-
公开(公告)号：KR101904192B1
公开(公告)日：2018-10-05
申请号：KR1020170067150
申请日：2017-05-30
Applicant: 한국과학기술원
Abstract: 본발명은공간형증강현실에서모델독립형얼굴랜드마크인식장치에관한것이다. 또한, 본발명에따르면, 컬러영상과깊이영상을포함하는얼굴영상을획득하는영상획득부; 상기영상획득부에서획득된얼굴영상에서얼굴영역을검출하여, 컬러영상에서국부이진패턴을산출하고, 깊이영상에서국부각 패턴을산출하여국부이전패턴과국부각 패턴이결합된얼굴영상의얼굴특징을산출하는얼굴특징검출부; 및상위얼굴랜드마크집합에서하위두 개의집합으로나누면서얼굴위상관계트리인얼굴토폴로지트리를형성하고, 상기얼굴토폴로지트리의루트노드와일치하는얼굴전체랜드마크를포함하는다수의테스트얼굴영상을가지고학습하는과정에서잠재회귀포레스트의각 노드에얼굴토폴로지트리를따라얼굴랜드마크부분집합을포함하는부분랜드마크영상을가지도록학습하고, 상기얼굴특징검출부에서검출한얼굴영상의얼굴특징을입력값으로했을때 학습된포레스트를순회(Traverse)하며디비전노드와리프노드에저장된위치값차이를축적하여부분랜드마크의결과값을산출하는랜드마크검출부를포함하는공간형증강현실에서모델독립형얼굴랜드마크인식장치가제공된다.
-
公开(公告)号：KR1020170135758A
公开(公告)日：2017-12-08
申请号：KR1020170067150
申请日：2017-05-30
Applicant: 한국과학기술원
Abstract: 본발명은공간형증강현실에서모델독립형얼굴랜드마크인식장치에관한것이다. 또한, 본발명에따르면, 컬러영상과깊이영상을포함하는얼굴영상을획득하는영상획득부; 상기영상획득부에서획득된얼굴영상에서얼굴영역을검출하여, 컬러영상에서국부이진패턴을산출하고, 깊이영상에서국부각 패턴을산출하여국부이전패턴과국부각 패턴이결합된얼굴영상의얼굴특징을산출하는얼굴특징검출부; 및상위얼굴랜드마크집합에서하위두 개의집합으로나누면서얼굴위상관계트리인얼굴토폴로지트리를형성하고, 상기얼굴토폴로지트리의루트노드와일치하는얼굴전체랜드마크를포함하는다수의테스트얼굴영상을가지고학습하는과정에서잠재회귀포레스트의각 노드에얼굴토폴로지트리를따라얼굴랜드마크부분집합을포함하는부분랜드마크영상을가지도록학습하고, 상기얼굴특징검출부에서검출한얼굴영상의얼굴특징을입력값으로했을때 학습된포레스트를순회(Traverse)하며디비전노드와리프노드에저장된위치값차이를축적하여부분랜드마크의결과값을산출하는랜드마크검출부를포함하는공간형증강현실에서모델독립형얼굴랜드마크인식장치가제공된다.
Abstract translation: 本发明涉及空间增强现实中的与模型无关的人脸标志识别装置。 此外，根据本发明，彩色图像和深度面部图像获得单元，用于获得图像包括所述图像; 检测从由图像获取单元获取的面部图像的面部区域，它计算彩色图像中的局部二元模式，并且计算在深度图像本地先前图案和站入射图案的站入射图案接合面部图像的面部特征 面部特征检测器，用于计算面部特征; 且在标记子集nanumyeonseo两组的上表面面对地相位关系树，其形成面对拓扑树和与大量测试面部图像的包括整个面部界标面部拓扑树的根节点匹配的学习 学习过程，以便具有包括面部界标的子集沿着面拓扑树的电位返回福雷斯特其中每个节点的地标图像的一部分，并且当作为输入值由面部特征检测单元检测到的面部图像的面部特征 遍历空间型AR所学习的森林（横动）和除法节点和叶模型累积存储在包含标志检测单元，用于计算所述部分的地标结果的节点独立位置面部界标识别设备时的位置值的差 提供。
-
公开(公告)号：KR101785650B1
公开(公告)日：2017-10-17
申请号：KR1020160004289
申请日：2016-01-13
Applicant: 한국과학기술원
Abstract: 사용자에게착용된단일깊이카메라에의해촬영되는손의제 1 시퀀스영상을획득하는단계; 획득한제 1 시퀀스영상내 복수의프레임으로부터제 1 시공간특징벡터(spatio-temporal feature vector)를획득하는단계; 클릭동작의발생여부및 클릭위치에대한정보를알고있는손의제 2 시퀀스영상의프레임으로부터추출된제 2 시공간특징벡터에기초하여, 랜덤포레스트(random forest)를구성하는단계; 및제 1 시공간특징벡터를랜덤포레스트에입력하여, 제 1 시퀀스영상에서손의클릭동작의발생여부및 클릭위치를판단하는단계를포함하는것을특징으로하는, 본발명의일 실시예에따른클릭감지장치에의한클릭감지방법이개시된다.
-
公开(公告)号：KR1020140001168A
公开(公告)日：2014-01-06
申请号：KR1020130074403
申请日：2013-06-27
Applicant: 한국과학기술원
CPC classification number: G06K9/4671 , G06K9/6298 , G06T3/0031 , G06T17/20
Abstract: According to the present invention, a method for extracting and generating feature points and feature descriptor from a RGB-D image comprises the steps of: generating a depth gradient based on a position for each pixel with a depth image function for each pixel of a three-dimensional image including depth information acquired by a camera; calculating adjacent three vertices based on a predetermined vertex of the depth image based on the generated depth gradient; producing a predetermined vector using the calculated three vertices and extracting a normalized surface normal vector by the cross product operation of the produced vector; performing rendering using corresponding depth image vector information composed of the extracted surface normal vector to convert a three-dimensional vector into a two-dimensional image; and applying a scale invariant feature transform (SIFT) algorithm to produce feature descriptor from the two-dimensional image. [Reference numerals] (110) Acquire an image by setting modes; (112) Three-dimensional image; (114) RGB image; (116) Generate a depth gradient based on a pixel position; (118) Convert into a grayscale image; (120) Calculate three adjacent vertexes based on a predetermined vertex of a depth image; (122) Produce a predetermined vector; (124) Extract a normalized surface normal vector; (126) Perform rendering; (128) Apply an SIFT algorithm; (130) Produce feature descriptor; (AA) Start; (BB) No; (CC) Yes; (DD) End
Abstract translation: 根据本发明，一种用于从RGB-D图像提取和生成特征点和特征描述符的方法包括以下步骤：基于每个像素的位置生成具有用于三维像素的每个像素的深度图像功能的深度梯度 包括由相机获取的深度信息的三维图像; 基于生成的深度梯度，基于深度图像的预定顶点计算相邻的三个顶点; 使用所计算的三个顶点产生预定向量，并通过所产生的向量的交叉积运算来提取归一化表面法向量; 使用由所提取的表面法线矢量构成的相应的深度图像矢量信息来执行再现，以将三维矢量转换为二维图像; 并应用尺度不变特征变换（SIFT）算法从二维图像中产生特征描述符。 （附图标记）（110）通过设定模式获取图像; （112）三维图像; （114）RGB图像; （116）基于像素位置生成深度梯度; （118）转换为灰度图像; （120）基于深度图像的预定顶点来计算三个相邻顶点; （122）生成预定矢量; （124）提取归一化表面法向量; （126）执行渲染; （128）应用SIFT算法; （130）生成特征描述符; （AA）开始; （BB）否 （CC）是; （DD）结束
-
-
-
Search Results
14
records
(took 0.021 seconds)
