFS, users should login to
the platform through a client browser. Then they can manage
videos, like upload, download, and delete video files if they
have such rights. It’s hard for users to operate HDFS directly.
A user-friendly interface is designed using the browser server
architecture style, and J2EE technology is used to implement
it to provide a simple and easy operation environment. When
a user’ requests are sent to a Web server via http protocol,
this Web server will call APIs to send reading and writing
requests to the HDFS cluster. Then Web server receives the
response from HDFS and display it in the browser. As shown
in Figure 4, the Web server acts as a relay in this process.
FIGURE 4. Work flow of user requests.
When uploading video files to HDFS cluster, a logged
in user will upload video files to the the Web server, and
then a UploadMovieToHDFS() method is used to upload files
to the HDFS cluster. After this, the Web server deletes the
cached files. When downloading video files from a HDFS
cluster, the Web server queries files address information
and apply DownloadFromHDFS() method to get files from
the HDFS cluster. When deleting video files, the request is
sent to the Web server, then the persistence layer query the
file address information and apply DelMovieFromHDFS()
method to delete the file from the HDFS cluster.
B. VIDEO PLAYING
Red5 is an open source streaming media server [9] based
on Flash streaming media service [10] developed in Java.
It uses RTMP, RTMPT and RTMPE to play audio (MP3)
and video (FLV, MP4, etc.) [11], [12]. When the request
of playing video is sent by users, the Web server will
transcode RTSP stream from cameras to RTMP stream using
FFmpeg and send it to Red5 streaming media server. Then
the Web server can get real time video stream from Red5
streaming media server using Flex. The process is shown
in Figure 5.
FIGURE 5. Work flow of video transcoding.
The transcoded video stream from camera is stored on
HDFS. A mount is made using fuse-dfs between Red5
streaming media server and HDFS cluster and a directory
named Streams for the mount directory is created. The
process is shown in Figure 6.
FIGURE 6. Work flow of video playing.
The Red5 streaming media server scans FLV files
from the Streams directory and transcodes them to video
stream. The Web server receives video stream from the Red5
streaming media server and plays it in a browser.
C. BACKGROUND SUBTRACTION
One of the most important video processing task is back-
ground subtraction. There are lots of algorithms to achieve
this. The mixture of Gaussians (MOG) is used widely. We use
an improved Gaussian model designed by Zivkovic [13]. The
advantages of this algorithm are that the number of Gaussian
components is constantly adapted per pixel and that it can
perform shadow detection.
IV. EVALUATION
We have built a distributed cloud cluster to test our the video
cloud platform. The cluster contains a gigabit Ethernet switch
and five Dell XPS 8700-R37N8 computers with the label of
node1 to node5. Some necessary softwares such as OpenCV
and CUDA are installed. Four roles are distributed on these
five nodes. They are NameNode (NN), DataNode (DN),
Resourcemanager (RM) and Nodemanager (NM) as shown
in Table 1.
TABLE 1. Distribution of each role on the test bed.
A. PERFORMANCE EVALUATION
The first test is Hadoop benchmark test (system I/O test).
We use MapReduce to test the concurrent reading and writing
2640 VOLUME 3, 2015

TABLE 2. Results of Hadoop benchmark test.
of the platform in order to show the throughput performance
and HDFS I/O performance. In the test, we upload and down-
load a file of size 100G. The results are shown in Table 2.
Throughput is the average throughput of each Map task.
Average I/O rate is the average I/O rate of each file, and
exec time shows the total time of the whole process. We can
see clearly that for the reading process, it costs more time
when the backup number is three and it costs less time when
the backup number is one. But for the writing process, it
is a different situation. Therefor the backup number should
be adjusted according to actual needs. When the backup
number is fixed, the more concurrency there are, the larger
throughput and less processing time the platform can achieve.
It is important to note that the writing speed is higher than
the reading speed when the backup number is one, but the
reading speed is higher than the writing speed when the
backup number is three. The reason why we get this result is
the difference between writing method and reading method of
HDFS. The writing process is a pipeline, that is to say, when
writing a video file, the video file is segmented to blocks and
the blocks are segmented to smaller blocks called package.
The package is written on the first DataNode and copied to the
second DataNode from the first DataNode. Then the package
on the second DataNode is copied to the third DataNode and
so on. So the writing time increases with the increasing of
backup number. However, the reading process is parallel, that
is, the data blocks can be read from each DataNode which
stores them and combined as a complete video file. So the
reading speed will be higher if the backup number becomes
larger.
In addition, a test of uploading a file of size 36G with single
process has been done to verify the conclusion of Table 2 that
the writing speed is higher than the reading speed when the
backup number is one, but the reading speed is higher than the
writing speed when the backup number is three. The results
are shown in Table 3, with different number of backups.
B. TUNING THE PERFORMANCE
In this test, we will conduct a video summary with a
17.5 hours video file, whose size is 37G bytes with a frame
rate of 25 FPS and the resolution is 1280*720 pixels. There
TABLE 3. Results of backup number test.
are two approaches to do this. The first one is segmenting
raw video data to clips and using the clip names as the input
of map, then reading video files according to the clip address
and processing them in the map function. The other one is
also segmenting raw video data to clips, but using clip address
as key and video files as value, then combining them into
SequenceFile and processing it as the input of MapReduce.
The second one is a good choice due to potentials to reduce
data transmissions. We compare the above two approaches
and the results are shown in Table 4. We can see that for the
second approach, the performance is better than the first one.
TABLE 4. Results of video summary.
In a practical application, we should optimize the platform
for a better performance. In general, video processing
needs a lot of memory. Hadoop provides usemapreduce.
map.memory.mb and mapreduce.reduce.memory.mb to mod-
ify the memory of each map task and reduce task depending
on actual requirements.
There are more than 190 parameters in a Hadoop con-
figuration file and 25 of them have remarkable effect
on the efficiency of Hadoop applications. We test some
parameters which are very important for video processing.
They are:
• yarn.nodemanager.resource.memory-mb represents the
total physical memory that is available to YARN on one
node. YARN can not detect the total available physical
memory of one node automatically. So we should adjust
it according to the available memory resource of a node.
• dfs.block.size represents the block size on HDFS.
Meanwhile, it determines the number of maps in the
default MapReduce framework.
• dfs.replication represents the backup number. Increasing
backup number can improve map’s hit probability of
local data. However, this figure should be not be big for
big data processing applications.
The parameter optimization results are shown as follows,
in which Table 5 is a detail illustration of schema 2 in Table 4.
It can be seen obviously that the processing speed increases
with the increasing of memory. And Table 6 shows the effects
of different block size on processing time. We can see that it
has the shortest processing time when the block size is 64M,
VOLUME 3, 2015 2641

TABLE 5. Effects of different memory size.
TABLE 6. Effects of different block size.
because the video files have been segmented to 64M. So the
the clip size should be close to the HDFS block size to reduce
processing time.
V. DISCUSSION
We have encountered many problems during the development
of the video platform.
• During the development of the video data storage mod-
ule, we can not upload video files to HDFS cluster with
HDFS APIs directly. The solution to the problem is
uploading video files to the cache of Web server. Then
the video files will be upload to the HDFS cluster from
the Web server using HDFS APIs.
• For retrieving real-time streaming, we can not get
streaming data from IP cameras directly. The solution
to the problem is transcoding RTSP stream to RTMP
stream with FFmpeg. Then we use Flex to receive video
stream from the Red5 streaming media server.
• We find that Ajax is a good solution to enhance user
experience because it can optimize the transmission
between browser and Web server, and reduce data trans-
mission footprint and load time of Web server.
• The parameter setting of Hadoop can affect video
processing a lot, and we need to carefully adjust some of
the parameters in order to reduce the data transmissions,
and improve the performance.
VI. RELATED WORK
Myoungjin Kim et al. proposed a Hadoop-based distributed
video transcoding system in a cloud computing environment,
which can transcode various video format to MPEG-4 [14].
They design and implement the platform efficiently using
MapReduce framework. Comparing with their work, besides
transcoding, we develop a user-friendly interface to operate
HDFS easily, details on how to use Hadoop efficiently by
tuning different parameters are also presented.
In [15], the authors proposed a Hadoop-based storage
architecture for massive MP3 files. They used classification
algorithm in pre-processing module to merge small files into
sequence files. They confirm that the introduced efficient
indexing mechanism is a good solution to the problem of
small files. Our work focus on Hadoop-based storage and
processing architecture for big video data, and usability is one
of our focuses.
Lin et al. [16] proposed a prototype of cloud based video
recording system. The system can provide scalable video
recording, data backup and feature detection. They also
used HDFS to store video data, which is similar with ours.
In our work, we apply J2EE technology to develop a user-
friendly interface for HDFS, and we use Red5 to play video.
Additionally we consider fine tuning Hadoop is an important
task for practical big data applications.
In [17], the authors studied and analyzed the open source
Red5 streaming media server and they used Flex and Red5
streaming media server for playing video successfully. In our
research, we mainly concentrate on Hadoop based manage-
ment of video data using Flex and Red5 streaming media
server. In our work we combine Flex, Red5 and Hadoop
together to implement our platform.
Liu et al. [18] present a framework for video playing and
video storage based on Hadoop. Their framework provides
high availability services, which could support concurrent
access and playing streaming media in mobile terminals. Our
work is similar with their work, but we target a user friendly
big video data management platform and we also present
our experiences on using J2EE for the management of video
data, and experiences on optimizing the video management
platform with Hadoop.
In [19], the authors presented a video monitoring sys-
tem that can meet the users’ demands of searching video,
uploading video, downloading video and transcoding video.
They also used FFmpeg to transcode video. In our work,
we focused on the usability of the video management plat-
form, plus the experiences on integration of J2EE, Flex, and
Hadoop.
In our previous work [20], we did some preliminary
research on video monitoring platform, where Hadoop based
processing was explored. In this paper, we focus on a user
friendly video management platform with its architecture,
and sharing our experiences on fine tuning the performance
of Hadoop based video processing.
VII. CONCLUSIONS AND FUTURE WORK
In this paper, we present a distributed cloud platform for stor-
ing and managing massive video data based on the integration
of Hadoop, Red5 streaming media server and J2EE. The
platform is economical because the cluster can be built by per-
sonal computer, and the software packages are open source.
We have proposed to use Red5 streaming media technology to
realize the video playback while downloading. We have also
evaluated the performance of the platform extensively and it
shows that the platform has good performance. We have also
found that the parameter setting of Hadoop can greatly affect
the performance of the video processing.
We are experimenting to use the presented approach for
smart transportation system in Qingdao City, China. Addi-
tionally we are adopting an online processing and offline
processing combined approach [4] in order to make full of
the cloud computing potentials. Another direction is that we
plan to use deep learning to do knowledge mining from video
data [21].
2642 VOLUME 3, 2015

REFERENCES
[1] J. Venner, Pro Hadoop. New York, NY, USA: Apress, 2009.
[2] T. White, Hadoop: The Definitive Guide. Sebastopol, CA, USA:
O’Reilly Media, Inc., 2012.
[3] H. Tan and L. Chen, ‘‘An approach for fast and parallel video processing
on Apache Hadoop clusters,’’ in Proc. IEEE Int. Conf. Multimedia
Expo (ICME), Jul. 2014, pp. 1–6.
[4] W. Zhang, L. Xu, P. Duan, W. Gong, Q. Lu, and S. Yang, ‘‘A video cloud
platform combing online and offline cloud computing technologies,’’ Pers.
Ubiquitous Comput., vol. 19, no. 7, pp. 1099–1110, Oct. 2015.
[5] R. Pereira, M. Azambuja, K. Breitman, and M. Endler, ‘‘An architecture
for distributed high performance video processing in the cloud,’’ in Proc.
IEEE 3rd Int. Conf. Cloud Comput. (CLOUD), Jul. 2010, pp. 482–489.
[6] X. Wang and J. Su, ‘‘Research of distributed data store based on HDFS,’’ in
Proc. 5th Int. Conf. Comput. Inf. Sci. (ICCIS), Jun. 2013, pp. 1457–1459.
[7] S. A. Weil, S. A. Brandt, E. L. Miller, D. D. E. Long, and C. Maltzahn,
‘‘Ceph: A scalable, high-performance distributed file system,’’ in Proc. 7th
Symp. Oper. Syst. Design Implement., 2006, pp. 307–320.
[8] F. Azzedin, ‘‘Towards a scalable HDFS architecture,’’ in Proc. Int. Conf.
Collaboration Technol. Syst. (CTS), May 2013, pp. 155–161.
[9] G. Shi, Z.-P. Li, and Z.-L. Xu, ‘‘Design and implementation of real-time
video streaming system based on Red5,’’ Instrum. Technol., vol. 6, p. 6,
Jun. 2010.
[10] J. Emigh, ‘‘New Flash player rises in the Web-video market,’’ Computer,
vol. 39, no. 2, pp. 14–16, Feb. 2006.
[11] M. Lu and X. D. Wang, ‘‘Design of video conference system based on
techniques of FLV streaming media,’’ J. Jilin Univ. Inf. Sci. Ed., vol. 28,
no. 2, pp. 186–190, 2010.
[12] D. Wang and K. Xu, ‘‘Red5 Flash server analysis and video call ser-
vice implementation,’’ in Proc. IEEE 2nd Symp. Web Soc. (SWS), 2010,
pp. 397–400.
[13] Z. Zivkovic, ‘‘Improved adaptive Gaussian mixture model for background
subtraction,’’ in Proc. 17th Int. Conf. Pattern Recognit. (ICPR), vol. 2.
Aug. 2004, pp. 28–31.
[14] M. Kim, Y. Cui, S. Han, and H. Lee, ‘‘Towards efficient design and
implementation of a Hadoop-based distributed video transcoding system in
cloud computing environment,’’ Int. J. Multimedia Ubiquitous Eng., vol. 8,
no. 2, pp. 213–224, 2013.
[15] X.-Y. Zhao, Y. Yang, L.-L. Sun, and Y. Chen, ‘‘Hadoop-based storage
architecture for mass MP3 files,’’ J. Comput. Appl., vol. 32, no. 6,
pp. 1724–1726, 2012.
[16] C.-F. Lin, S.-M. Yuan, M.-C. Leu, and C.-T. Tsai, ‘‘A framework for
scalable cloud video recorder system in surveillance environment,’’ in
Proc. 9th Int. Conf. Ubiquitous Intell. Comput., 9th Int. Conf. Auto. Trusted
Comput. (UIC/ATC), Sep. 2012, pp. 655–660.
[17] D. Wang and K. Xu, ‘‘Red5 Flash server analysis and video call service
implementation,’’ in Proc. IEEE 2nd Symp. Web Soc. (SWS), Aug. 2010,
pp. 397–400.
[18] C. Liu, K. Fan, Z. Yang, and J. Xiu, ‘‘A distributed video share system
based on Hadoop,’’ in Proc. IEEE 3rd Int. Conf. Cloud Comput. Intell.
Syst. (CCIS), Nov. 2014, pp. 587–590.
[19] Y. Liu, S. Liang, M. Zhao, and Y. Sun, ‘‘Hadoop applications in video
monitoring,’’ in Advanced Materials Research, vols. 1030–1032. Dürnten,
Switzerland: Trans Tech Publications, 2014, pp. 1717–1720.
[20] W. Zhang, X. Ding, and R. Hou, ‘‘Hadoop and PaaS collaborative prac-
tice research in video monitoring platform,’’ in Applied Mechanics and
Materials, vols. 543–547. Dürnten, Switzerland: Trans Tech Publications,
2014, pp. 3549–3554.
[21] W. Zhang, P. Duan, Z. Li, Q. Lu, W. Gong, and S. Yang, ‘‘A deep
awareness framework for pervasive video cloud,’’ IEEE Access, vol. 3,
pp. 2227–2237, Nov. 2015.
XIN LIU is currently an Assistant Chief Engi-
neer with Hisense TransTech Company, Ltd. His
main research interests include smart transporta-
tion, video processing, and smart cities.
DEHAI ZHAO is currently pursuing the master’s
degree with a focus on big data processing sys-
tems, food recognition using deep learning, and
big data mining topics. He has published two
papers on these areas.
LIANG XU is currently pursuing the master’s
degree with a focus on big data processing sys-
tems, food recognition using deep learning, and
big data mining topics. He has published five
papers on these areas, including reputed journal
like Personal and Ubiquitous Computing.
WEISHAN ZHANG was a Research Associate
Professor/Senior Researcher with the Computer
Science Department, University of Aarhus (till
2010), where he had been working on the EU
FP6 Hydra pervasive middleware project, for
which he was a Technical Manager (2008–2009).
He was a Visiting Scholar with the Depart-
ment of Systems and Computer Engineering, Car-
leton University, Canada (2006–2007). He was
an Associate Professor with the School of Soft-
ware Engineering, Tongji University, Shanghai, China (2003–2007). He
was a NSTB Post-Doctoral Research Fellow with the Department of
Computer Science, National University of Singapore (2001 to 2003).
He is currently a Full Professor, the Deputy Head of research with
the Department of Software Engineering, School of Computer and
Communication Engineering, China University of Petroleum. He is the
Director of the Big Data Intelligent Processing Innovation Team of
Huangdao District, the Director of the Smart City Research Center
of Fuwode Electronic Corporation, and the Founding Director of the
Autonomous Service Systems Laboratory. He has an h-index of 13, accord-
ing to Google Scholar, and the number of total citations is around 490 in
2015.
JIJUN YIN is currently a Researcher with Hisense
TransTech Company, Ltd. His main research inter-
ests include smart transportation, video process-
ing, and smart cities.
XIUFENG CHEN received the Ph.D. degree from
the University of Jilin, in 2013. He holds a post-
doctoral position with Hisense TransTech Com-
pany, Ltd., Qingdao, China. His research inter-
ests include traffic system analysis and intelligent
transportation technology.
VOLUME 3, 2015 2643

A distributed video management cloud platform using hadoop

Recommended

More Related Content

Similar to A distributed video management cloud platform using hadoop (20)

More from redpel dot com (20)

Recently uploaded (20)

A distributed video management cloud platform using hadoop