2013 Annual Report of Advanced Networked Environment Division, Cybermedia Center, Osaka University, Japan

This web page is the 2013 annual report of Advanced Networked Environment Division, Cybermedia Center, Osaka University (Ubiquitous Network Laboratory, Graduate School of Information Science and Technology, Osaka University / Matsuoka Laboratory).

Japanese version of this web page is here.

Table of contents

  1. Members
    1. Laboratory staffs
    2. Joint researchers
    3. Students
  2. Achievements
    1. Research on green data centers
      1. An evaluation of the impact of task allocation on power consumption in data centers
      2. Performance evaluation of task assignment method for reducing air conditioning power consumption at data centers
      3. Performance evaluation of HEMS over ASP service utilizing websocket protocol
      4. Robust virtual machine arrangement method for multi-tenant datacenter inspired by degeneracy in biological networks
    2. Research on overlay network architectures
      1. Reduction of inter-ISP transit cost caused by overlay routing
      2. Distributed and conflict-aware measurement based on local information exchange in overlay networks
      3. Experimental evaluation of an available bandwidth measurement method for arbitrary parts of end-to-end path
    3. Research on architecture of transport layer for next-generation high-speed network
      1. Analysis of power consumption in TCP data transmission over a wireless LAN environment
      2. Improvement of end-to-end bandwidth measurement method for power-saving routers
    4. Research on wireless networks
      1. Performance evaluation of heuristic algorithms for time slot assignment in IEEE 802.16j relay networks
      2. Performance study of IEEE 802.16j relay networks by multiple regression analysis
      3. Energy efficient information dissemination scheme based on the received signal strength in wireless sensor networks
      4. Performance evaluation of tunnel aggregation method for reducing control plane cost in mobile core networks
    5. Research on pedestrian counting
      1. Counting pedestrians passing through a line in crowded scenes by extracting optical flows
      2. Pedestrian counting based on the number of salient points considering non-linear effect of occlusions
      3. Pedestrian counting based on Monte Carlo method with binary sensors
  3. Publications
    1. Journal papers
    2. International conference papers
    3. Theses
      1. Ph.D. Theses
      2. Masters' Theses
      3. Bachelors' Theses

1. Members

1.1. Laboratory staffs

Morito Matsuoka
Morito Matsuoka
Professor
Go Hasegawa
Go Hasegawa
Associate Professor
Yoshiaki Taniguchi
Yoshiaki Taniguchi
Assistant Professor
Mari Onishi
Mari Onishi
Secretary

1.2. Joint researchers

Masayuki Murata
Masayuki Murata
Osaka University
Professor
Norihiro Hagita
Norihiro Hagita
ATR IRC
Director
Masazumi Ueba
Masazumi Ueba
Muroran Inst. Tech.
Professor
Norimichi Ukita
Norimichi Ukita
NAIST
Associate Professor
Harumasa Tada
Harumasa Tada
Kyoto Univ. of Edu.
Associate Professor
Yutaka Nakamura
Yutaka Nakamura
Osaka University
Associate Professor

1.3. Students

Daisuke Kobayashi
Daisuke Kobayashi
Doctoral Course 2nd
Wataru Toorisaka
Wataru Toorisaka
Doctoral Course 1st
Yusuke Iijima
Yusuke Iijima
Master Course 2nd
Shoichi Takagi
Shoichi Takagi
Master Course 2nd
Yuki Tsubouchi
Yuki Tsubouchi
Master Course 2nd
Kohei Higo
Kohei Higo
Master Course 2nd
Shuto Fujii
Shuto Fujii
Master Course 2nd
Miki Mizushima
Miki Mizushima
Master Course 2nd
Kai Koyama
Kai Koyama
Master Course 1st
Shun Sakurai
Shun Sakurai
Master Course 1st
Kazuyuki Hashimoto
Kazuyuki Hashimoto
Master Course 1st
Kouji Suganuma
Kouji Suganuma
Bachelor Course 4th
Ryota Soe
Ryota Soe
Bachelor Course 4th
Takaaki Deguchi
Takaaki Deguchi
Bachelor Course 4th
Qiling Tong
Qiling Tong
Bachelor Course 4th
Akira Morimoto
Akira Morimoto
Bachelor Course 4th

2. Achievements

2.1. Research on green data centers

2.1.1. An evaluation of the impact of task allocation on power consumption in data centers

An evaluation of the impact of task allocation on power consumption in data centers

In data centers, the power consumption of air conditioners and that of fans in the server are affected by the heat distribution in the data center. Since server task distribution in data centers affects to the heat distribution, the total power consumption in data centers varies according to the task distribution even if the total amount of tasks remains unchanged. In this report, we evaluate the impact of task allocation on power consumption in data centers. We calculate the total power consumption in data centers based on the observation results obtained through real data center experiments. Through numerical evaluations using various data center models, we show that the power consumption in a tandem data center model is decreased by 26 % compared to that in a traditional data center model.

2.1.2. Performance evaluation of task assignment method for reducing air conditioning power consumption at data centers

Performance evaluation of task assignment method for reducing air conditioning power consumption at data centers

In this research, we propose a server task assignment method for leveling the temperature distribution of data center servers to reduce air conditioning power consumption. Our method is based on the simulation of data center's airflow to find a server with the highest temperature and that with the lowest temperature and moves the server's task from the former server to the latter server. Evaluation results show that we can decrease the power consumption of air conditioners by 9.6% and 45.6% for the traditional data center model and the novel data center model developed by our research group, respectively. Furthermore, by considering the effect of waste heat utilization, the total power consumption in our novel data center model can be decreased by up to 22.4%.

2.1.3. Performance evaluation of HEMS over ASP service utilizing websocket protocol

Performance evaluation of HEMS over ASP service utilizing websocket protocol

In our previous research, we realize a Home Energy Management System (HEMS) as an ASP service with the Websocket protocol, which can provide Web-based bidirectional communication channels. We showed that the proposed architecture can reduce the server load and the network traffic significantly. However, the performance evaluation has been conducted by only simple mathematical analysis. In this research, we designed the cloud-based HEMS assuming energy control from HEMS server to Air conditioners and LED lightings in households, and evaluated its performance by implementation experiments. We found that the memory usages of the HEMS server increase in nearly proportion to the number of households. On the other hand, CPU utilization and end-to-end message delay fluctuate unpredictably according to the number of CPU cores. We concluded that the usage of multi-core processor largely affects the performance of HEMS server.

2.1.4. Robust virtual machine arrangement method for multi-tenant datacenter inspired by degeneracy in biological networks

Robust virtual machine arrangement method for multi-tenant datacenter inspired by degeneracy in biological networks

In this research, we propose a virtual machine deployment method for datacenter inspired by degeneracy found in biological networks. The main idea is to exploit the concept of networked buffering observed in biological networks, , providing many-to-many redundancy with small amount of additional resources, unlike the one-to-one or one-to-many redundancy. We evaluate the fundamental characteristics of the proposed method by computer simulation, and show that the proposed virtual machine deployment method can largely improve the recovery performance from large-scale failures and reduce additional resources compared with existing methods.

2.2. Research on overlay network architectures

2.2.1. Reduction of inter-ISP transit cost caused by overlay routing

Reduction of inter-ISP transit cost caused by overlay routing

The current Internet consists of a numerous number of Internet Service Providers (ISPs), each of which operates own network to maximize its own benefit. Each ISP configures the IP-level routes for the network traffic based on monetary cost of inter-ISP links. In recent years, on the other hand, new types of traffic routing mechanisms are getting much attention, which largely impact on the ISPs' cost structure. Application-level traffic routing is one of such mechanisms. Because the application-level traffic routing is conducted by end-to-end approach without any care of the monetary costs of inter-ISP links, it can be harmful to the ISPs' cost structure. Content-centric networking (CCN), which routes packets based on content name, also largely affects to the ISPs' cost structure. On the contrary to application-level traffic routing, CCN brings positive effect on inter-ISP transit cost due to its in-network caching mechanism, because that it can reduce the traversing traffic on transit links by replying the cached contents when there is a cache hit. However, CCN is not developed considering ISPs cost structure directly, and in-network caching of CCN does not help reducing the transit cost of ISPs with peering relationships.

This research begins by developing a method to reduce transit cost of application-level traffic routing conducted by individual end user. To determine the relationships among ASes, which are required for ISP cost-aware routing, we first construct a method to estimate the transit cost of application-level paths from end-to-end network performance values.

In the next part of this research, we aim to realize an application-level traffic routing conducted by individual operators in multiple ISPs in a distributed fashion. In the proposed method, we assume that the operators of application-level traffic routing cooperate with each other on their route selection, so that we can avoid performance degradation caused by route overlaps due to their selfish decisions. We then propose an application-level traffic routing method based on distributed simulated annealing to obtain near-optimal solutions to the problem.

The third part of this research, we focus on the in-network caching mechanism in CCN and propose a new mechanism to reduce the transit cost by cache sharing mechanism, that is, cooperative cache sharing among CCN routers in multiple ISPs. It aims to improve cache hit ratio, which leads the further reduction in the inter-ISP transit cost. In the proposed method, the CCN routers share the memory space for content caching.

[Related paper]

  1. Kazuhito Matsuda, Go Hasegawa and Masayuki Murata, ``An application-level routing method with transit cost reduction based on a distributed heuristic algorithm,'' IEICE Transactions on Communications, vol.E96-B, no.6, pp.1481-1491, June 2013. [paper]

2.2.2. Distributed and conflict-aware measurement based on local information exchange in overlay networks

Distributed and conflict-aware measurement based on local information exchange in overlay networks

Measuring network resource information, including available bandwidth, propagation delay, and packet loss ratio, is an important task for efficient operation of overlay network services. Although measurement accuracy can be enhanced by frequent measurements, performing measurements with high frequency can cause measurement conflict problem that increases the network load and degrades measurement accuracy. In this research, we propose a low-cost, distributed and conflict-aware measurement method that reduces measurement conflicts while maintaining high measurement accuracy. The main idea is that the overlay node exchanges the route information and the measurement results with its neighboring overlay nodes while decreasing the measurement frequency. This means our method trades the overhead of conducting measurements for the overhead of information exchange to enhance measurement accuracy. Simulation results show that the relative error in the measurement results of our method can be decreased by half compared with the existing method when the total measurement overheads of both methods are equal. We also confirm that exchanging measurement results contributes more to the enhancement of measurement accuracy than performing measurements.

[Related paper]

  1. Dinh Tien Hoang, Go Hasegawa and Masayuki Murata, ``A distributed measurement method exploiting path overlapping in large scale network systems,'' in Proceedings of the 1st Workshop on Large Scale Network Measurements (31st NMRG meeting), Zurich, Switzerland, Oct. 2013. [slide]
  2. Dinh Tien Hoang, Go Hasegawa and Masayuki Murata, ``Monitoring available bandwidth in overlay networks using local information exchange,'' in Proceedings of the 2013 Australasian Telecommunications Networks and Applications Conference (ATNAC 2013), pp.172-177, Christchurch, New Zealand, Nov. 2013. [paper]
  3. Dinh Tien Hoang, ``Distributed solution approaches for large-scale network measurement exploiting local information exchange.'' Ph.D. Thesis, Graduate School of Information Science and Technology, Osaka University, Jan. 2014.
  4. Yusuke Iijima, ``Measuring packet loss ratio on overlay networks based on spatial composition,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.
  5. Dinh Tien Hoang, Go Hasegawa and Masayuki Murata, ``A distributed mechanism for probing overlay path bandwidth using local information exchange,'' IEICE Transactions on Communications, vol.~E97-B, no.~5, May 2014. (to appear)
  6. Dinh Tien Hoang, Go Hasegawa and Masayuki Murata, ``Spatial and temporal solutions for fault diagnosis in large-scale network systems,'' submitted for publication.

2.2.3. Experimental evaluation of an available bandwidth measurement method for arbitrary parts of end-to-end path

Experimental evaluation of an available bandwidth measurement method for arbitrary parts of end-to-end path

The available bandwidth on an end-to-end network path is an important performance metric for various network control. Existing methods for measuring available bandwidth can only the available bandwidth value of the path and it cannot acquire the information of bottleneck location on the path, or the detailed hop-by-hop available bandwidth distribution of the path. Our research group has proposed a novel measurement method of the available bandwidth for arbitrary parts of an end-to-end path. In this research, we show the evaluation results of the proposed method with an experimental network environment. We present that the proposed method can measure the available bandwidth with less than 3.6% error even when the receiver-side available bandwidth is twice larger than the sender-side available bandwidth on the end-to-end path.

2.3. Research on architecture of transport layer for next-generation high-speed network

2.3.1. Analysis of power consumption in TCP data transmission over a wireless LAN environment

Analysis of power consumption in TCP data transmission over a wireless LAN environment

With recent developments in network technologies, wireless LANs (WLANs) have been installed widely and many mobile devices have been able to access the Internet everywhere everytime through WLANs. As a result, fairness among users and energy efficiency grow in importance in WLANs. In this research, in order to resolve these issues on transport-layer protocols and WLANs, we study on transport-layer solutions to achieve fairness and energy efficiency in WLAN environments. This research begins by proposing, designing, and implementing a transport-layer solution to alleviate unfairness among TCP flows in WLANs. To alleviate such unfairness, we present a transport-layer solution that TCP congestion control mechanisms are activated against congestion at an AP. Through computer simulations and experiments in real WLAN environments, the proposed method successfully achieves both fairness among upstream TCP flows and fairness between upstream and downstream flows regardless of vendor implementations of APs and wireless network interface (WNIs). The second part of this research deals with energy efficiency of wireless clients in WLANs. Energy consumption models are constructed for TCP data transfer over a WLAN in order to assess impacts of the TCP behavior on energy efficiency. Because of importance of the TCP behavior, the proposed models consist of the combination of two layer models: a MAC level model and a TCP-level model. In order to improve the sleep efficiency, TCP-level burst transmission behavior is proposed, which reduces the number of state transitions between active and sleep modes by transmitting multiple data packets in groups. By comparing the energy consumption between WNIs with CAM and with ideal sleep mode, numerical examples reveal the upper bound of energy savings under these sleep strategies. In particular, TCP-level burst transmission successfully reduces energy consumption with increasing moderate delay. To realize the energy saving by the above work, we need to alleviate uncoordinated behavior by multiple applications running on a single client, which produces the difficulty in control of sleep timing. For that purpose, as the last part of this research, SCTP tunneling is proposed and designed. SCTP tunneling has two features: flow aggregation and burst transmission. It aggregates multiple TCP flows into a single SCTP association between a wireless client and an AP to control packet transmission and reception timings. Furthermore, to enhance the sleep efficiency, SCTP tunneling reduces the number of state transitions by handling multiple packets in a bursty fashion. We extend the above energy consumption model and construct a mathematical model of the energy consumed by SCTP tunneling to assess its energy efficiency. Numerical examples based on the model show that the proposed method can reduce energy consumption by up to 69%. Implementation issues of SCTP tunneling are also discussed. As stated above, we propose, design, and implement transport-layer solutions to achieve per-flow fairness among TCP flows and energy efficiency for multiple TCP flows in a WLAN. Through examining the issues on both fairness and energy efficiency, this research shows that these solutions are feasible approaches and can be applied to real

[Related paper]

  1. Masafumi Hashimoto, Go Hasegawa and Masayuki Murata, ``SCTP tunneling: Flow aggregation and burst transmission to save energy for multiple TCP flows over a WLAN,'' IEICE Transactions on Communications, vol.E96-B, no.10, pp.2615-2624, Oct. 2013. [paper]

2.3.2. Improvement of end-to-end bandwidth measurement method for power-saving routers

Improvement of end-to-end bandwidth measurement method for power-saving routers

The increase in energy consumption associated with network traffic explosion is becoming a major problem. To realize energy-efficient networking, a number of researchers have focused on technologies that dynamically adjust the processing performance and the link speed of routers and switches according to the network traffic load. However, when such power-saving routers and switches are present in the network, the existing network protocols and control methods may not work well since they do not assume the existence of such routers and switches. We focuses on the measurement method of an end-to-end available bandwidth, which is an important performance metric for various network controls.

In this research, we introduce a parameter tuning method for Pathload, which is a popular tool for measuring the end-to-end available bandwidth, to maintain the measurement accuracy in the network with power-saving routers, while not affecting the behavior of power-saving routers. For that purpose, we first investigate the interactions between the bandwidth measurement behavior of Pathload and the energy efficiency of power-saving routers by conducting extensive simulation experiments. In the simulation experiments, we examine the effects of the power-saving router on the measurement accuracy and the effects of the Pathload measurement on the behavior of the power-saving router. Based on simulation results, we exhibit that both the measurement accuracy of Pathload and the energy efficiency of power-saving routers degrade, particularly when the power saving functions are triggered in short cycles.

As the second contribution of this research, we propose a parameter tuning method for Pathload that maintain measurement accuracy without affecting the behavior of power-saving routers. We accomplish this by giving simple mathematical analysis of the traffic load of Pathload on the power-saving router. We confirm the accuracy of our analysis by comparing the analysis results with corresponding simulation results. We also show that Pathload can measure an available bandwidth without affecting the behavior of power-saving routers by tuning the number of probing packets on the basis of the analysis results.

[Related paper]

  1. Daisuke Kobayashi, Go Hasegawa and Masayuki Murata, ``End-to-end bandwidth measurement method considering effects on power-saving routers,'' in Proceedings of the 2013 Australasian Telecommunications Networks and Applications Conference (ATNAC 2013), pp.112-117, Christchurch, New Zealand, Nov. 2013. [paper]

2.4. Research on wireless networks

2.4.1. Performance evaluation of heuristic algorithms for time slot assignment in IEEE 802.16j relay networks

Performance evaluation of heuristic algorithms for time slot assignment in IEEE 802.16j relay networks

In wireless networks, radio interference occurs when multiple nodes communicate on the same channel at the same time, that causes network performance degradation. In IEEE 802.16j relay networks, wireless communications are carried out based on TDMA where the wireless network resources are divided into multiple time slots and they are assigned to wireless links between relay nodes as transmission opportunities. The network performance is improved by decreasing the total number of different time slots assigned to all links in a single scheduling cycle, because it brings the increase in the transmission opportunities of the links per unit time. Although it can be achieved when multiple links utilize the same time slot, the capacity of such links is degraded due to the radio interference. On the other hand, since all links in the network need to have enough time slots to accommodate their traffic load, degrading the link capacity may increase the total number of different time slots in the scheduling cycle. Therefore, we should determine the time slot assignment by considering the above-mentioned tradeoff relationship. Furthermore, finding the optimum solution for time slot assignment is NP-hard problem, we require the efficient heuristic algorithms.

In this research, we propose heuristic algorithms for time slot assignment problem in IEEE 802.16j relay networks, and evaluate them through extensive simulation experiments. Two algorithms based on different heuristic are introduced. One algorithm assigns a set of time slots to links by a greedy approach. The other algorithm determines a set of links that use a time slot by a brute-force search for maximizing the total link capacity. Performance evaluation results exhibit that the proposed algorithms reduces around 34% and 39% of the total time slots compared with the case where no link utilizes the same time slot, respectively. Meanwhile, they also show that calculation time of the latter algorithm is longer than that of the former algorithm to reduce the total time slots. Thus, we show that there is a tradeoff between performance and calculation time.

[Related paper]

  1. Shoichi Takagi, Go Hasegawa, Yoshiaki Tanigchi, Hirotaka Nakano and Morito Matsuoka, ``Comparative evaluation of heuristics for time slot assignment in IEEE 802.16j relay networks,'' in Proceedings of the 7th International Conference on Mobile Computing and Ubiquitous Networking (ICMU 2014), pp.83-84, Bras Basah, Singapore, Jan. 2014. (Poster)
  2. Shoichi Takagi, ``Performance evaluation of heuristic algorithms for time slot assignment in IEEE 802.16j relay networks,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.

2.4.2. Performance study of IEEE 802.16j relay networks by multiple regression analysis

Performance study of IEEE 802.16j relay networks by multiple regression analysis

IEEE 802.16j relay networks (referred as relay networks below) provide the broadband wireless access environment by using wireless multi-hop communications. In the relay network, to avoid radio wave interference, Time Division Multiple Access (TDMA) is adopted for sharing wireless network resource, where time slots are assigned to wireless links between relay nodes as communication opportunities. The communication quality of links, which is evaluated by Signal to Interference and Noise Ratio (SINR), depends highly on time slot assignment since the link quality at a certain time slot is affected by interference signals from relay nodes which utilize the same time slot. Furthermore, when the adaptive modulation and coding is applied to the data transmission, the transmission bitrate is highly dependent on the SINR. The relay network performance is also influenced by various factors such as the background noise level, pass loss exponent, density of relay nodes, and transmission signal strength. Therefore, for assessing the performance of the relay networks, the effect of those network environment parameters should be deeply investigated.

In this research, we present results of throughput analysis of IEEE 802.16j relay networks. Specifically, we first summarize the network environment parameters that affect the throughput of relay networks. Then, we conduct extensive simulation experiments with various sets of those parameter values. From the simulation results, we investigate the relative effect of parameters on throughput performance by multiple regression analysis. Analysis results show that the background noise level, the path loss exponent, and the transmission signal strength are the key parameters determining the throughput of the relay networks. The analysis results also give the regression equation for estimating the throughput of the relay networks. In addition, we provide some insights for finding appropriate heuristic algorithms for time slot assignment based on the analysis results.

[Related paper]

  1. Kohei Higo, Go Hasegawa, Yoshiaki Tanigchi, Hirotaka Nakano and Morito Matsuoka, ``Multiple regression analysis of IEEE 802.16j relay network throughput,'' in Proceedings of the 16th International Conference on Advanced Communication Technology (ICACT 2014), pp.437-441, Pyeongchang, Korea, Feb. 2014.
  2. Kohei Higo, ``Performance study of IEEE 802.16j relay networks by multiple regression analysis,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.

2.4.3. Energy efficient information dissemination scheme based on the received signal strength in wireless sensor networks

Energy efficient information dissemination scheme based on the received signal strength in wireless sensor networks

In this research, we propose an information dissemination scheme to facilitate electric power consumption for wireless sensor networks. Our new dissemination scheme is based on the flooding and uses the receiver signal strength for determining the broadcast timing. We evaluated our proposed scheme by simulations. We used a network model with 100 ~ 100 square lattices, where sensor nodes are arranged on the square lattices. As a result, it is shown that our proposed scheme improves the information delivery ratio by increasing the power consumption slightly in the network. Furthermore, we ran simulations using the network model, where sensor nodes are randomly located and showed the proposed scheme had the same characteristics as simulation results using the network model, where sensor nodes are arranged on the square lattices.

2.4.4. Performance evaluation of tunnel aggregation method for reducing control plane cost in mobile core networks

Performance evaluation of tunnel aggregation method for reducing control plane cost in mobile core networks

Machine-to-Machine (M2M) communication has been paid much attention as a new communication paradigm to increasing wireless network traffic. When considering M2M devices are accommodated to mobile cellular networks, since the Average Revenue Per User (ARPU) for M2M devices is quite smaller than traditional cellular phones, the cost for accommodating M2M devices to the mobile cellular networks cannot be ignored for mobile network carriers. In this research, we apply a tunnel aggregation method to suppress the cost for accommodating M2M devices in the mobile cellular networks. The proposed method decreases the number of tunnels in mobile network system without compromising communication performances of terminals. We evaluate the performance of this method by mathematical analysis and show the basic characteristics. We also estimate the monetary cost for realizing the proposed method and reveal that it can reduce the cost by up to 80% compared with the traditional mobile core networks, while keeping the call blocking rate less than 0.1%.

[Related paper]

  1. Shun Sakurai, Go Hasegawa, Naoki Wakamiya and Takanori Iwai, ``Performance evaluation of tunnel sharing method for accommodating M2M communication in mobile cellular networks,'' in Proceedings of the International Workshop on Emerging Technologies for LTE-Advanced and Beyond-4G, Atlanta, USA, Dec. 2013.

2.5. Research on pedestrian counting

2.5.1. Counting pedestrians passing through a line in crowded scenes by extracting optical flows

Counting pedestrians passing through a line in crowded scenes by extracting optical flows

Pedestrian counting is needed in many situations such as safety control in airports and stations, traffic control in streets, and surveillance at event sites. For example, the number of people in an office building or a public facility can be used for safety management and energy saving control. This information can be obtained by monitoring the number of pedestrians passing through specific areas such as doors, passageways, and so on. Since manual counting requires labor costs and is difficult to apply in crowded scenes due to human errors, automatic counting has attracted a lot of attention. Especially, video processing-based counting methods have been widely studied since it has some advantages such as flexibility of camera placement and accuracy compared to other sensors. In video processing-based counting methods, counting in crowded scenes is an open issue since in crowded scenes occlusion occurs and that makes counting pedestrians difficult. Recently, for estimating the density of pedestrians in crowded scenes, a method based on optical flow extraction was proposed and the effectiveness of the method was shown. Therefore, there is a possibility that optical flow extraction is also effective for estimating the number of pedestrians in crowded scenes.

In this research, we propose a counting method of pedestrians passing through a specific line in the video sequences, called counting line, in crowded scenes based on optical flow extraction. In our proposed method, feature points are firstly detected from video sequences. Secondly, optical flows are extracted by detecting corresponding points of feature points in next frame. Then, optical flows whose segment crosses the counting line are considered as a valid optical flows for estimating the number of pedestrians. In addition, optical flows detected falsely or extracted from background regions are deleted. Finally, the number of pedestrians is estimated based on the number of optical flows using the pre-learned correlation between the number of optical flows and the actual number of pedestrians. Through experimental evaluations using actual video sequences, we show that the correlation coefficient between the number of optical flows and the number of pedestrians passing through the counting line is more than 0.835. We also show that the mean relative error of our proposed method is less than 0.068 for crowded scenes.

[Related paper]

  1. Miki Mizushima, Yoshiaki Tanigchi, Go Hasegawa, Hirotaka Nakano and Morito Matsuoka, ``Counting pedestrians passing through a line in video sequences based on optical flow extraction,'' in Proceedings of the 12th International Conference on Circuits, Systems, Electronics, Control & Signal Processing (CSECS 2013), pp.129-136, Budapest, Hungary, Dec. 2013. [paper]
  2. Miki Mizushima, ``Counting pedestrians passing through a line in crowded scenes by extracting optical flows,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.

2.5.2. Pedestrian counting based on the number of salient points considering non-linear effect of occlusions

Pedestrian counting based on the number of salient points considering non-linear effect of occlusions

In recent studies for video processing-based pedestrian counting, the Albiol method, which is based on the number of salient point detection, is one of the famous pedestrian counting methods. In the Albiol method, a linear relationship is assumed between the number of salient points in a frame and the number of pedestrians. However, the relationship is not always linear because the number of salient points from a pedestrian is decreased when the pedestrian is hidden by another pedestrians in the front, i.e. occlusion. In this research, to improve estimation accuracy of the Albiol method, we investigate the relationship between the number of salient points and the actual number of pedestrians. We first make a virtual pedestrian model which is an occurrence probability distribution of salient points. We next randomly place virtual pedestrians on a virtual frame, and obtain the regression equations. Then, we propose a pedestrian counting method based on the regression equations. Through evaluations using several actual video sequences, we show that the pedestrian counting accuracy is improved by up to 40% by the proposed method compared to the Albiol method.

[Related paper]

  1. Kazuyuki Hashimoto, Yoshiaki Tanigchi, Go Hasegawa, Hirotaka Nakano and Morito Matsuoka, ``Pedestrian counting based on the number of salient points considering non-linear effect of occlusions,'' in Proceedings of the 12th International Conference on Circuits, Systems, Electronics, Control & Signal Processing (CSECS 2013), pp.26-33, Budapest, Hungary, Dec. 2013. [paper]

2.5.3. Pedestrian counting based on Monte Carlo method with binary sensors

Pedestrian counting based on Monte Carlo method with binary sensors

In various fields such as marketing research, traffic control, and safety management, there is a demand for methods for counting the number of pedestrians. Although manual counting is often used for monitoring the number of pedestrians, it entails high labor cost and cannot be used in crowded environments. Therefore, automatic methods for estimating the number of pedestrians using sensors have attracted considerable attention. In particular, binary sensors are among the simplest sensors, capable of detecting only the presence or absence of objects within the sensing region. Although binary sensors can neither detect the number of pedestrians nor identify individual pedestrians within the sensing region, binary sensors possess advantages such as low cost, small size, and energy efficiency in comparison with other types of sensors. Therefore, binary sensors have been expected for estimating the number of pedestrians. In traditional binary sensor-based pedestrian counting systems, accuracy improvement in crowded environments where a large number of pedestrians move is a challenging issue.

In this research, we proposed two methods for estimating the number of pedestrians bi-directionally using binary sensors in two representative environments. The first method is for indoor environments where a large number of pedestrians walk in two directions, such as in a narrow corridor. The proposed method uses ceiling-mounted binary sensors, such as pyroelectric infrared sensors, located on a straight line. The second method is for wider streets, such as in a shopping street, a sidewalk in a downtown area, and so on. The proposed method uses mat-type binary sensors, such as piezo sensors, installed on grid lines. In these proposed methods, sensor data from binary sensors are sent to a monitoring server, and the number of pedestrians is estimated on the monitoring server. To estimate the number of pedestrians from sensor data, we utilize a Monte Carlo method. In particular, a number of simulations on which virtual pedestrians are randomly generated and move are conducted, and virtual sensor data are obtained in each simulation. Simulation scenarios with small difference between actual sensor data and virtual sensor data are selected for estimating the number of pedestrians. We evaluated the accuracy of proposed methods through simulation experiments by changing the number of simulation fields, the specifications of binary sensors, and the number of binary sensors. Simulation results of the first proposed method showed that the relative error in crowded situations decreased by up to 45% as compared with an existing method. In addition, those of the second proposed method exhibited that the relative error is the smallest by using 2x8 binary sensors when a width of fields is 8[m].

[Related paper]

  1. Shuto Fujii, Yoshiaki Taniguchi, Go Hasegawa and Hirotaka Nakano, ``Monte Carlo-based bidirectional pedestrian counting method for compound-eye sensor systems,'' Journal of Emerging Trends in Computing and Information Sciences, vol.4, pp.48-57, Apr. 2013. [paper]
  2. Shuto Fujii, ``Pedestrian counting based on Monte Carlo method with binary sensors,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.

3. Publications

3.1. Journal papers

  1. Shuto Fujii, Yoshiaki Taniguchi, Go Hasegawa and Hirotaka Nakano, ``Monte Carlo-based bidirectional pedestrian counting method for compound-eye sensor systems,'' Journal of Emerging Trends in Computing and Information Sciences, vol.4, pp.48-57, Apr. 2013. [paper]
  2. Xun Shao, Go Hasegawa, Yoshiaki Taniguchi and Hirotaka Nakano, ``The implications of overlay routing for ISPs' peering strategies,'' IEICE Transactions on Information and Systems, vol.E96-D, no.5, pp.1115-1124, May 2013. [paper]
  3. Kazuhito Matsuda, Go Hasegawa and Masayuki Murata, ``An application-level routing method with transit cost reduction based on a distributed heuristic algorithm,'' IEICE Transactions on Communications, vol.E96-B, no.6, pp.1481-1491, June 2013. [paper]
  4. Yoshiaki Taniguchi, Go Hasegawa and Hirotaka Nakano, ``Self-organizing transmission scheduling considering collision avoidance for data gathering in wireless sensor networks,'' Journal of Communications, vol.8, no.6, pp.389-397, June 2013. [paper]
  5. Go Hasegawa, Yuuki Ise, Yoshiaki Taniguchi and Hirotaka Nakano, ``Effect of radio wave obstruction by obstacles on performance of IEEE 802.16j wireless multi-hop relay networks,'' International Journal on Advances in Networks and Services, vol.6, no.1-2, pp.17-26, June 2013. [paper]
  6. Yoshiaki Taniguchi, Masahiro Sasabe, Satoshi Aihara and Hirotaka Nakano, ``Bayesian estimation-based pedestrian tracking in microcells,'' The Scientific World Journal, vol.2013, article ID 187479, 10 pages, Sep. 2013. [paper]
  7. Masafumi Hashimoto, Go Hasegawa and Masayuki Murata, ``SCTP tunneling: Flow aggregation and burst transmission to save energy for multiple TCP flows over a WLAN,'' IEICE Transactions on Communications, vol.E96-B, no.10, pp.2615-2624, Oct. 2013. [paper]
  8. Harumasa Tada, Masayuki Murata, Go Hasegawa, Tadatoshi Babasaki, Jiro Nakamura, Morito Matsuoka, Kazuhiro Matsuda, ``Evaluation of the effect of router caching on network energy consumption,'' IEICE Transactions on Communications, vol.J96-B, no.11, pp.1260-1271, Nov. 2013. (in Japanese) [paper]
  9. Yoshiaki Taniguchi and Hirotaka Nakano, ``Modeling and evaluation of a ceiling-mounted compound-eye sensor,'' Information, vol.17, no.2, pp.663-676, Feb. 2014.

3.2. International conference papers

  1. Dinh Tien Hoang, Go Hasegawa and Masayuki Murata, ``A distributed measurement method exploiting path overlapping in large scale network systems,'' in Proceedings of the 1st Workshop on Large Scale Network Measurements (31st NMRG meeting), Zurich, Switzerland, Oct. 2013. [slide]
  2. Daisuke Kobayashi, Go Hasegawa and Masayuki Murata, ``End-to-end bandwidth measurement method considering effects on power-saving routers,'' in Proceedings of the 2013 Australasian Telecommunications Networks and Applications Conference (ATNAC 2013), pp.112-117, Christchurch, New Zealand, Nov. 2013. [paper]
  3. Dinh Tien Hoang, Go Hasegawa and Masayuki Murata, ``Monitoring available bandwidth in overlay networks using local information exchange,'' in Proceedings of the 2013 Australasian Telecommunications Networks and Applications Conference (ATNAC 2013), pp.172-177, Christchurch, New Zealand, Nov. 2013. [paper]
  4. Kazuyuki Hashimoto, Yoshiaki Tanigchi, Go Hasegawa, Hirotaka Nakano and Morito Matsuoka, ``Pedestrian counting based on the number of salient points considering non-linear effect of occlusions,'' in Proceedings of the 12th International Conference on Circuits, Systems, Electronics, Control & Signal Processing (CSECS 2013), pp.26-33, Budapest, Hungary, Dec. 2013. [paper]
  5. Miki Mizushima, Yoshiaki Tanigchi, Go Hasegawa, Hirotaka Nakano and Morito Matsuoka, ``Counting pedestrians passing through a line in video sequences based on optical flow extraction,'' in Proceedings of the 12th International Conference on Circuits, Systems, Electronics, Control & Signal Processing (CSECS 2013), pp.129-136, Budapest, Hungary, Dec. 2013. [paper]
  6. Shun Sakurai, Go Hasegawa, Naoki Wakamiya and Takanori Iwai, ``Performance evaluation of tunnel sharing method for accommodating M2M communication in mobile cellular networks,'' in Proceedings of the International Workshop on Emerging Technologies for LTE-Advanced and Beyond-4G, Atlanta, USA, Dec. 2013.
  7. Shoichi Takagi, Go Hasegawa, Yoshiaki Tanigchi, Hirotaka Nakano and Morito Matsuoka, ``Comparative evaluation of heuristics for time slot assignment in IEEE 802.16j relay networks,'' in Proceedings of the 7th International Conference on Mobile Computing and Ubiquitous Networking (ICMU 2014), pp.83-84, Bras Basah, Singapore, Jan. 2014. (Poster)
  8. Kazuhito Matsuda, Go Hasegawa and Masayuki Murata, ``Cooperative cache sharing among ISPs for reducing inter-ISP transit cost in content-centric networking,'' in Proceedings of the 28th International Conference on Information Networking (ICOIN 2014), pp.208-213, Phuket, Thailand, Feb. 2014.
  9. Kohei Higo, Go Hasegawa, Yoshiaki Tanigchi, Hirotaka Nakano and Morito Matsuoka, ``Multiple regression analysis of IEEE 802.16j relay network throughput,'' in Proceedings of the 16th International Conference on Advanced Communication Technology (ICACT 2014), pp.437-441, Pyeongchang, Korea, Feb. 2014.

3.5. Theses

3.5.1. Ph.D. Thesis

None

3.5.2. Masters' Theses

  1. Yusuke Iijima, ``Measuring packet loss ratio on overlay networks based on spatial composition,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.
  2. Shoichi Takagi, ``Performance evaluation of heuristic algorithms for time slot assignment in IEEE 802.16j relay networks,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.
  3. Kohei Higo, ``Performance study of IEEE 802.16j relay networks by multiple regression analysis,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.
  4. Shuto Fujii, ``Pedestrian counting based on Monte Carlo method with binary sensors,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.
  5. Miki Mizushima, ``Counting pedestrians passing through a line in crowded scenes by extracting optical flows,'' Master's Thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2014.

3.5.3. Bachelors' Theses

  1. Kouji Suganuma, ``An evaluation of the impact of task allocation on power consumption in data centers,'' Bachelor's Thesis, School of Engineering Science, Osaka University, Feb. 2014. (in Japanese)
  2. Ryota Soe, ``Energy-efficient virtual machine deployment method for datacenters based on degenerate robustness in biological networks,'' Bachelor's Thesis, School of Engineering Science, Osaka University, Feb. 2014. (in Japanese)
  3. Takaaki Deguchi, ``Performance evaluation of task assignment method for reducing air conditioning power consumption at data centers,'' Bachelor's Thesis, School of Engineering Science, Osaka University, Feb. 2014. (in Japanese)
  4. Qiling Tong, ``Performance evaluation of cloud-based energy management system utilizing websocket protocol,'' Bachelor's Thesis, School of Engineering Science, Osaka University, Feb. 2014. (in Japanese)
  5. Akira Morimoto, ``Experimental evaluation of available bandwidth measurement method for arbitrary parts of end-to-end path,'' Bachelor's Thesis, School of Engineering Science, Osaka University, Feb. 2014. (in Japanese)