비디오에 의한 감시 및 관측(VSAM) 시스템을 위한 사람의 계수기법

도용태 한국센서학회 2002 센서학회지 Vol.11 No.1

People are important targets for video surveillance and monitoring(VSAM) but difficult to be analyzed. In this paper, a technique to count people in image sequences is dealt as a prerequisite pro cede for automatic tracking and behaviour analysis. A group of people is divided at local minima of the line connecting the highest pixels on, the binary image of the people extracted from the image taken by a stationary video camera. As the properties of the divided regions vary according to the relative positions of the people in a group, different states are assigned for the completely occluded, partially occluded, completed separated individual, and wrongly divided regions. By analyzing the transition of the states of divided regions, the number of people on the site monitored is red. The technique is checked in real experimental situations.

소수 데이터의 신경망 학습에 의한 카메라 보정

도용태 한국센서학회 2019 센서학회지 Vol.28 No.3

When a camera is employed for 3D sensing, accurate camera calibration is vital as it is a prerequisite for the subsequent steps of the sensing process. Camera calibration is usually performed by complex mathematical modeling and geometric analysis. On the other contrary, data learning using an artificial neural network can establish a transformation relation between the 3D space and the 2D camer image without explicit camera modeling. However, a neural network requires a large amount of accurate data for its learning. A significantly large amount of time and work using a precise system setup is needed to collect extensive data accurately in practice. In this study, we propose a two-step neural calibration method that is effective when only a small amount of learning data is available. In the first step, the camera projection transformation matrix is determined using the limited available data. In the second step, the transformation matrix is used for generating a large amount of synthetic data, and the neural network is trained using the generated data. Results of simulation study have shown that the proposed method as valid and effective.

초음파 거리 센서의 계측오차 감소를 위한 연구

도용태,김태호,유석환 대한전자공학회 1997 電子工學會論文誌, S Vol.s34 No.11

Ultrasonic sensors are widely employed in detecting range to a target by the virtue of their low cost and simplicity. However, the sensor's measurements are corrupted by systematic errors due mainly to the dependency of sound speed upon surrounding conditions and random errors of uncertain origin. In this paper, we present the results of research carried out to reduce these errors for increasing the reliability of an untrasonic sensor system to be used in orbotic or other automated system's range finding. The sensor system designed herein is in a peuliar structure having a reference target and two receivers. Echoes from a small reference target placed at a known distance are used for compensating the variations of sound speed according to the changes of sensing conditions. Unlike existing ones, the technique proposed can compensate the effects of temperature or any other physical parameters without an additional sensor dedicated to the compensation. The measurements by two redundantly employed receivers are fused to reduce random errors in a statistical sense. The correlation of the signals from the receivers sharing a hardware in part is considered in the fusion process. The methodology desicribed in this paepr is conceptually simple, easy to be implemented, and effetive to increase the accuracy of the sensor measurements as experimental results confirm.

시각기반 웜 자세의 기구학적 모형화

도용태,탄 콕 키옹 제어·로봇·시스템학회 2015 제어·로봇·시스템학회 논문지 Vol.21 No.3

We present a novel method to model the body posture of a worm for vision-based automatic monitoring and analysis. The worm considered in this study is a Caenorhabditis elegans (C. elegans), which is popularly used for research in biological science and engineering. We model the posture by an open chain of a few curved or rigid line segments, in contrast to previously published approaches wherein a large number of small rigid elements are connected for the modeling. Each link segment is represented by only two parameters: an arc angle and an arc length for a curved segment, or an orientation angle and a link length for a straight line segment. Links in the proposed method can be readily related using the Denavit-Hartenberg convention due to similarities to the kinematics of an articulated manipulator. Our method was tested with real worm images, and accurate results were obtained.

확률적 신경망을 사용한 중첩된 사람 영상의 분할

도용태 Springer 2006 Lecture Notes in Computer Science Vol.4113 No.-

3차원 위치측정을 위한 스테레오 카메라 시스템의 인공 신경망을 이용한 보정

도용태,이대식,류석환 ( Yong Tae Do,Dae Sik Lee,Seog Hwan Yoo ) 한국센서학회 1998 센서학회지 Vol.7 No.6

Stereo cameras are the most widely used sensing systems for automated machines including robots to interact with their three-dimensional(3D) working environments. The position of a target point in the 3D world coordinates can be measured by the use of stereo cameras and the camera calibration is an important preliminary step for the task. Existing camera calibration techniques can be classified into two large categories - linear and nonlinear techniques. While linear techniques are simple but somewhat inaccurate, the nonlinear ones require a modeling process to compensate for the lens distortion and a rather complicated procedure to solve the nonlinear equations. In this paper, a method employing a neural network for the calibration problem is described for tackling the problems arisen when existing techniques are applied and the results are reported. Particularly, it is shown experimentally that by utilizing the function approximation capability of multi-layer neural networks trained by the back-propagation(BY) algorithm to learn the error pattern of a linear technique. the measurement accuracy can be aimply and efficiently increased.

스테레오 화상점으로부터 삼차원 점의 직접적 회복

도용태 대구대학교 산업기술연구소 1994 産業技術硏究 Vol.13 No.-

본 논문에서는 정합된 두 화상점의 좌표로부터 직접적으로 삼차원 점의 좌표를 결정하는 기법이 제안된다. 기존의 기법들과는 달리 두 카메라는 동일선상에 있지 않아도 되며, 각 카메라가 보정되기 보다는 스테레오 장치 자체가 보정되었다. 바늘구멍 카메라 모델에 기초한 선형시스템의 가정이 사용되었으며, 이러한 가정은 측정의 정확도에 한계를 가져오는 대신 측정속도를 향상시킬 수 있다. 따라서 제안된 기법의 주된 응용영역은 실시간 계측이 특히 중요한 로보트를 비롯한 자동화 시스템의 3차원 작업이 될 것이다. 제안된 기법은 기존의 방식과 비교함으로서 검토되었으며, 실험을 통해 유용성이 입증되었다.

의료용 혈압계의 소형화와 지능형 계측 프로그램 개발

도용태 대구대학교 공과대학 1998 산학연 공동기술개발컨소시엄 최종보고서 Vol. No.

방사형 렌즈 왜곡에 기인한 카메라 보정 오차의 수정

도용태 대구대학교 정보통신연구소 2003 情報通信硏究 Vol.1 No.2

카메라의 보정은 3차원 실세계와 2차원 영상 사이의 관계를 결정짓는 중요한 절차로 많은 3차원 기계시각의 문제에서 기본이 된다. 기존의 카메라 보정 기법은 선형과 비선형의 해법을 사용한 것들로 크게 나눌 수 있는데, 선형의 기법이 간단하고 빠른 장점이 있는 반면 시스템에 존재하는 비선형성이 클 경우 이를 무시함으로서 생기는 오차가 문제가 될 수 있다. 카메라의 비선형성은 주로 렌즈의 왜곡에 의하며, 특히 방사형 왜곡이 주요한 요소임이 보고된 바 있다. 본 논문에서는 방사형 왜곡을 수정하기 위한 새로운 기법을 제안한다. 제안된 방식은 선형의 보정기법으로 얻어진 출력을 보상하여 실제 카메라 출력을 얻기 위한 목적으로 사용될 수 있으며, 이는 모의실험을 통하여 확인할 수 있었다. Camera calibration is an important procedure determining the relationship between 3D real world and 2D image. This is thus a prerequisite for many 3D machine vision problems. Existing camera calibration techniques can be categorized largely into two groups, linear or nonlinear. Linear techniques are simple and fast but problematic because of the error due to ignoring nonlinearity existing in the system when it is large. The nonlinearity of a camera is caused mainly by lens distortion and it is reported that especially radial distortion is a major factor. In this paper, a new technique is proposed for correcting the radial distortion. The proposed technique can be used to compensate for the output of a linear calibration technique and to get real camera output as shown in simulation.

영상내 사람 영역의 분할기법

도용태 대구대학교 정보통신연구소 2003 情報通信硏究 Vol.1 No.2

본 논문에서는 카메라에 의해 얻어진 영상에서 사람 영역을 분할하기 위한 기법을 제시한다. 우선 배경 영역을 적응적으로 기억한 후, 여기에 사람이 출연하여 생기는 전경 영역을 추출하구 이어 전경 영역을 각 개인별로 구분하는 작업을 차례로 행하게 된다. 제시된 방법은 추출된 이진 전경 영역의 수직 투사 그래프를 얻고 이의 지역최소를 찾아 분할하는 것이다. 각 분할은 화실성의 척도를 정의함으로서 그 신뢰도를 추정하였다. 카메라의 조망 구역에서 전경 영역은 사람에 의해서만 생긴다는 가정이 사용되었고, 제안된 방식은 실제 실험을 통하여 그 유용성을 점검하였다. In this paper a technique for dividing the region of people in images obtained by a camera is proposed. First, the background region is adaptively stored and the foreground region made by people appear on the image is extracted in the next step. The foreground region is divided then into segments of individual people. The proposed method obtains a graph by vertical projection of the binary foreground image and divides the image at the local minima of the graph. A certainty measure is defined to estimate the credence of division. The foreground region is assumed here to be made by only people. The method described is checked in actual experiments.