• The KIPS Transactions:PartD
• /
• v.11D no.1
• /
• pp.95-104
• /
• 2004

This paper proposes an efficient remote video evaluation system that is matched well with personalized characteristics of students using information filtering based on user profile. For making a question in forms of video, a key frame extraction method based on coordinate, size and color information is proposed. And Question-mating intervals are extracted using gray-level histogram difference and time window. Also, question-making method that combined category-based system with keyword-based system is used for efficient evaluation. Therefore, students can enhance their study achievement through both supplementing their inferior area and preserving their interest area.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.48 no.3
• /
• pp.50-59
• /
• 2011

In this paper, we present a spectrum sensing method by exploiting the relationship of previous and current detected data sets in frequency domain. Most of the traditional spectrum sensing methods only consider the current detected data sets of Primary User (PU). Previous state of PU is a kind of conditional probability that strengthens the reliability of the detector. By considering the relationship of the previous and current spectrum sensing, cross entropy-based spectrum sensing is proposed to detect PU signal more effectively, which has a strengthened performance and is robust. When previous detected signal is noise, the discriminating ability of cross entropy-based spectrum sensing is no better than conventional entropy-based spectrum sensing. To address this problem, we propose an improved cross entropy-based frequency-domain spectrum sensing. Regarding the spectrum sensing scheme, we have derived that the proposed method is superior to the cross entropy-based spectrum sensing. We proceed a comparison of the proposed method with the up-to-date entropy-based spectrum sensing in frequency-domain. The simulation results demonstrate the performance improvement of the proposed spectrum sensing method.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.39 no.4
• /
• pp.455-461
• /
• 2002

This paper presents a new parametric log line equation of polar form for Hough transform in log-polar plane, in which it can remove the well-known unboundedness problem of Hough parameters. Bolduc's method is used to generate a log-polar image dividing the fovea and periphery from a Cartesian image. Edges of the fovea and periphery are detected by using the Sobel mask and the proposed space-variant gradient mask, and are combined in the log-polar plane. The sampled points that might constitute a log line are quite sparse in a deep peripheral region due to severe under-sampling, which is an inherent property of LPM. To cope with such under-sampling, we determine the values of cumulative cells in Hough space by using the space-variant weighting. In our experiments, the proposed method demonstrates its validity of detecting not only the lines passing through both the fovea and periphery but also the lines in a deep periphery.

• Journal of Cadastre & Land InformatiX
• /
• v.48 no.2
• /
• pp.213-225
• /
• 2018

For constructing three-dimensional (3D) spatial information occlusion region problem arises in the process of taking the texture of the building. In order to solve this problem, it is necessary to investigate the automation method to automatically recognize the occlusion region, issue it, and automatically complement the texture. In fact there are occasions when it is possible to generate a very large number of structures and occlusion, so alternatives to overcome are being considered. In this study, we attempt to apply an approach to automatically create an occlusion region based on learning by patterning the blocked region using the recently emerging deep learning algorithm. Experiment to see the performance automatic detection of people, banners, vehicles, and traffic lights that cause occlusion in building walls using two advanced algorithms of Convolutional Neural Network (CNN) technique, Faster Region-based Convolutional Neural Network (R-CNN) and Mask R-CNN. And the results of the automatic detection by learning the banners in the pre-learned model of the Mask R-CNN method were found to be excellent.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2008.05a
• /
• pp.353-358
• /
• 2008

청각장애인들은 건청인에 비해 의사소통의 기회가 적어 원만한 상호관계를 유지하는데 어려움이 있다. 이러한 문제는 청각장애인들이 구화를 대신해 몸짓이나 손짓을 이용하여 의사를 전달하는 수화를 건청인들이 대부분 습득하고 있지 않아 청각장애인들과 의사소통이 거의 불가능 한 것이 현실이다. 따라서 본 논문에서는 건청인과 청각장애인들 간의 의사소통을 가능하게 하기 위한 전단계로 FCM 알고리즘을 이용한 지화 인식 방법을 제안한다. 제안된 방법은 화상 카메라를 통해 얻어진 영상에서 YCbCr 컬러 공간과 HSI 컬러 공간을 이용하여 피부영역을 검출한 후 추출된 피부영역을 4 방향 윤곽선 추적 알고리즘을 적용하여 두 손의 위치를 추적한다. 그리고 추적한 두 손의 영역에 대해 형태학적 정보를 이용하여 잡음을 제거한 후, 최종적으로 두 손의 영역을 추출한다. 추출된 손의 영역은 FCM 알고리즘을 적용하여 지화의 특징들을 분류하고 인식한다. 제안된 방법의 성능을 평가하기 위해 화상카메라에서 획득한 지화 영상을 대상으로 실험한 결과, 두 손 영역의 추출과 지화 인식에 있어서 효과적인 것을 확인하였다.

• Journal of Convergence for Information Technology
• /
• v.11 no.9
• /
• pp.28-34
• /
• 2021

In the case of object detection using deep learning, both accuracy and real-time are required. However, it is difficult to use a deep learning model that processes a large amount of data in a limited resource environment. To solve this problem, this paper proposes an object detection model for small embedded devices. Unlike the general detection model, the model size was minimized by using a structure in which the pre-trained feature extractor was removed. The structure of the model was designed by repeatedly stacking lightweight convolution blocks. In addition, the number of region proposals is greatly reduced to reduce detection overhead. The proposed model was trained and evaluated using the public dataset PASCAL VOC. For quantitative evaluation of the model, detection performance was measured with average precision used in the detection field. And the detection speed was measured in a Raspberry Pi similar to an actual embedded device. Through the experiment, we achieved improved accuracy and faster reasoning speed compared to the existing detection method.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.466-472
• /
• 2021

Speaker Change Detection (SCD) refers to finding the moment when the main speaker changes from one person to the next in a speech conversation. In speaker change detection, difficulties arise due to overlapping speakers, inaccuracy in the information labeling, and data imbalance. To solve these problems, TIMIT corpus widely used in speech recognition have been concatenated artificially to obtain a sufficient amount of training data, and the detection of changing speaker has performed after identifying overlapping speakers. In this paper, we propose an speaker change detection system that considers the speaker overlapping. We evaluated and verified the performance using various approaches. As a result, a detection system similar to the X-Vector structure was proposed to remove the speaker overlapping region, while the Bi-LSTM method was selected to model the speaker change system. The experimental results show a relative performance improvement of 4.6 % and 13.8 % respectively, compared to the baseline system. Additionally, we determined that a robust speaker change detection system can be built by conducting related studies based on the experimental results, taking into consideration text and speaker information.

• Journal of the Institute of Convergence Signal Processing
• /
• v.9 no.1
• /
• pp.1-9
• /
• 2008

One of the problems for implementing the spot detection phase in the 2-DE gel image analysis program is the eliminating noises in the image. Remained noises after the preprocessing phase cause the over-segmented regions by the segmentation phase. To identify and exclude the over-segmented background regions, if we use the fixed thresholding method that is choosing an intensity value for the threshold, the spots that is invisible by the eyes but mean a very small amount proteins which have important role in the biological samples could be eliminated. This paper propose an adaptive thresholding method that come from an idea that is got on statistical analysing for the prominences of the peaks. The adaptive thresholding method works as following. Firstly we calculate an average prominence value curve and fit it to exponential function curve, as a result we get parameters for the exponential function. And then we calculate a threshold value by using the parameters and probability distribution of errors. Lastly we apply the threshold value to the region for determining the region is a noise or not. According to the probability distribution of errors, the reliability is 99.85% and we show the correctness of the proposed method by representing experiment results.

• The KIPS Transactions:PartB
• /
• v.15B no.2
• /
• pp.87-94
• /
• 2008

In this paper, we propose a new vision-based fire detection method for a real-life application. Most previous vision-based methods using color information and temporal variation of pixel produce frequent false alarms because they used a lot of heuristic features. Furthermore there is also computation delay for accurate fire detection. To overcome these problems, we first detected candidated fire regions by using background modeling and color model of fire. Then we made probabilistic models of fire by using a fact that fire pixel values of consecutive frames are changed constantly and applied them to a Bayesian Network. In this paper we used two level Bayesian network, which contains the intermediate nodes and uses four skewnesses for evidence at each node. Skewness of R normalized with intensity and skewnesses of three high frequency components obtained through wavelet transform. The proposed system has been successfully applied to many fire detection tasks in real world environment and distinguishes fire from moving objects having fire color.

• 전자공학회논문지 IE
• /
• v.43 no.2
• /
• pp.63-69
• /
• 2006

Islanding phenomenon of utility-connected PV power conditioning systems(PV PCS) can cause a variety of problems and must be prevented. If the real and reactive powers supplied by PV PCS are closely matched to those of load, islanding detection by passive methods becomes difficult. The active frequency drift(AFD) method, called the frequency bias method, enables islanding detection by forcing the frequency of the voltage in the islanding to drift up or down. In this paper, non-detection zone(NDZ) of AFD is analyzed for the islanding detection method of utility-connected PV PCS by simulation tool PSIM.