• 한국터널지하공간학회 논문집
• /
• 제19권1호
• /
• pp.95-107
• /
• 2017

본 논문에서는 제도적으로 운영 중인 터널내 CCTV들로부터 실시간으로 들어오는 영상들을 최신 딥러닝 알고리즘을 이용, 학습시켜 다양한 조건의 터널환경에서 돌발 상황을 감지하고 그 돌발 상황의 종류들을 분류해 내는 시스템 개발을 위한 사전검토 연구를 수행하였다. 사전검토 연구를 위해, 2개의 도로현장의 교통류 CCTV영상 일부를 이용하여 가용한 전통적인 영상처리기법으로 영상내부로 집입하는 차량을 감지하고, 이동경로를 추적하여 일정 시간간격의 이동 차량의 좌표와 시간정보를 추출하고 학습자료를 구성하였다. 각 차량의 이동정보는 차선변경, 정차 등 6가지의 이벤트 정보와 연계된다. 차량 이동정보와 이벤트로 구성된 학습자료는 레질리언스(resilience) 기계학습 알고리즘을 이용하여 학습하였다. 2개의 은닉층을 설정하고, 각 은닉층의 노드수에 대한 9개의 은닉구조 모델을 설정하여 매개변수 연구를 수행하였다. 본 사전검토의 경우에는 첫 번째, 두 번째 은닉층 노드수가 각각 300개와 150개로 설정된 모델이 합리적으로 가장 추론정확도가 높은 것으로 평가되었다. 이로부터 일반화되기 매우 힘든 복잡한 교통류 상황을 기계학습을 이용하여 어떠한 사전 규칙설정 없이도 교통류의 특징들을 정확히 자동으로 감지할 수 있는 가능성을 보였다. 본 시스템은 시스템의 운용을 통해 지속적으로 교통류 영상과 이벤트 정보가 늘어난다면, 자동으로 그 시스템의 인지능력과 정확도가 자동으로 향상되는 효과도 기대할 수 있다.

• 천문학회지
• /
• 제29권spc1호
• /
• pp.63-64
• /
• 1996

The nature of distant faint blue field galaxies remains a mystery, despite the fact that much attention has been devoted to this subject in the last decade. Galaxy counts, particularly those in the optical and near ultraviolet bandpasses, have been demonstrated to be well in excess of those expected in the 'no-evolution' scenario. This has usually been taken to imply that galaxies were brighter in the past, presumably due to a higher rate of star formation. More recently, redshift surveys of galaxies as faint as B$\~$24 have shown that the mean redshift of faint blue galaxies is lower than that predicted by standard evolutionary models (de-signed to fit the galaxy counts). The galaxy number count data and redshift data suggest that evolutionary effects are most prominent at the faint end of the galaxy luminosity function. While these data constrain the form of evolution of the overall luminosity function, they do not constrain evolution in individual galaxies. We are carrying out a series of observations as part of a long-term program aimed at a better understanding of the nature and amount of luminosity evolution in individual galaxies. Our study uses the luminosity-linewidth relation (Tully-Fisher relation) for disk galaxies as a tool to study luminosity evolution. Several studies of a related nature are being carried out by other groups. A specific experiment to test a 'no-evolution' hypothesis is presented here. We have used the AUTOFIB multifibre spectro-graph on the 4-metre Anglo-Australian Telescope (AAT) and the Rutgers Fabry-Perot imager on the Cerro Tolalo lnteramerican Observatory (CTIO) 4-metre tele-scope to measure the internal kinematics of a representative sample of faint blue field galaxies in the red-shift range z = 0.15-0.4. The emission line profiles of [OII] and [OIII] in a typical sample galaxy are significantly broader than the instrumental resolution (100-120 km $s^{-l}$), and it is possible to make a reliable de-termination of the linewidth. Detailed and realistic simulations based on the properties of nearby, low-luminosity spirals are used to convert the measured linewidth into an estimate of the characteristic rotation speed, making statistical corrections for the effects of inclination, non-uniform distribution of ionized gas, rotation curve shape, finite fibre aperture, etc.. The (corrected) mean characteristic rotation speed for our distant galaxy sample is compared to the mean rotation speed of local galaxies of comparable blue luminosity and colour. The typical galaxy in our distant sample has a B-band luminosity of about 0.25 L$\ast$ and a colour that corresponds to the Sb-Sd/Im range of Hub-ble types. Details of the AUTOFIB fibre spectroscopic study are described by Rix et al. (1996). Follow-up deep near infrared imaging with the 10-metre Keck tele-scope+ NIRC combination and high angular resolution imaging with the Hubble Space Telescope's WFPC2 are being used to determine the structural and orientation parameters of galaxies on an individual basis. This information is being combined with the spatially resolved CTIO Fabry-Perot data to study the internal kinematics of distant galaxies (Ing et al. 1996). The two main questions addressed by these (preliminary studies) are: 1. Do galaxies of a given luminosity and colour have the same characteristic rotation speed in the distant and local Universe? The distant galaxies in our AUTOFIB sample have a mean characteristic rotation speed of $\~$70 km $s^{-l}$ after correction for measurement bias (Fig. 1); this is inconsistent with the characteristic rotation speed of local galaxies of comparable photometric proper-ties (105 km $s^{-l}$) at the > $99\%$ significance level (Fig. 2). A straightforward explanation for this discrepancy is that faint blue galaxies were about 1-1.5 mag brighter (in the B band) at z $\~$ 0.25 than their present-day counterparts. 2. What is the nature of the internal kinematics of faint field galaxies? The linewidths of these faint galaxies appear to be dominated by the global disk rotation. The larger galaxies in our sample are about 2"-.5" in diameter so one can get direct insight into the nature of their internal velocity field from the $\~$ I" seeing CTIO Fabry-Perot data. A montage of Fabry-Perot data is shown in Fig. 3. The linewidths are too large (by. $5\sigma$) to be caused by turbulence in giant HII regions.