• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2021년도 추계학술발표대회
• /
• pp.878-879
• /
• 2021

Most state-of-the-art CNNs for action recognition are based on a two-stream architecture: RGB frames stream represents the appearance and the optical flow stream interprets the motion of action. However, the cost of optical flow computation is very high and then it increases action recognition latency. We introduce a design strategy for action recognition inspired by a two-stream network and teacher-student architecture. There are two sub-networks in our neural networks, the optical flow sub-network as a teacher and the RGB frames sub-network as a student. In the training stage, we distill the feature from the teacher as a baseline to train student sub-network. In the test stage, we only use the student so that the latency reduces without computing optical flow. Our experiments show that its advantages over two-stream architecture in both speed and performance.

• 한국자기학회:학술대회 개요집
• /
• 한국자기학회 2008년도 Asian Magnetics Conference
• /
• pp.314.1-314.1
• /
• 2008

• Journal of information and communication convergence engineering
• /
• 제7권3호
• /
• pp.377-382
• /
• 2009

In this paper, theoretical analysis for metropolitan optical networks is performed. First, analytical optical SNR is derived assuming each node consists of an EDFA, an optical filter, an optical switch, and a VOA, and then the relationship between OSNR and BER is studied. In a metropolitan optical network, an optical signal can be dropped to deliver data, and we also studied the effect of drop loss on system performance. When the drop loss is relatively small, the receiver structure of the node can be treated as a preamplifier receiver which is widely used in long-haul systems. In that case, ASE noise from EDFAs is the dominant noise source in the receiver. However, system performance is relatively insensitive to OSNR when the drop loss is significant because of the noise sources in the receiver (thermal and shot noise).

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제10권1호
• /
• pp.1-17
• /
• 2016

The separated management and operation of commercial IP/optical multilayer networks makes network operators look for a unified control plane (UCP) to reduce their capital and operational expenditure. Software-defined networking (SDN) provides a central control plane with a programmable mechanism, regarded as a promising UCP for future optical networks. The general control and scheduling mechanism in SDN-based optical burst switching (OBS) networks is insufficient so the controller has to process a large number of messages per second, resulting in low network resource utilization. In view of this, this paper presents the burst-flow scheduling mechanism (BFSM) with a proposed scheduling algorithm considering channel usage. The simulation results show that, compared with the general control and scheduling mechanism, BFSM provides higher resource utilization and controller performance for the SDN-based OBS network in terms of burst loss rate, the number of messages to which the controller responds, and the average latency of the controller to process a message.

• 제어로봇시스템학회논문지
• /
• 제8권3호
• /
• pp.256-263
• /
• 2002

Optical Multistage Interconnection Networks(OMINs) comprising photonic switches have been studied extensively as important interconnecting building blocks for communication networks and parallel computing systems. A basic element of photonic switching networks is a 2$\times$2 directional coupler with two inputs and two outputs. This paper is concerned with the diagnosis of cross-talk-faults in OMINs. As the size of today's network becomes very large, the conventional diagnosis methods based on tests and simulation have become inefficient, or even more, impractical. In this paper, we propose a simple and easily implementable algorithm for detection and isolation of the single crosstalk-fault in OMINs. Specifically, we develope an algorithm fur the isolation of the source fault in switching elements whenever the single crosstalk-fault is detected in OMINS. The proposed algorithm is illustrated by an example of 16$\times$16 banyan network.

• Journal of Communications and Networks
• /
• 제13권3호
• /
• pp.286-297
• /
• 2011

A novel converged optical network unit (ONU)-base station (BS) architecture has been contemplated for next-generation optical-wireless networks. It has been demonstrated through high quality studies that data traffic carried by both wired and wireless networks exhibit self-similar and long range dependent characteristics; attributes that classical teletraffic theory based on simplistic Poisson models fail to capture. Therefore, in order to apprehend the proposed converged architecture and to reinforce the provisioning of tightly bound quality of service (QoS) parameters to end-users, we substantiate the analysis of the QoS behavior of the ONU-BS under self-similar and long range dependent traffic conditions using custom queuing which is a common queuing discipline. This paper extends our previous work on priority queuing and brings novelty in terms of presenting performance analysis of the converged ONU-BS under realistic traffic load conditions. Further, the presented analysis can be used as a network planning and optimization tool to select the most robust and appropriate queuing discipline for the ONU-BS relevant to the QoS requirements of different applications.

• 한국광학회:학술대회논문집
• /
• 한국광학회 2003년도 제14회 정기총회 및 03년 동계학술발표회
• /
• pp.246-247
• /
• 2003

In recent years, soaring traffic volumes over optical communications networks lead to the rapid advances in information communications equipments. In backbone communications networks, there has been an advance in high-density transmission through DWDM, which can simultaneously transmit numerous signals with different wavelengths. When the channel spacing is narrower, the cross-talk between channels an important parameter that guarantees to the high performance of a whole system, becomes critical. (omitted)

• 정보처리학회논문지C
• /
• 제10C권4호
• /
• pp.447-454
• /
• 2003

광 버스트 교환(OBS : Optical Burst Switching)은 광 지연선(FDL: Fiber Delay Line) 사용의 감소와 차세대 전광 네트워크에서의 광 스위칭 패러다임을 실현하기 위해 제안되고 있다. OBS는 버스트를 시분할 다중화함으로써 파장 라우팅보다 대역폭의 효율성을 제공하고 코어 네트워크에 대한 확장성을 제공한다. 최근, 광 버스트 교환망에서 어떻게 QoS(Quality of Service)를 제공할 것인가가 이슈가 되고 있다. 본 논문에서는 혼잡 발생시 우회 라우팅(deflection routing) 경로로 우선순위 0, 1, 2의 버스트를 우회시켜 QoS를 제공하는 우회 라우팅 방법에 대해 제안한다. 제안된 방법의 큰 장점은 QoS를 간단히 제공할 수 있다는 것이고, 이는 QoS 보장 알고리즘으로부터 제공된다. 또한, 제안한 QoS 제공방식은 각 코어 라우터에서 광 지연선의 사용을 최소로 하면서 트래픽을 공평하게 분배함으로써 효율적인 네트워크를 구현하게 한다. 제안 QoS 제공 방식의 우선순위 0, 1, 2 버스트의 QoS 보장과 효율적인 네트워크 사용에 대하여 OPNET을 사용한 컴퓨터 시뮬레이션을 이용하여 평가하였다. 이에 따른 결과로써, 높은 우선 순위 버스트의 종단간 지연은 향상되었으며, 또한 효율적인 네트워크 이용이 확인되었다.

• 한국광학회지
• /
• 제24권4호
• /
• pp.184-188
• /
• 2013

본 논문에서는 스마트 홈 네트워크, 자동차용 광네트워크, 선박용 광네트워크 등의 멀티모드 광네트워크 구성을 위한 멀티모드 광분배기를 제작하였다. 멀티모드 광분배기는 폴리머 기반의 자외선 임프린트(ultra violet imprint, UV-imprint) 공정에 의해 $200{\mu}m$ 코어 사이즈를 갖는 대구경으로 제작하였으며, 신호 분기를 위한 1:1 균등 분배기 및 네트워크 모니터링을 위한 9:1 비균등 분배기를 제작하였다. 대구경 광분배기를 제작하기 위해 실리콘 웨이퍼에 Bosch process에 의한 광분배기 패턴을 식각하여 imprint 마스터를 제작 후 폴리디메틸실록산(polydimethylsiloxane, PDMS)를 이용한 탄성체 몰드를 역상 복제 하였다. 역상 복제된 PDMS 몰드를 이용해 UV-imprint 공정에 의한 대구경 광분배기 칩을 제작하였다. 또한, 폴리메틸메타크릴레이트(polymethylmethacrylate, PMMA) 광섬유 블록을 제작하고 광분배기 칩의 입출력 단에 광섬유를 정렬 부착하여, 최종적으로 광분배기는 광분배기 칩과 패키징을 위한 광섬유 블록을 폴리머 기반의 저가형으로 제작하였다. 제작된 폴리머 기반 광분배기의 도파손실, 삽입손실 및 출력 균일도를 측정하고, 내환경 특성 평가를 위해 $-40^{\circ}C{\sim}+60^{\circ}C$ 주기의 temperature cycling test 후 삽입손실 변화를 측정하였다.

• 전자공학회논문지B
• /
• 제32B권3호
• /
• pp.48-56
• /
• 1995

This paper presents a method of implementing efficient optical flow estimation for dynamic scene analysis using the hierarchical Hopfield neural networks. Given the two consequent inages, Zhou and Chellappa suggested the Hopfield neural network for computing the optical flow. The major problem of this algorithm is that Zhou and Chellappa's network accompanies self-feedback term, which forces them to check the energy change every iteration and only to accept the case where the lower the energy level is guaranteed. This is not only undesirable but also inefficient in implementing the Hopfield network. The another problem is that this model cannot allow the exact computation of optical flow in the case that the disparities of the moving objects are large. This paper improves the Zhou and Chellapa's problems by modifying the structure of the network to satisfy the convergence condition of the Hopfield model and suggesting the hierarchical algorithm, which enables the computation of the optical flow using the hierarchical structure even in the presence of large disparities.