-
-
公开(公告)号：KR101434170B1
公开(公告)日：2014-08-26
申请号：KR1020120106685
申请日：2012-09-25
Applicant: 한국전자통신연구원
CPC classification number: G06N99/005 , G06K9/6231 , G06K9/6235 , G06K9/6256 , G06K9/6282 , G06K2009/6236
Abstract: 학습 과정을 단순화하거나 추정의 정확도를 높이기 위한 추출된 데이터 특징을 이용한 학습 방법이 개시된다. 추출된 데이터 특징을 이용한 학습 방법은 입력된 학습 데이터를 미리 설정된 기준으로 두 개의 그룹으로 분할하는 단계와, 분할된 두 개의 그룹을 구분할 수 있는 데이터 특징을 추출하는 단계 및 추출된 특징을 이용하여 학습을 수행하는 단계를 포함한다. 학습 데이터를 단계적으로 세분하여 각 그룹을 용이하게 구분할 수 있는 특징을 추출하기 때문에 학습 과정을 단순화 시킬 수 있으며, 추정의 정확도를 높일 수 있다.
-
公开(公告)号：KR1020140090795A
公开(公告)日：2014-07-18
申请号：KR1020130002836
申请日：2013-01-10
Applicant: 한국전자통신연구원
CPC classification number: G06K9/00295 , G06K9/00335 , G06T7/246 , G06T7/292 , G06T2207/10016 , G06T2207/30196 , G06T2207/30232 , H04N5/23296 , H04N5/247 , H04N7/181 , H04N7/188
Abstract: The present invention relates to a method for tracking an object in a multi-camera environment. According to the present invention, the method for tracking an object in a multi-camera environment comprises the steps of: generating a first feature information of an object from an image inputted from a first camera; detecting a second camera in which the identification information of the object is recognized when the object escapes from a photographing area of the first camera; and comparing the first feature information with a second feature information of an object generated from an image which is inputted from the second camera and tracking the object in the image inputted from the second camera. The present invention performs image-based object tracking in an image obtained by a camera, and recognizes the identification information of a terminal owned by the object, hands over the camera, and continuously tracks the same object, when the object moves out of the camera. The camera can be handed over quickly and accurately by recognizing the identification information of the terminal and handing over the camera. This contributes to the improvement of object tracking performance.
Abstract translation: 本发明涉及一种用于跟踪多摄像机环境中的物体的方法。 根据本发明，用于在多摄像机环境中跟踪对象的方法包括以下步骤：从从第一摄像机输入的图像生成对象的第一特征信息; 检测当所述对象从所述第一相机的拍摄区域逸出时识别所述对象的识别信息的第二相机; 以及将所述第一特征信息与从所述第二照相机输入的图像生成的物体的第二特征信息进行比较并且跟踪从所述第二照相机输入的图像中的物体。 本发明在通过照相机获得的图像中执行基于图像的对象跟踪，并且当对象从相机移出时识别对象所拥有的终端的识别信息，双手相机，并且连续地跟踪相同的对象 。 通过识别终端的识别信息并交出相机，即可快速准确地切换相机。 这有助于提高对象跟踪性能。
-
公开(公告)号：KR1020140076815A
公开(公告)日：2014-06-23
申请号：KR1020120145282
申请日：2012-12-13
Applicant: 한국전자통신연구원
CPC classification number: G06T7/20 , G06T2207/10016 , G06T2207/30232
Abstract: The present invention relates to an abnormal behavior recognizing technology in a CCTV image monitoring service. A pixel-based abnormal motion detecting method according to the present invention includes the following steps of: generating motion vector information about each pixel of an inputted image; generating motion information in the image by using the generated motion vector information; and determining whether an abnormal motion exists or not by comparing the generated motion information with predetermined or pre-learned reference motion information. The present invention is able to recognize an abnormal situation in which many people abnormally move by a sudden accident or event in a CCTV image through an abnormal behavior detection based on the pixel information analysis of an inputted image.
Abstract translation: 本发明涉及CCTV图像监视服务中的异常行为识别技术。 根据本发明的基于像素的异常运动检测方法包括以下步骤：产生关于输入图像的每个像素的运动矢量信息; 通过使用所生成的运动矢量信息来生成图像中的运动信息; 以及通过将所生成的运动信息与预先确定的或预先学习的参考运动信息进行比较来确定是否存在异常运动。 本发明能够通过基于输入图像的像素信息分析的异常行为检测来识别许多人因CCTV图像中的突然事故或事件而异常移动的异常情况。
-
公开(公告)号：KR1020140035660A
公开(公告)日：2014-03-24
申请号：KR1020120102186
申请日：2012-09-14
Applicant: 한국전자통신연구원
IPC: G06T7/00
CPC classification number: G06K9/3233 , G06T7/251 , G06T7/254
Abstract: A method for extracting an object region of the present invention comprises the steps of: extracting an object ROI region from an input image; generating panoramic information for the input image via background modeling; generating an object mask using the generated panoramic information; and splitting the object region by matching the object ROI region and the object mask. [Reference numerals] (102) ROI region extraction block; (104) Front view generation block; (1062) Image reconstruction unit; (1064) Object mask generation unit; (1082) Binary image generation unit; (1084) Binary mask generation unit; (1086) Object region separation unit; (AA) Input image
Abstract translation: 本发明的提取对象区域的方法包括以下步骤：从输入图像中提取对象ROI区域; 通过背景建模产生输入图像的全景信息; 使用所生成的全景信息生成对象掩模; 并通过匹配对象ROI区域和对象掩码来分割对象区域。 （102）ROI区域提取块; （104）前视图生成块; （1062）图像重建单元; （1064）对象掩码生成单元; （1082）二进制图像生成单元; （1084）二进制掩码生成单元; （1086）对象区域分离单元; （AA）输入图像
-
Patent Agency Ranking
公开(公告)号：KR1020130067842A
公开(公告)日：2013-06-25
申请号：KR1020110134815
申请日：2011-12-14
Applicant: 한국전자통신연구원
IPC: G06T7/00
Abstract: PURPOSE: An image-based object information detecting method is provided to offer distance information of human objects and height information of a person in addition to the existing image information by using a single camera, thereby capable of being utilized for an intelligent monitoring service or a pedestrian-vehicle collision preventing service. CONSTITUTION: An object information detecting device forms a distance reference pattern in a recorded image. The device detects an object contained in the image with an object pattern corresponding to the size of the object(S20). The device detects a distance mapped on a point corresponding to the location information of the object pattern from the distance reference pattern(S50). The device detects actual height mapped on the height of the object pattern from the detected distance(S60). [Reference numerals] (AA) Start; (BB) End; (S10) Take images from object; (S20) Detect object; (S30) Detect location information; (S40) Extract point corresponding to location information of object information among distance based pattern; (S50) Detect distance; (S60) Detect height of object pattern in detected distance; (S70) Detect actual height; (S80) Display distance with object and actual height
Abstract translation: 目的：提供一种基于图像的对象信息检测方法，通过使用单个照相机，除了现有的图像信息之外，还提供人物的距离信息和人物的高度信息，从而能够用于智能监视服务或 行人车辆防撞服务。 构成：物体信息检测装置在记录图像中形成距离基准图案。 该装置利用与该物体的尺寸对应的对象图案来检测图像中包含的对象（S20）。 设备从距离参考图案检测对应于对象图案的位置信息的点处的距离（S50）。 该设备从检测到的距离检测对象图案的高度上映射的实际高度（S60）。 （附图标记）（AA）开始; （BB）结束; （S10）从物体拍摄图像; （S20）检测对象; （S30）检测位置信息; （S40）基于距离的图案对应于对象信息的位置信息的提取点; （S50）检测距离; （S60）以检测距离检测物体图案的高度; （S70）检测实际高度; （S80）显示与物体的距离和实际高度
-
公开(公告)号：KR101151435B1
公开(公告)日：2012-06-01
申请号：KR1020090108441
申请日：2009-11-11
Applicant: 한국전자통신연구원
Abstract: 본발명의얼굴인식방법은, 얼굴영상내에서지역적특징을추출하는단계; 추출된지역적특징을이용하여얼굴영상을표현하는단계; 표현된얼굴영상내에서각 영역을순차적으로분할하는단계; 및분할된각 영역들에대한히스토그램을생성하는단계; 및생성된히스토그램을통해템플릿매칭을수행하여얼굴을인식하는단계를포함한다.
-
公开(公告)号：KR1020110132834A
公开(公告)日：2011-12-09
申请号：KR1020100052393
申请日：2010-06-03
Applicant: 한국전자통신연구원
CPC classification number: G06K9/6215 , A61B5/1172 , G06F11/10
Abstract: PURPOSE: A method for recognizing a fingerprint is provided to protect an original feature point from a fingerprint by adding a plurality of fake fingerprints. CONSTITUTION: A reference feature point(310) is selected from the extracted original feature point. A feature shape is formed based on the information of the selected reference feature point. A space between feature points(330) is calculated by determining the number of fake feature points to be added to the shape. A fingerprint template is generated by adding the fake feature points with the calculated space.
Abstract translation: 目的：提供一种用于识别指纹的方法，通过添加多个假指纹来保护原始特征点免受指纹的影响。 构成：从提取的原始特征点中选择参考特征点（310）。 基于所选择的参考特征点的信息形成特征形状。 通过确定要添加到形状的假特征点的数量来计算特征点（330）之间的空间。 通过将假特征点与计算空间相加来生成指纹模板。
-
公开(公告)号：KR1020110109147A
公开(公告)日：2011-10-06
申请号：KR1020100028739
申请日：2010-03-30
Applicant: 한국전자통신연구원
CPC classification number: G08B13/19669 , G08B13/1966 , G08B25/10 , H04N7/18
Abstract: 본 발명은 방범 시스템에 관한 것으로, 영상 촬영 장치를 통해 촬영된 촬영 영상이 현장 제어 장치에 입력되면 해당 촬영 영상에서 이벤트가 발생하는지를 체크하고, 이벤트가 발생하지 않은 경우 네트워크 전송 용량에 따라 축소 인코딩된 촬영 영상을 실시간으로 전송하여 출력하거나, 이벤트가 발생한 경우 해당 촬영 영상에 대한 이벤트 정보, 클리핑 영상 및 축소 인코딩된 촬영 영상을 전송하고, 이러한 이벤트 발생에 따른 이벤트 경보를 원격 관제 장치에서 출력하며, 해당 클리핑 영상 및 촬영 영상을 출력함으로써, 고해상도 영상을 출력할 수 있는 방범 시스템을 쉽게 구축할 수 있는 것이다.
-
公开(公告)号：KR1020110067480A
公开(公告)日：2011-06-22
申请号：KR1020090124090
申请日：2009-12-14
Applicant: 한국전자통신연구원
CPC classification number: G06K9/00281 , G06K9/6212 , G06K9/6234 , G06K2009/4666
Abstract: PURPOSE: A feature points detecting method for detecting a face is provided to increase the accuracy of detecting a face by learning LBP feature points having various points and size. CONSTITUTION: A feature points detecting method for detecting a face comprises the steps of: assigning an initial value according to the kind of an input image, face image or non-face image(400); setting the weight value of the input image according to the number of face images and non-face images in the input images(S402); standardizing the weight value(S404); calculating an error value of the feature points using the weak learning machine for the feature points, the standardized weight value, and the initial value(S406); selecting a weak learning machine having a least error value(S408); checking whether the input image is determined accurately on the basis of the weak learning machine(S410); and changing the weight value depending on the check results(S416).
Abstract translation: 目的：提供用于检测脸部的特征点检测方法，以通过学习具有各种点和大小的LBP特征点来提高对脸部的检测精度。 构成：用于检测面部的特征点检测方法包括以下步骤：根据输入图像的种类，面部图像或非脸部图像（400）分配初始值; 根据输入图像中的脸部图像和非脸部图像的数量来设置输入图像的权重值（S402）; 标准化重量值（S404）; 使用弱学习机计算特征点的误差值，标准化权重值和初始值（S406）; 选择具有最小误差值的弱学习机（S408）; 基于弱学习机检查是否准确地确定输入图像（S410）; 并根据检查结果改变重量值（S416）。
-
-
-
-
-
-
-
-
-
Search Results
87
records
(took 0.029 seconds)
