http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2010 한국정보기술학회논문지 Vol.8 No.1
In this paper, the proposed method is based on detecting moving region by applying a combined three frames difference and background subtraction. We can adaptively extract the object which is not sensitive to change of illumination and has a fast motion or slow motion by applying these two methods. In preprocess we used gaussian filter in order to obtain clear image. Moving object region was detected in using the proposed method from the image which is removed noise. About background modeling method we used MOG(Mixture Of Gaussian) method, to modeling and update the background so that applies to real-time change of background. The processing time for each frame in the experimental results of proposed method is 33.012ms, and precision is 93%. The experimental results show that the proposed method is more robust, accurate and powerful than existed methods.
색상과 Texture 정보를 이용한 Particle Filter 기반 객체 추적 방법
안명수(Ming-Shou An),하성욱(Seong-Wook Ha),강대성(Dae-Seong Kang) 한국정보기술학회 2010 한국정보기술학회논문지 Vol.8 No.11
In particle filter, the sampling initialization has many noises. Due to this problem, there are also many noises in weight update step. Therefore, it is important that extracting the feature for sampling. So far, the feature extraction method with HSV color information is already widely used. However, it is difficult to extract the feature accurately only with the color information. In our research, we proposed the method that is fusing the HSV color information and texture information to feature detection and object tracking. For this feature, we use the Particle filter to tracking effectively in brightness changes or complex background. In the experimental results, we confirmed that the proposed method can increase the precision of the Particle filter, and has the good robustness to object tracking.
기상 및 비전 센서 기반의 산불 방재 시스템에 관한 연구
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2014 한국정보기술학회논문지 Vol.12 No.6
The forest fires and weather conditions have a close relationship. The meteorological factors which make impact on forest fire are relative humidity, temperature, wind speed, recipitation and so on. In this paper, we proposed an effective fire monitoring system by utilizing the meteorological information and image information. First, the sample of forest fire image converted from RGB color space to HSI color space, and a H-S histogram model generated by color characteristics of forest fire. And that, for the input image, H-S histogram patch applied to detect the forest fire based on histogram back-projection method. Then, we can aware the weather environment of location of forest fire occurred by analyzing the temperature and humidity data, wind direction and speed data collected from meteorological observation instruments. And it can be used as useful information to predict the direction and magnitude of forest fire for possible expansion. Through the experimentation, we evaluated the performance of meteorological observation instruments, and confirmed the effectiveness of fire detection algorithm.
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2014 한국정보기술학회논문지 Vol.12 No.11
In this paper, we proposed a method of object detection based on analysis of integral feature of region of interest (ROI). First, the background modelling algorithm was applied to detect the foreground and define as a ROI. Second, we used integral information analysis method to define a more accurate region of object. In this method, in order to obtain the information of the optimized region of object, the integral image features only analyzed by Haar-like features and a simple subtraction operation which have less computation was performed. So, our proposed method has those advantages such as low complexity and high accuracy. In experimental results, through a variety of results of object detection the effectiveness of the proposed method also demonstrated.
몰입형 인터페이스를 위한 손동작 인식 알고리즘에 관한 연구
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2012 한국정보기술학회논문지 Vol.10 No.3
Immersive hand gesture interface is mostly used for augmented reality and kinect game environment which are introduced recently. Implementing hand gesture interface in this domain requires not only the region of hand must be recognized quickly in real time, but also a high degree of accuracy. In this paper, the hand detection and tracking methods are proposed for immersive hand gesture interface. For detecting the region of hand correctly, the detection method uses the data trained by Adaboost algorithm with Haar-like feature. We improved the pairwise geometrical histogram(PGH) algorithm for recognizing the hand. The advantage of this algorithm is that processing speed is high due to efficient calculation and the recognition accuracy is high. Experimental results show that the recognition rate for test image data is high.
다 특징 정보 융합에 의한 스테레오 기반 다중 객체 추적 기법
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2011 한국정보기술학회논문지 Vol.9 No.7
In this paper, we propose a method that can be effectively tracking the multi-object based on stereo vision by the information fusion of color, depth and texture. A CAMShift algorithm based on the color is an representative algorithm for the object tracking. But, it is not an effective method for multi-object tracking only by backprojection image information. So, we used three features to describe the independent features of object. The first feature is the backprojection image information; the second is the depth map by stereo camera; the third is HLBP(histogram of local binary patterns). We use the method of the three features fusion through CAMShift algorithm for tracking. It is robust to the illumination changes and occlusion problems. The comparative results from experiments show the proposed method can extract the independent features of object, and track effectively.
풍력발전기 모니터링 시스템을 위한 WSN 기반 발전기 고장 신호 분석 알고리즘
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2013 한국정보기술학회논문지 Vol.11 No.6
With the development of science and technology, the offshore wind power is to become a main of wind power field. Although the large-scale offshore wind farm will be achieved, but its installation and maintenance are increasingly difficult. So it is necessary to research the fault diagnosis technology for offshore wind turbines. In this paper, we used the WSN(Wireless Sensor Network) in order to overcome the environment of positional constraints. And then we collect measured signal data from distributed nodes of the installed ethernet gateway within wind turbine farms and analyze those data in center monitoring system. The real-time signal analysis and pattern feature extraction were made through wavelet transform. The information of classified signal pattern is used to implement an automatic fault diagnosis system by using neural network algorithm.
개선된 MOG 모델과 예측 방법에 의한 실시간 움직임 객체 검출 및 추적에 대한 연구
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2009 한국정보기술학회논문지 Vol.7 No.4
The analysis of the moving object is based on the detection system and the problem which is to have a challenge and widely applied in the tracking and safety. how to detect moving objects has recently studied a lot. In this paper, MOG (Mixture Of Gaussian), a way of Modeling the background, is described using statistical characteristics of the information obtained over time surveillance zone. And the Modeling background and the object are detected by proposing how to improve the MOG (Mixture Of Gaussian) of the disadvantages. The noise process is practiced by applying the operation morphology in extracted object image. The experiment to track the moving object is progressed by calculating the special features of detected objects accurately and appling the block matching and the Kalman Filter prediction methods for proving.
HHT 기반 신경망을 이용한 풍력터빈 고장진단 방법에 대한 연구
안명수(Ming-Shou An),강대성(Dae-Seong Kang) 한국정보기술학회 2013 한국정보기술학회논문지 Vol.11 No.11
For the wind power, the global research has been activating, and the large scale has been implementing with the development of technology. But due to the development of large-scale structure of wind turbines, the possibility of the failure was increased. So the most effective way is real time condition monitoring and advanced fault diagnosis to ensure the stable operation and reduce the maintenance costs. In this paper, the Hilbert-Huang Transform(HHT) which is the signal analysis for fault diagnosis extracts the feature information from the classified patterns of normal signal and fault signal. The extracted feature information would be used for input layer of neural network algorithm. The failure or not can be identified through the neural network training. In addition, simple faults as well as complex faults that occur over a long time can be diagnosed by proposed method of fault analysis early.