-
公开(公告)号：KR101371275B1
公开(公告)日：2014-03-26
申请号：KR1020120124276
申请日：2012-11-05
Applicant: 재단법인대구경북과학기술원
Abstract: Disclosed by the present invention is an object tracking method capable of solving a problem in which an object tracking performance is dependent on the performance of an object detector. The multi-object tracking method based on a stereo image according to an embodiment of the present invention includes: a normal object recognizing step for extracting an object candidate group from the left and right images of a current frame of a stereo vision; a specific object recognizing step for extracting a specific object which is an interesting object from the object candidate group; a specific target recognizing step for estimating an interesting area of the current frame by using the interest area which is extracted from a previous frame image; and a multi-object tracking step for tracking a multi-object in the current frame image by setting the recognized object candidate group, the recognized specific object, and the estimated interest area as an observation model and setting the interest area of the previous frame image as a motion model. [Reference numerals] (AA) Previous frame post-state vector; (BB) Previous frame; (CC) Current frame; (S21) Normal object recognition; (S23) Specific object recognition; (S25) Target feature recognition; (S27) Data connection; (S29) Multi-object tracking
Abstract translation: 本发明公开的是一种能够解决物体跟踪性能取决于物体检测器的性能的问题的物体追踪方法。 根据本发明的实施例的基于立体图像的多对象跟踪方法包括：正常对象识别步骤，用于从立体视觉的当前帧的左图像和右图像中提取对象候选组; 特定对象识别步骤，用于从对象候选组中提取作为感兴趣对象的特定对象; 特定目标识别步骤，用于通过使用从先前帧图像提取的兴趣区域来估计当前帧的感兴趣区域; 以及多对象跟踪步骤，用于通过将所识别的对象候选组，所识别的特定对象和所估计的兴趣区域设置为观察模型来跟踪当前帧图像中的多个对象，并设置前一帧图像的兴趣区域 作为运动模型。 （附图标记）（AA）上一帧后置状态向量; （BB）上一帧; （CC）当前帧; （S21）正常对象识别; （S23）具体对象识别; （S25）目标特征识别; （S27）数据连接; （S29）多物体追踪
-
公开(公告)号：KR1020130052405A
公开(公告)日：2013-05-22
申请号：KR1020110117821
申请日：2011-11-11
Applicant: 재단법인대구경북과학기술원
Abstract: PURPOSE: An obstacle detecting and separating method of a stereo vision based image and an apparatus performing the same are provided to easily process about a part having a low parallax value through histogram smoothing, thereby improving long distance detection performance. CONSTITUTION: An obstacle detecting method of a stereo vision based image includes following steps. A parallax map is generated by receiving stereo based left and right images(110). A histogram smoothing process is performed in the parallax map(120). The parallax map in which the histogram smoothing is performed is transformed into a V-parallax map, and road information is extracted from the V-parallax map(130). An obstacle area is detected by performing column detection using extracted road information and the parallax map in which the histogram smoothing is performed. An obstacle detecting apparatus includes a stereo camera unit, an image processing unit, and a monitoring unit. The stereo camera unit consisting of left and right cameras obtains an image from outside. The image processing unit detects an obstacle by performing the obstacle detecting method about the image obtained by the stereo camera unit. The monitoring unit displays a detection result of the image processing unit. The obstacle detecting apparatus is installed in a vehicle, and the car transmits the detection result of the image processing unit to a control device. The stereo camera unit obtains the image by using an active shutter method or by separating left and right images using a polarizing filter. [Reference numerals] (100) Stereo camera unit; (200) Image processing unit; (300) Monitoring unit; (AA) External image
Abstract translation: 目的：提供基于立体视觉的图像的障碍物检测和分离方法及其执行装置的装置，以通过直方图平滑来容易地处理具有低视差值的部分，从而改善长距离检测性能。 构成：基于立体视觉的图像的障碍物检测方法包括以下步骤。 通过接收基于立体声的左和右图像（110）来产生视差图。 在视差图（120）中执行直方图平滑处理。 执行直方图平滑处理的视差图被变换为V视差图，并从V视差图（130）中提取道路信息。 通过使用提取的道路信息和执行直方图平滑化的视差图执行列检测来检测障碍物区域。 障碍物检测装置包括立体摄像机单元，图像处理单元和监视单元。 由左右相机组成的立体相机单元从外部获得图像。 图像处理单元通过对由立体摄像机单元获得的图像执行障碍物检测方法来检测障碍物。 监视单元显示图像处理单元的检测结果。 障碍物检测装置安装在车辆中，轿厢将图像处理单元的检测结果发送到控制装置。 立体摄像机单元通过使用主动快门方法或通过使用偏振滤光器分离左和右图像来获得图像。 （附图标记）（100）立体相机单元; （200）图像处理单元; （300）监控单元; （AA）外部图像
-
公开(公告)号：KR1020130052404A
公开(公告)日：2013-05-22
申请号：KR1020110117820
申请日：2011-11-11
Applicant: 재단법인대구경북과학기술원
CPC classification number: H04N13/161 , H04N13/128 , H04N13/275
Abstract: PURPOSE: A distance estimating device using features based on 3D images and a method thereof are provided to estimate the distance and object recognition by extracting the features of an object. CONSTITUTION: A distance estimating unit estimates the information about a distance from a photographed spot to an object using 3D depth information estimated from an input image(S102,S104). A depth information processing unit allocates the 3D depth information of each scale space layer to the featured points of the scale space layer(S112). A scale space selecting unit searches for a scale space layer matched with the feature points extracted from a 2D object image(S116). The distance estimating unit estimates the distance of the object using the depth information of the searched scale space layer(S118). [Reference numerals] (AA) Input image; (BB) 2D object image; (CC) Object image including 3D depth information; (DD) 2D image; (S102) Estimate 3D depth information; (S104) Estimate the distance of an object from a photographing spot; (S106) Compose a scale space; (S108) Estimate depth information for each scale space layer using the depth information of a reference image; (S110) Extract the feature point of each scale space; (S112) Depth information corresponding to the feature point; (S114) Extract the feature point; (S116) Select a scale space layer matched to the feature point; (S118) Estimate an object distance using the depth information of the selected scale space layer
Abstract translation: 目的：提供使用基于3D图像的特征的距离估计装置及其方法，通过提取对象的特征来估计距离和对象识别。 构成：距离估计单元使用从输入图像估计的3D深度信息来估计关于从拍摄点到对象的距离的信息（S102，S104）。 深度信息处理单元将尺度空间层的3D深度信息分配给比例空间层的特征点（S112）。 比例空间选择单元搜索与从2D对象图像提取的特征点匹配的比例空间层（S116）。 距离估计单元使用所搜索的比例空间层的深度信息来估计对象的距离（S118）。 （附图标记）（AA）输入图像; （BB）2D对象图像; （CC）包括3D深度信息的对象图像; （DD）2D图像; （S102）估计3D深度信息; （S104）估计物体与拍摄点的距离; （S106）构成比例空间; （S108）使用参考图像的深度信息估计每个比例空间层的深度信息; （S110）提取每个刻度空间的特征点; （S112）与特征点对应的深度信息; （S114）提取特征点; （S116）选择与特征点匹配的比例空间层; （S118）使用所选择的比例尺空间层的深度信息估计对象距离
-
公开(公告)号：KR101264282B1
公开(公告)日：2013-05-22
申请号：KR1020100126756
申请日：2010-12-13
Applicant: 재단법인대구경북과학기술원
Abstract: 본발명은도로상차량검출방법에관한것으로, (a) 검출차량에설치된적어도하나의카메라를통해상기차량의전방또는후방을촬영하여영상을입력하는단계; (b) 상기입력된영상을영상처리알고리즘을이용하여도로의소실점(vanishing point)을찾는단계; (c) 상기소실점을기준으로관심영역(ROI)을설정하는단계; 및 (d) 상기관심영역(ROI)을영상처리하여차량을검출하는단계를포함한다. 이와같은본 발명을제공하면, 검출의정확도를높이고, 검출알고리즘이차량을검출하기위해탐색하는범위를관심영역의범위로줄임으로써전체영상을탐색할경우보다처리하는시간을줄이게된다.
-
公开(公告)号：KR101208647B1
公开(公告)日：2012-12-06
申请号：KR1020100106136
申请日：2010-10-28
Applicant: 재단법인대구경북과학기술원
Abstract: 장애물체검출방법은좌 영상및 우영상에대하여스테레오매칭을수행하여장애물체를포함하는시차맵(depth map)을생성하는단계, 시차맵의열에대하여시차값의분포를누적하여장애물체에따른 U-변위맵을생성하는단계, 및 U-변위맵 상에서미리설정된크기에해당하는장애물체를검출하는단계를포함한다.
-
Patent Agency Ranking
公开(公告)号：KR101032660B1
公开(公告)日：2011-05-06
申请号：KR1020090117259
申请日：2009-11-30
Applicant: 재단법인대구경북과학기술원
IPC: G06T7/40
CPC classification number: H04N13/225 , G06K9/00805 , G06T7/32 , H04N13/221
Abstract: PURPOSE: A method for extracting an obstacle is provided to detect only vehicles using a contrast features about the surrounding obstacles. CONSTITUTION: The number of peak and valley and location of the same in a calculated histogram are detected(S220). Images for each peak is generated using a frequency value within the section location of an obstacle object(S230). The width and length location of the obstacle object are detected within each peak image(S240). The location of obstacle is detected from a first peak image and a second peak image(S260). The criteria between centered points of the obstacle is calculated(S270). After the recognition of the obstacles, the merge or the separation of two images are performed(S280).
Abstract translation: 目的：提供一种用于提取障碍物的方法，用于仅使用关于周围障碍物的对比度特征来检测车辆。 构成：检测在计算的直方图中的峰谷数和位置数（S220）。 使用在障碍物物体的部分位置内的频率值来生成每个峰值的图像（S230）。 在每个峰值图像内检测障碍物的宽度和长度位置（S240）。 从第一峰值图像和第二峰值图像检测障碍物的位置（S260）。 计算障碍物中心点的标准（S270）。 在识别到障碍物之后，执行两个图像的合并或分离（S280）。
-
公开(公告)号：KR100991671B1
公开(公告)日：2010-11-04
申请号：KR1020080127212
申请日：2008-12-15
Applicant: 재단법인대구경북과학기술원
Abstract: 본 발명은 영상 정합 방법 및 장치와 그를 이용한 객체 검출 장치에 관한 것이다.
본 발명은 스테레오 영상을 정합하여 초기 시차지도를 생성하고, 초기 시차지도에서 저 신뢰도의 시차값을 제거하며, 초기 시차지도에서 제거된 저 신뢰도의 시차값을 v 시차 영상 및 u 시차 영상을 이용하여 보간하여 보간 시차지도를 생성하고 출력하는 시차 보간기를 포함하는 것을 특징으로 하는 영상 정합 장치를 제공한다.
본 발명에 의하면, 신뢰도 높은 시차지도를 장애물 검출이나 인식에서 이용할 수 있으므로, 스테레오 비전 시스템의 성능을 향상시킬 수 있다.
스테레오 영상, 정합, 시차지도, u 시차 영상, v 시차 영상, 보간-
公开(公告)号：KR1020100056597A
公开(公告)日：2010-05-28
申请号：KR1020080115471
申请日：2008-11-20
Applicant: 재단법인대구경북과학기술원
CPC classification number: G06Q50/30 , G01D9/00 , G06Q40/08 , G08G1/0141
Abstract: PURPOSE: An information collecting apparatus, a method thereof and an information analyzer using the same for collecting information about a road are provided to perform the accurate analysis by collecting information about a surrounding environment. CONSTITUTION: A central center server(130) or an information providing server(120) stores the external environment information. An information collecting device(110) stores and generates accident collection information by sensing the accident of a transportation. The information collecting device generates and stores road information by sensing a surrounding road state. The information collecting device generates and stores the inside environmental information.
Abstract translation: 目的：提供一种信息收集装置，其方法和使用该信息收集装置的信息分析器，用于收集关于道路的信息，以通过收集关于周围环境的信息来执行准确的分析。 构成：中央中心服务器（130）或信息提供服务器（120）存储外部环境信息。 信息收集装置（110）通过感测交通事故来存储并产生事故收集信息。 信息收集装置通过感测周围道路状态来生成并存储道路信息。 信息收集装置生成并存储内部环境信息。
-
公开(公告)号：KR100928906B1
公开(公告)日：2009-11-30
申请号：KR1020070140634
申请日：2007-12-28
Applicant: 재단법인대구경북과학기술원
Abstract: A distance estimating device and method using variance are provided to be used for a safe driving system by reducing an estimated error. A distance estimating method using a stereo vision comprises: a step of selecting a target object and calculating the center of the selected target object using a photographed image(S702,S704); a step of setting a window for the variance estimation of a reference image using the center of the object and brightness distributed intelligence of the image(S706); a step of setting the window of the remaining image not the reference image using the variance mean value of a pixel within the set window(S716); a step of calculating similarity of the pixel within the set window on each image; and a step of estimating the precision variance using the similarity(S718).
Abstract translation: 通过减少估计误差，提供了用于安全驾驶系统的距离估计装置和使用方差的方法。 使用立体视觉的距离估计方法包括：选择目标对象并使用拍摄图像计算选择的目标对象的中心的步骤（S702，S704）; 使用对象的中心和图像的亮度分布式智能为参考图像的方差估计设置窗口（S706）的步骤; 使用所设置的窗口内的像素的方差平均值来设置剩余图像的窗口而不是参考图像的步骤（S716）; 计算每个图像上的设定窗口内的像素的相似性的步骤; 以及使用相似度估计精度差异的步骤（S718）。
-
公开(公告)号：KR100926287B1
公开(公告)日：2009-11-12
申请号：KR1020070140656
申请日：2007-12-28
Applicant: 재단법인대구경북과학기술원
Abstract: 본 발명은 거리 추정 장치 및 거리 추정 방법에 관한 것이다. 본 발명은 스테레오 비전 영상의 시차 정보와 원근 정보를 이용하여 픽셀의 양자화로 인한 오차를 최소화할 수 있는 수단을 제공함으로써, 시차의 오차를 줄이고, 객체의 정확한 위치, 거리, 속도 등을 추정할 수 있는 효과가 있다. 또한, 제안된 추정 방법을 이용하면 추가로 다른 센서를 사용하지 않아도 객체의 위치, 거리 추정의 정밀도와 신뢰도를 높일 수 있는 효과가 있다. 상기와 같은 추정 방법은 다양한 분야에 응용하여 정확한 위치, 거리를 추정할 수 있는 효과가 있다. 예를 들어, 로봇이나 자동차 등에 적용하는 경우 자동차의 추돌 방지, 주행 등의 운전 시스템이나 안전 운행 시스템 등에 적용하여 사용할 수 있다.
스테레오 비전, 거리 추정
-
-
-
-
-
-
-
-
-
Search Results
112
records
(took 0.029 seconds)
