A5000296697- Face and Expression Recognition
- Face recognition and analysis
- Biometric Identification and Security
- Neural Networks and Applications
- Video Surveillance and Tracking Methods
- Advanced Algorithms and Applications
- Image Processing Techniques and Applications
- Vehicle License Plate Recognition
- Advanced Vision and Imaging
- Image and Object Detection Techniques
- Internet of Things and Social Network Interactions
- Handwritten Text Recognition Techniques
- Fuzzy Logic and Control Systems
- Advanced Sensor and Control Systems
- 3D Shape Modeling and Analysis
- Industrial Vision Systems and Defect Detection
- Image and Video Stabilization
- Advanced Image and Video Retrieval Techniques
- Fire Detection and Safety Systems
- Blind Source Separation Techniques
- Video Coding and Compression Technologies
- Tactile and Sensory Interactions
- Gaze Tracking and Assistive Technology
- Remote-Sensing Image Classification
- Teleoperation and Haptic Systems
Pohang University of Science and Technology
2017-2019
Seoul National University
2019
Hongik University
1999-2014
Soongsil University
2008-2014
Korea Atomic Energy Research Institute
2005
Purdue University West Lafayette
1992
This paper presents a data-driven thermal rendering framework for highly realistic re-creation of the dynamics real objects in virtual environment. We begin with physics-based model and then extend it to compute desired temperature display without identification parameters objects. Our approach is data-driven: we collect responses between hand object many different combinations initial temperatures interpolate several render stimuli target temperatures. also present perceptual evaluation...
In this paper, we propose a new real-time human detection under omni-directional cameras for visual surveillance purpose, based on CNN with unified and AGMM. Compared to CNN-based state-of-the-art object methods. YOLO model-based method boasts of very fast detection, but less accuracy. The proposed adapts the detecting model so as be intensified by additional foreground contextual information obtained from pre-stage Increased computational time incurred AGMM processing is compensated...
A method for obtaining the power energy with high quality is to keep factor a load as close unity feasible. In this paper, we present new improve load. The proposed uses fuzzy control techniques in order determine how many parallel capacitors are be connected correction of factor.
In this paper, we propose an audio surveillance system to detect and localize dangerous sound in real-time so as be able direct a PTZ camera catch snapshot image about the location of source instantly. The proposed firstly detects foreground based on adaptive Gaussian mixture background model, classifies it into one pre-trained classes sounds GMM model. Next, decides whether belongs class group or not. If does, then localization algorithm Dual delay-line is applied source. Finally, uses...
We propose a data-driven method for realistic texture rendering on an electrovibration display. To compensate the nonlinear dynamics of display, we use autoregressive with external input (NARX) neural networks as inverse model The are trained lateral forces resulting from actuating display pseudo-random binary signal (PRBS). collected textured surface various scanning velocities and normal fed into network to generate actuation For arbitrary velocity force, apply two-step interpolation...
In this paper, we propose an efficient and robust lane detection method for detecting immediate left right boundaries of the in roads. The proposed are based on hyperbolic model reliable line segment clustering. cluster is determined from most probable obtained clustering segments extracted by LSD algorithm. Experiments show that works robustly against lanes with difficult environments such as ones occlusions or cast shadows addition to dashed marks, performs better compared other methods...
In this paper, we propose an effective design and implementation of a versatile embedded multimedia live streaming decoder system based on TI's DM814x processor, which can not only decode video streams various codecs(H.264, MPEG-4, etc.), audio codecs(AAC, MP3, G.711, but also demultiplex (demux) MPEG-2 TS, stream inside, all carried over RTP after the servers are contacted by RTSP URLs. Those decoded raw displayed out through attached HDMI, HD-SDI, NTSC, line-out interface decoder. The S/W...
Conventional neural networks utilize all the dimensions of original input patterns for training and classification. However, a particular attribute does not necessarily contribute to classification may even cause misclassification in certain cases. A new ensemble competitive learning method using reduced dimension is proposed. In contrast previous which adjust parameters, proposed takes advantage information each patterns. Since degree contribution known beforehand, different data sets with...
눈좌표 검출은 얼굴 인식 및 관련된 응용 분야 등에서 필요한 작업이다. 현재까지 보고된 대부분의 검출 방법은 성공적인 적용을 위해서는 여전히 정확도 속도의 개선을 필요로 한다. 본 논문에서는 다중스케일 가버 특징 벡터 모델 기반의 개선된 방법을 제안한다. 제안된 먼저 다운샘플링된 입력 이미지에서 초기 눈좌표에서의 벡터와 해당 스케일의 눈 번치와의 가버젯 유사도를 이용하여 눈좌표를 추정한다. 이후 추정된 상위 이미지에서의 초기값으로 취하고 스케일 같은 방법으로 찾으며, 이를 반복적으로 하여 최종적으로 원래 확정한다. 실험을 통해, 논문에서 제안한 기반 방법이 계산량은 크게 증가시키지 않으면서 기존 연구들에서 다른 방법에 비해 정확도가 방법임을 확인하였다. Eye localization is necessary for face recognition and related application areas. Most of eye algorithms reported thus far...
Current IP network cameras support live video and audio streaming through separate channels so that it is not easy to achieve video/audio synchronization at clients. In this paper, we propose a design implementation of versatile multimedia for DM36x-based camera which can only existing various formats (H.264, MPEG4, MJPEG) in many sizes (AAC, PCM) but also MPEG2-TS. order MPPEG2-TS streaming, first modify some the MPEG2-TS related classes LIVE555 (an open source framework) implement new SW...
본 논문은 얼굴인식 시스템 상에서 마스크를 착용한 변장이미지가 입력 감지될 경우 나머지 노출된 부분의 특징만을 가지고 가려진 사람의 신원을 추정하는 방법을 기술한다. 얼굴영역 검출 후에 마스크상단의 눈 주변 이미지만을 특징점 추출을 실시하여 등록된 얼굴 인증 데이터 베이스와의 비교를 통해 추정한다. 매칭에 쓰일 추출에는 조명에 강인하고 영상의 크기와 회전에도 변하지 않는 특성을 가진 SIFT(Scale Invariant Feature Transform) 알고리즘을 이용한다. 매칭을 정확한 매칭률은 전체 실험결과를 평가한다. This paper presents a new human identification algorithm using partial features of the uncovered portion face when person wears mask. After area is detected, feature extracted from eye above The...
Zoom tracking involves the automatic adjustment of focus motor in response to zoom movements for purpose keeping an object interest focus, and is typically achieved by moving motors a lens module so as follow so-called "trace curve", which shows in-focus positions versus specific distance. Thus, one can simply implement following most closest trace curve after all data are stored memory. However, this approach often prohibitive practical implementation because its large memory requirement....
기존 디지털 출력 방식의 PIR 센서를 이용한 침입감지 시스템은 사람이 아닌 다른 물체에 대한 침입 탐지 오류가 많았다. 본 논문은 이를 극복하기 위하여 아날로그 센서 기반 감지 시스템을 제안한다. 방식 센서는 임계값을 기준으로 이진 출력값 대신, 일정 범위 내의 다양한 전압 준위로 출력값을 내보낸다. 이용하여 획득된 신호의 샘플링된 신호값으로부터 FFT(Fast Fourier Transform) 또는 MFCC(Mel-frequency cepstrum codfficents)을 주파수 성분을 추출하여, 인공 신경회로망(Artificial Neural Network)의 특징벡터로 사용된다. 인간의 움직임과 애완동물의 움직임에 신호 패턴들을 학습한 신경회로망을 통해서 침입상황에서 침입한 객체가 사람인지 애완동물인지 판별하게 된다. The intruder detection system using digital sensor has the problem that it can't...