• Korea Petroleum Association Journal
• /
• no.5 s.212
• /
• pp.41-44
• /
• 1999

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.439-451
• /
• 2021

We propose a deep learning-based beamformer combined with spectral subtraction for continuous speech recognition operating in noisy environments. Conventional beamforming systems were mostly evaluated by using pre-segmented audio signals which were typically generated by mixing speech and noise continuously on a computer. However, since speech utterances are sparsely uttered along the time axis in real environments, conventional beamforming systems degrade in case when noise-only signals without speech are input. To alleviate this drawback, we combine online beamforming algorithm and spectral subtraction. We construct a Continuous Speech Enhancement (CSE) evaluation set to evaluate the online beamforming algorithm in noisy environments. The evaluation set is built by mixing sparsely-occurring speech utterances of the CHiME3 evaluation set and continuously-played CHiME3 background noise and background music of MUSDB. Using a Kaldi-based toolkit and Google web speech recognizer as a speech recognition back-end, we confirm that the proposed online beamforming algorithm with spectral subtraction shows better performance than the baseline online algorithm.

• Journal of Digital Convergence
• /
• v.10 no.1
• /
• pp.365-369
• /
• 2012

We propose a vision-based method that estimates train position and movement for railway monitoring in which we use a fuzzy classifier to determine train states. The proposed method employs frame difference and background subtraction for estimating train motion and presence, respectively. These features are used as the linguistic variables of the fuzzy classifier. Experimental results show that the proposed method can correctly estimate train position and movement. Therefore the method can be used for railway monitoring systems which estimate crowd density or protect safety.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.173-174
• /
• 2018

배경제거는 영상에서 움직이는 객체를 분리할 때 유용한 방법이며, 대표적인 예인 Mixture of Gaussian (MOG) 알고리즘은 픽셀 당 3-5 가우스 모델을 혼합해 배경과 움직이는 객체를 구분한다. 소형 표적을 추적하기 위해서는 화소 혹은 작은 블록 단위로 시/공간적 밝기 변화량을 이용하는 옵티컬 플로우 기법이 적절하다. 본 논문에서는 소형 표적의 강인한 객체 추적을 위해 MOG2와 옵티컬 플로우의 결합 방법을 소개한다. 제안된 방법은 MOG2를 사용하여 전경 영역을 획득하고 전경 영역에만 옵티컬 플로우를 적용한다. 실험 결과는 제안 방법이 잡음과 배경의 미세 변화가 있더라도 무인 비행체를 잘 추적할 수 있음을 보여준다.

• Journal of the Institute of Convergence Signal Processing
• /
• v.19 no.4
• /
• pp.192-198
• /
• 2018

Recently, the use of drone has been increasing rapidly in many ways. A drone can capture remote objects efficiently so it is suitable for surveillance and security systems. This paper discusses three methods for detecting moving vehicles using a drone. We compare three target detection methods using a background frame, preceding frames, or moving average frames. They are subtracted from a current frame. After the frame subtraction, morphological filters are applied to increase the detection rate and reduce the false alarm rate. In addition, the false alarm region is removed based on the true size of targets. In the experiments, three moving vehicles were captured by a drone, and the detection rate and the false alarm rate were obtained by three different methods and the results are compared.

• The Journal of the Korea Contents Association
• /
• v.6 no.9
• /
• pp.50-59
• /
• 2006

This paper proposes an automatic video monitoring system and its application to emergency detection by analyzing human behavior using neural network. The object area is identified by subtracting the statistically constructed background image from the input image. The identified object area then is transformed to the feature vector. Neural network has been adapted for analyzing the human behavior using the feature vector, and is designed to classify the behavior in rather simple numerical calculation. The system proposed in this paper is able to classify the three human behavior: stand, faint, and squat. Experiment results shows that the proposed algorithm is very efficient and useful in detecting the emergency situation.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.27 no.5
• /
• pp.426-431
• /
• 2007

When a high-energy ultrasound propagates through a solid body that contains a crack or a delamination, the two faces of the defect do not ordinarily vibrate in unison, and dissipative phenomena such as friction, rubbing and clapping between the faces will convert some of the vibrational energy to heat. By combining this heating effect with infrared imaging, one can detect a subsurface defect in material in real time. In this paper a realtime detection of the brazing defect of thin Inconel plates using the UIR (ultrasonic infrared imaging) technology is described. A low frequency (23 kHz) ultrasonic transducer was used to infuse the welded Inconel plates with a short pulse of sound for 280 ms. The ultrasonic source has a maximum power of 2 kW. The surface temperature of the area under inspection is imaged by an infrared camera that is coupled to a fast frame grabber in a computer. The hot spots, which are a small area around the bound between the two faces of the Inconel plates near the defective brazing point and heated up highly, are observed. And the weak thermal signal is observed at the defect position of brazed plate also. Using the image processing technology such as background subtraction average and image enhancement using histogram equalization, the position of defective brazing regions in the thin Inconel plates can be located certainly.

• Journal of KIISE:Software and Applications
• /
• v.33 no.11
• /
• pp.942-952
• /
• 2006

We propose an automatic segmentation method for identifying pulmonary structures using gray-level information of chest CT images. Our method consists of following five steps. First, to segment pulmonary structures based on the difference of gray-level value, we select the threshold using optimal thresholding. Second, we separate the thorax from the background air and then the lungs and airways from the thorax by applying the inverse operation of 2D region growing in chest CT images. To eliminate non-pulmonary structures which has similar intensities with the lungs, we use 3D connected component labeling. Third, we segment the trachea and left and right mainstem bronchi using 3D branch-based region growing in chest CT images. Fourth, we can obtain accurate lung boundaries by subtracting the result of third step from the result of second step. Finally, we select the threshold in accordance with histogram analysis and then segment radio-dense pulmonary vessels by applying gray-level thresholding to the result of the second step. To evaluate the accuracy of proposed method, we make a visual inspection of segmentation result of lungs, airways and pulmonary vessels. We compare the result of the conventional region growing with the result of proposed 3D branch-based region growing. Experimental results show that our proposed method extracts lung boundaries, airways, and pulmonary vessels automatically and accurately.