• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.41 no.2
• /
• pp.79-88
• /
• 2004

Gaze detection is to locate the position on a monitor screen where a user is looking by computer vision. Gaze detection systems have numerous fields of application. They are applicable to the man-machine interface for helping the handicapped to use computers and the view control in three dimensional simulation programs. In our work, we implement it with a computer vision system setting a IR-LED based single camera. To detect the gaze position, we locate facial features, which is effectively performed with IR-LED based camera and SVM(Support Vector Machine). When a user gazes at a position of monitor, we can compute the 3D positions of those features based on 3D rotation and translation estimation and affine transform. Finally, the gaze position by the facial movements is computed from the normal vector of the plane determined by those computed 3D positions of features. In addition, we use a trained neural network to detect the gaze position by eye's movement. As experimental results, we can obtain the facial and eye gaze position on a monitor and the gaze position accuracy between the computed positions and the real ones is about 4.8 cm of RMS error.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.5 s.305
• /
• pp.123-136
• /
• 2005

We propose a recognition system for superimposed patterns based on selective attention model and SVM which produces better performance than artificial neural network. The proposed selective attention model includes attention layer prior to SVM which affects SVM's input parameters. It also behaves as selective filter. The philosophy behind selective attention model is to find the stopping criteria to stop training and also defines the confidence measure of the selective attention's outcome. Support vector represents the other surrounding sample vectors. The support vector closest to the initial input vector in consideration is chosen. Minimal euclidean distance between the modified input vector based on selective attention and the chosen support vector defines the stopping criteria. It is difficult to define the confidence measure of selective attention if we apply common selective attention model, A new way of doffing the confidence measure can be set under the constraint that each modified input pixel does not cross over the boundary of original input pixel, thus the range of applicable information get increased. This method uses the following information; the Euclidean distance between an input pattern and modified pattern, the output of SVM, the support vector output of hidden neuron that is the closest to the initial input pattern. For the recognition experiment, 45 different combinations of USPS digit data are used. Better recognition performance is seen when selective attention is applied along with SVM than SVM only. Also, the proposed selective attention shows better performance than common selective attention.

• Journal of the Korea Concrete Institute
• /
• v.20 no.4
• /
• pp.513-522
• /
• 2008

The fiber dispersion performance in fiber-reinforced cementitious composites is a crucial factor with respect to achieving desired mechanical performance. However, evaluation of the fiber dispersion performance in the composite PVA-ECC (Polyvinyl alcohol-Engineered Cementitious Composite) is extremely challenging because of the low contrast of PVA fibers with the cement-based matrix. In the present work, an enhanced fiber detection technique is developed and demonstrated. Using a fluorescence technique on the PVA-ECC, PVA fibers are observed as green dots in the cross-section of the composite. After capturing the fluorescence image with a Charged Couple Device (CCD) camera through a microscope. The fibers are more accurately detected by employing a series of process based on a categorization, watershed segmentation, and morphological reconstruction.

• Resources Recycling
• /
• v.26 no.6
• /
• pp.73-83
• /
• 2017

Used small household appliances have a wide variety of product types and component materials, and contain high percentage of black plastics. However, they are not being recycled efficiently as conventional sensors such as near-infrared ray (NIR), etc. are not able to detect black plastic by types. In the present study, an automatic sorting system was developed based on laser-induced breakdown spectroscopy (LIBS) to promote the recycling of waste plastics. The system we developed mainly consists of sample feeder, automatic position recognition system, LIBS device, separator and control unit. By applying laser pulse on the target sample, characteristic spectral data can be obtained and analyzed by using CCD detectors. The obtained data was then treated by using a classifier, which was developed based on artificial intelligent algorithm. The separation tests on waste plastics also were carried out by using a lab-scale automatic sorting system and the test results will be discussed. The classification rate of the radial basis neural network (RBFNNs) classifier developed in this study was about > 97%. The recognition rate of the black plastic by types with the automatic sorting system was more than 94.0% and the sorting efficiency was more than 80.0%. Automatic sorting system based on LIBS technology is in its infant stage and it has a high potential for utilization in and outside Korea due to its excellent economic efficiency.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.39 no.11
• /
• pp.1123-1129
• /
• 2015

A pair of two-way valves typically is used in automotive washing machines, where the water flow direction is frequently reversed and highly pressurized clean water is sprayed to remove the oil and dirt remaining on machined engine and transmission blocks. Although this valve system has been widely used because of its competitive price, its application is sometimes restricted by surging effects, such as pressure ripples occurring in rapid changes in water flow caused by inaccurate valve control. As an alternative, one three-way reversing valve can replace the valve system because it provides rapid and accurate changes to the water flow direction without any precise control device. However, a cavitation effect occurs because of the complicated bottom plug shape of the valve. In this study, the cavitation index and percent of cavitation (POC) were introduced to numerically evaluate fluid flows via computational fluid dynamics (CFD) analysis. To reduce the cavitation effect generated by the bottom plug, the optimal shape design was carried out through a parametric study, in which a simple computer-aided engineering (CAE) model was applied to avoid time-consuming CFD analysis and difficulties in achieving convergence. The optimal shape design process using full factorial design of experiments (DOEs) and an artificial neural network meta-model yielded the optimal waist and tail length of the bottom plug with a POC value of less than 30%, which meets the requirement of no cavitation occurrence. The optimal waist length, tail length and POC value were found to 6.42 mm, 6.96 mm and 27%, respectively.

• Phonetics and Speech Sciences
• /
• v.12 no.2
• /
• pp.29-37
• /
• 2020

Recently, the neural network-based deep learning algorithm has dramatically improved performance compared to the classical Gaussian mixture model based hidden Markov model (GMM-HMM) automatic speech recognition (ASR) system. In addition, researches on end-to-end (E2E) speech recognition systems integrating language modeling and decoding processes have been actively conducted to better utilize the advantages of deep learning techniques. In general, E2E ASR systems consist of multiple layers of encoder-decoder structure with attention. Therefore, E2E ASR systems require data with a large amount of speech-text paired data in order to achieve good performance. Obtaining speech-text paired data requires a lot of human labor and time, and is a high barrier to building E2E ASR system. Therefore, there are previous studies that improve the performance of E2E ASR system using relatively small amount of speech-text paired data, but most studies have been conducted by using only speech-only data or text-only data. In this study, we proposed a semi-supervised training method that enables E2E ASR system to perform well in corpus in different domains by using both speech or text only data. The proposed method works effectively by adapting to different domains, showing good performance in the target domain and not degrading much in the source domain.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.2
• /
• pp.215-221
• /
• 2006

This paper is concerned with the control of multiple nonlinearities included in a humanoid robot system. A humanoid robot has some problems such as the structural instability, which leads to consider the control of multiple nonlinearities caused by driver parts as well as gear reducer. Saturation and backlash are typical examples of nonlinearities in the system. The conventional algorithms of backlash control were fuzzy algorithm, disturbance observer and neural network, etc. However, it is not easy to control the system by employing only single algorithm since the system usually includes multiple nonlinearities. In this paper, a switching Pill is considered for a control of saturation and a dual feedback algorithm is proposed for a backlash control. To implement the above algorithms, the system identification is firstly performed for the minimization of the difference between the results of simulation and experiment, and then the switching Pill gains are determined using genetic algorithm with some heuristic approach. The performance of the switching Pill controller for saturation and the dual feedback for backlash control is investigated through the simulation. Finally, it is shown that the implemented control system has good results and can be applied to the real humanoid robot system ISHURO.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.8
• /
• pp.1947-1954
• /
• 2013

Previous works for detecting arrhythmia have mostly used nonlinear method such as artificial neural network, fuzzy theory, support vector machine to increase classification accuracy. Most methods require accurate detection of P-QRS-T point, higher computational cost and larger processing time. But it is difficult to detect the P and T wave signal because of person's individual difference. Therefore it is necessary to design efficient algorithm that classifies different arrhythmia in realtime and decreases computational cost by extrating minimal feature. In this paper, we propose arrhythmia detection based on binary coding using QRS feature varibility. For this purpose, we detected R wave, RR interval, QRS width from noise-free ECG signal through the preprocessing method. Also, we classified arrhythmia in realtime by converting threshold variability of feature to binary code. PVC, PAC, Normal, BBB, Paced beat classification is evaluated by using 39 record of MIT-BIH arrhythmia database. The achieved scores indicate the average of 97.18%, 94.14%, 99.83%, 92.77%, 97.48% in PVC, PAC, Normal, BBB, Paced beat classification.

• The Korean Journal of Physiology and Pharmacology
• /
• v.2 no.3
• /
• pp.297-305
• /
• 1998

This study used in vivo intracellular recording in rat hippocampus to evaluate the effect of lidocaine and MK-801 on the membrane properties and the synaptic responses of individual neurons to electrical stimulation of the commissural pathway. Cells in control group typically fired in a tonic discharge mode with an average firing frequency of $2.4{\pm}0.9$ Hz. Neuron in MK-801 treated group (0.2 mg/kg, i.p.) had an average input resistance of $3.28{\pm}5.7\;M{\Omega}$ and a membrane time constant of $7.4{\pm}1.8$ ms. These neurons exhibited $2.4{\pm}0.2$ ms spike durations, which were similar to the average spike duration recorded in the neurons of the control group. Slightly less than half of these neurons were firing spontaneously with an average discharge rate of $2.4{\pm}1.1$ Hz. The average peak amplitude of the AHP following the spikes in these groups was $7.4{\pm}0.6$ mV with respect to the resting membrane potential. Cells in MK-801 and lidocaine treated group (5 mg/kg, i.c.v.) had an average input resistance of $3.45{\pm}6.0\;M{\Omega}$ and an average time constant of $8.0{\pm}1.4$ ms. The cells were firing spontaneously at an average discharge rate of $0.6{\pm}0.4$ Hz. Upon depolarization of the membrane by 0.8 nA for 400 ms, all of the tested cells exhibited accommodation of spike discharge. The most common synaptic response contained an EPSP followed by early-IPSP and late-IPSP. Analysis of the voltage dependence revealed that the early-IPSP and late-IPSP were putative $Cl^--and\;K^+-dependent$, respectively. Systemic injection of the NMDA receptor blocker, MK-801, did not block synaptic responses to the stimulation of the commissural pathway. No significant modifications of EPSP, early-IPSP, or late-IPSP components were detected in the MK-801 and/or lidocaine treated group. These results suggest that MK-801 and lidocaine manifest their CNS effects through firing pattern of hippocampal pyramidal cells and neural network pattern by changing the synaptic efficacy and cellular membrane properties.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.1
• /
• pp.169-177
• /
• 2020

In South Korea, the results of R&D in science and technology are submitted to the National Science and Technology Information Service (NTIS) in reports that have Korea national science and technology standard classification codes (K-NSCC). However, considering there are more than 2000 sub-categories, it is non-trivial to choose correct classification codes without a clear understanding of the K-NSCC. In addition, there are few cases of automatic document classification research based on the K-NSCC, and there are no training data in the public domain. To the best of our knowledge, this study is the first attempt to build a highly performing K-NSCC classification system based on NTIS report meta-information from the last five years (2013-2017). To this end, about 210 mid-level categories were selected, and we conducted preprocessing considering the characteristics of research report metadata. More specifically, we propose a convolutional neural network (CNN) technique using only task names and keywords, which are the most influential fields. The proposed model is compared with several machine learning methods (e.g., the linear support vector classifier, CNN, gated recurrent unit, etc.) that show good performance in text classification, and that have a performance advantage of 1% to 7% based on a top-three F1 score.