• 한국통신학회논문지
• /
• 제30권6C호
• /
• pp.570-576
• /
• 2005

얼굴 인식 분야에서 포즈의 변화는 인식률을 저하시키는 가장 심각한 문제로 알려져 있다. 본 논문에서는 이러한 포즈가 변화된 얼굴 영상에 대한 인식률을 높이기 위한 전처리 단계로 정면이 아닌 얼굴 영상을 정면 얼굴 영상으로 변환시키는 방법을 제안한다. 제안한 방법은 PCA 계수를 선형 변환 시키는 변환 행렬을 사용되는데 이 변환 행렬은 PCA 계수 사이의 선형적인 관계를 이용하여 구한다. 제안된 방법은 PCA/LDA를 이용한 얼굴 인식 알고리즘으로 검증하였으며, 실험 결과 제안된 방법이 얼굴 인식률을 $20\%$ 정도 향상시킴을 알 수 있었다.

• Structural Engineering and Mechanics
• /
• 제31권2호
• /
• pp.211-223
• /
• 2009

Transient response analysis can be conducted either in the time domain, or via the frequency domain. Sometimes a frequency domain method (FDM) has advantages over a time domain method. A practical issue in the FDM is to find out an appropriate extended period, which may be affected by several factors, such as the excitation duration, the system damping, the artificial damping, the period of interest, etc. In this report, the extended period of the FDM based on the Duhamel's integral is investigated. This Duhamel's integral based FDM does not involve the unit impulse response function (UIRF) beyond the period of interest. Due to this fact, the ever-lasting UIRF can be simply set as zero beyond the period of interest to shorten the extended period. As a result, the preferred extended period is the summation of the period of interest and the excitation duration. This conclusion is validated by numerical examples. If the extended period is too short, then the front portion of the period of interest is more prone to errors than the rear portion, but the free vibration segment is free of the wraparound error.

• Wind and Structures
• /
• 제18권1호
• /
• pp.69-82
• /
• 2014

Flow around a small white fir tree was investigated with varying the length of the bottom trunk (hereafter referred to as bottom gap). The velocity fields around the tree, which was placed in a closed-type wind tunnel test section, were quantitatively measured using particle image velocimetry (PIV) technique. Three different flow regions are observed behind the tree due to the bottom gap effect. Each flow region exhibits a different flow structure as a function of the bottom gap ratio. Depending on the gap ratio, the aerodynamic porosity of the tree changes and the different turbulence structure is induced. As the gap ratio increases, the maximum turbulence intensity is increased as well. However, the location of the local maximum turbulence intensity is nearly invariant. These changes in the flow and turbulence structures around a tree due to the bottom gap variation significantly affect the shelter effect of the tree. The wind-speed reduction is increased and the height of the maximum wind-speed reduction is decreased, as the gap ratio decreases.

• Structural Engineering and Mechanics
• /
• 제52권5호
• /
• pp.903-918
• /
• 2014

Large space structures may have resonant low eigenvalues and often these appear with closely-spaced natural frequencies. Owing to the coupling among modes with closely-spaced natural frequencies, each eigenvector corresponding to closely-spaced eigenvalues is ill-conditioned that may cause structural instability. The subspace to an invariant subspace corresponding to closely-spaced eigenvalues is well-conditioned, so a method is presented to design the feedback control law of intelligent structures with closely-spaced eigenvalues in this paper. The main steps are as follows: firstly, the system with closely-spaced eigenvalues is transformed into that with repeated eigenvalues by the spectral decomposition method; secondly, the computation for the linear combination of eigenvectors corresponding to repeated eigenvalues is obtained; thirdly, the feedback control law is designed on the basis of the system with repeated eigenvalues; fourthly, the system with closely-spaced eigenvalues is regarded as perturbed system on the basis of the system with repeated eigenvalues; finally, the feedback control law is applied to the original system, the first order perturbations of eigenvalues are discussed when the parameter modifications of the system are introduced. Numerical examples are given to demonstrate the application of the present method.

• 제어로봇시스템학회논문지
• /
• 제15권5호
• /
• pp.525-529
• /
• 2009

Most of FLC received input data from error e and change-of-error e' with no relation with system complexity. Basic scheme follows typical PD and PI or PID Controller and that has been developed through fixed ME In this paper, We studied the relationship between MF and system response and system response through changing Fuzzy variable of consequence MF and propose the simple FLC using this relationship. The response of FLC is changed according to the width of Fuzzy variable of consequence MF. As changing the Fuzzy variable of consequence MF shows various nonlinear characteristic, we studied the relation between response and MF using analytical method. We designed the effective FLC using three-variable MF and nine rules and took simulation for verification. In this study, we propose the method to design system with FLC in stability point which is an impotent characteristic of designing system. The circle criterion which is adapted to analysis the nonlinear system is put to use for proposed method. Since SISO FLC has a time-invariant and odd characteristic we can use the critical point not disk which is generally used to determine the stability in the circle criterion, to determine the stability. Using this, we can get the maximum critical point plot of SISO FLC with changing the consequence fuzzy variables. The predetermined critical point plot of FLC can be used to decide the region of the system to be stable. This method is effectively used to design the SISO FLC.

• 대한전자공학회논문지
• /
• 제22권6호
• /
• pp.1-6
• /
• 1985

기준입력은 정상상태 오차에 영향을 주는 요인 중의 하나다. 본 논문은 표준입력에 의해 야기피는 정상상태 오차를 제거할 수 있는 새로운 디지날 제각기 설계 방법을 제시한다. 먼저 연준입력 함수 형태에 따라, 기준입력 항이 제거된 새로운 제어 시스템 식이 유도되고. 최려제어를 이용하여 유도된 시스템의 출력을 최소로 하는 제어법칙이 구해진다. 강태공간 방법에 의해. 제시된 제어 알고리즘은 불안정 시스템을 포함해서 시불변 선형 시스템에 적용될 구 있다.

• JSTS:Journal of Semiconductor Technology and Science
• /
• 제13권2호
• /
• pp.157-169
• /
• 2013

Scale Invariant Feature Transform (SIFT) generates image features widely used to match objects in different images. Previous work on hardware-based SIFT implementation requires excessive internal memory and hardware logic [1]. In this paper, a new hardware organization is proposed to implement SIFT with less memory and hardware cost than the previous work. To this end, a parallel Gaussian filter bank is adopted to eliminate the buffers that store intermediate results because parallel operations allow all intermediate results available at the same time. Furthermore, the processing order is changed from the raster-scan order to the block-by-block order so that the line buffer size storing the source image is also reduced. These techniques trade the reduction of memory size with a slight increase of the execution time and external memory bandwidth. As a result, the memory size is reduced by 94.4%. The proposed hardware for SIFT implementation includes the Descriptor generation block, which is omitted in the previous work [1]. The addition of the hardwired descriptor generation improves the computation speed by about 30 times when compared with the previous work.

• 말소리와 음성과학
• /
• 제12권2호
• /
• pp.17-27
• /
• 2020

This paper examines kinematic characteristics of /pk/ clusters, as compared to /kk/ and /pp/ with varying vowel contexts and speech rate. The results of EMMA data from eight Seoul-Korean speakers indicate as follows. Firstly, comparing /pk/ to /pp/ sequences, lips closing movement was faster and spatially greater in the /a/-to-/a/ context while temporally longer in the /i/-to-/i/ context. It was smaller in spatial displacement and shorter in temporal duration in /pk/ sequences. Peak velocity did not vary. Secondly, comparing /pk/ with /pp/ and /kk/ controls, lip aperture was less constricted in the /a/-to-/a/ context than /i/-to-/i/, but the maximum contact between the upper and lower lips was invariant across different vocalic contexts within /pk/ sequences (/apka/=/ipki/). Categorical reduction of C1 in /pk/ sequences fell in with the low-vowel and fast-rate conditions with across-/within-speaker variability. Gradient reduction of C1 was observed in all C1C2 types, being more frequent in fast rate. Lastly, the jaw articulator was a stable indicator of rate effects. The implication of the current study is that gestural reduction occurs with categorical reduction and general spatiotemporal weakening in the assimilating contexts, while quantitative properties of gestures may be a reason for gradient reduction, not necessarily confined to place assimilation.

• 한국산업정보학회논문지
• /
• 제6권2호
• /
• pp.16-22
• /
• 2001

본 논문에서는 화상자료의 특성인 이웃 화소간의 종속성을 표현하는데 적합한 깁스분포를 바탕으로 특징벡터를 추출하여 변형된 글자를 인식하는 새로운 방법을 제안하였다. 추출된 특징벡터는 이미지의 크기, 위치, 회전에 대하여 불변한 특성을 갖는 2차원 조건부 모멘트로 구성된다. 변형된 글자 인식을 위한 알고리즘은 특징벡터 추출하는 과정과 패턴을 인식하는 과정으로 구성하였다. (i) 특징벡터는 하나의 이미지에 대하여 추정된 조건부 깁스분포를 바탕으로 2차원 조건부 모멘트를 계산하여 추출한다. (ii) 변형된 문자 인식은 제안된 판별거리함수를 계산하여 최소거리를 산출한 미지의 변형된 문자를 원형문자로 인식한다. 제안된 방법에 대한 성능평가를 위하여, 생성된 훈련 데이터를 만들어 Workstation에서 실험 한 결과 96%이상의 인식성능이 있음을 밝혔다.

• 한국산업정보학회논문지
• /
• 제4권2호
• /
• pp.1-6
• /
• 1999

화상 데이터의 특성을 표현하는데 적합한 깁스분포를 바탕으로 특징벡터를 추출하여 패턴을 분류하는 새로운 알고리즘을 제안하였다. 특징벡터는 화상의 크기, 위치, 회전에 대해서 불변이며 접영에 대해서도 덜 민감한 특징을 갖는 2차원 모멘트들의 원소로 만들어진다. 알고리즘은 공간정보를 갖는 2차원 모멘트를 이용하여 특징벡터를 추출하는 과정과 거리함수를 이용하여 패턴을 분류하는 과정으로 구축하였다. 특징벡터는 깁스분포의 묘수를 추정하여 2차원 조건부 모멘트를 추출하여 구성한다. 패턴 분류 과정은 추출된 특징벡터로부터 제안된 판별거리함수를 계산하여 여러 원형 패턴 가운데 최소거리를 산출한 미지의 패턴을 원형패턴으로 분류한다. 제안된 방법의 성능을 검증하기 위하여 대문자와 소문자 52자로 구성된 훈련 데이터를 만들어 SUN ULTRA 10 워크스테이션에서 실험을 한 결과 98%이상의 분류성능이 있음을 밝혔다.