CN103559791A - Vehicle detection method fusing radar and CCD camera signals - Google Patents

Vehicle detection method fusing radar and CCD camera signals Download PDF

Info

Publication number
CN103559791A
CN103559791A CN 201310530503 CN201310530503A CN103559791A CN 103559791 A CN103559791 A CN 103559791A CN 201310530503 CN201310530503 CN 201310530503 CN 201310530503 A CN201310530503 A CN 201310530503A CN 103559791 A CN103559791 A CN 103559791A
Authority
CN
Grant status
Application
Patent type
Prior art keywords
vehicle
image
coordinate system
radar
camera
Prior art date
Application number
CN 201310530503
Other languages
Chinese (zh)
Other versions
CN103559791B (en )
Inventor
鲍泓
徐成
田仙仙
张璐璐
Original Assignee
北京联合大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date

Links

Abstract

The invention discloses a vehicle detection method fusing radar and CCD camera signals. The method includes the steps of inputting the radar and CCD camera signals, correcting a camera to obtain a projection matrix of a road plane coordinate and a projection matrix of an image coordinate, converting a road plane world coordinate into an image plane coordinate, building positive and negative sample sets suitable for a vehicle HOG describer, carrying out batch feature extraction on the vehicle sample sets to build an HOG sample set, building an SVM classification model of a linear support vector machine to train an SVM, extracting regions of interest of barriers detected by radar in a video image, inputting the regions of interest into an SVM classifier to judge a target category, outputting an identifying result, and measuring the distance of a target which is judged to be a vehicle by means of the radar. Combination detection is carried out by means of the radar and CCD camera signals, depth information of the vehicle is obtained, meanwhile, profile information of the vehicle can be well detected, and reliability and accuracy of vehicle detection and positioning are improved.

Description

一种融合雷达和CCD摄像机信号的车辆检测方法 A fusion of the radar and the CCD camera of the vehicle detection signal

技术领域 FIELD

[0001] 本发明属于计算机视觉领域,涉及一种基于多传感器信息融合技术的智能车前方车辆检测方法,特别涉及一种融合雷达和CCD摄像机信号的车辆检测方法。 [0001] The present invention belongs to the field of computer vision, relates to a method of detecting the forward vehicle intelligent vehicle multi-sensor information fusion technique, particularly to a method for detecting a fusion vehicle radar signals and the CCD camera.

背景技术 Background technique

[0002] 智能驾驶技术中,主流的环境检测传感器为惯导、激光雷达、毫米波雷达、红外和彩色(XD摄像机等。 [0002] intelligent driving technology, mainstream environmental detection sensor inertial navigation, laser radar, millimeter-wave radar, infrared and color (XD cameras.

[0003] 采用雷达可以快速精确地获取智能车前方二维水平面上的车辆距离信息,同时雷达的工作频率高,测量的距离数据精度高,价格相对便宜,能够满足车辆检测的实时性。 [0003] The radar can quickly and accurately obtain the vehicle before the horizontal plane Fanger Wei intelligent vehicle distance information, the radar while the high operating frequency, high-precision distance measurement data, relatively cheap, can meet real-time detection of the vehicle. 但是雷达获取车辆的信息量较少,单独使用雷达只能检测出车辆扫描平面上的深度信息。 However, the vehicle information acquired fewer radar, radar depth information alone only detected on the scanning plane of the vehicle.

[0004] 视频传感器可以提供二维可见光图像,通过对特定物体(车辆)已知的明显特征(如边缘、角点、纹理、位置和形状等特征信息)进行检测判定,在一些特定的条件下,可以有效地检测出车辆。 [0004] a two-dimensional video sensor may provide a visible image, it is determined by detecting a particular object (vehicle) a known significant feature (e.g., edge, corner, texture, shape and position of the feature information), under certain conditions , the vehicle can be effectively detected. 使用视频传感器检测车辆通常包含三个过程:首先从图像中确定出目标物体,然后对物体分类识别,最后是跟踪车辆。 Using a video sensor for detecting a vehicle generally consists of three processes: first determining from an image of the target object, then the object classification, and finally the tracking vehicle. 视频传感器的缺点是无法得到车辆的距离信肩、O Disadvantage video sensor can not obtain the distance to the shoulder of the vehicle, O

[0005] 在实际的智能车辆检测应用中,单独使用某一种检测传感器都难以全面而精确地完成车辆检测和定位,所以有必要利用多种传感器的数据进行融合,实现优势互补以提高车辆检测和定位的可靠性、精确性。 [0005] In the actual application of intelligent vehicle detection, a separate sensor for detecting a comprehensive and difficult to accurately detect and locate the vehicle is completed, it is necessary to use a variety of sensor data fusion, complementary advantages to improve the detection of the vehicle and reliability of positioning accuracy.

发明内容 SUMMARY

[0006] 针对现有技术中存在的单独使用一种检测传感器难以全面而精确地完成车辆检测和定位问题,本发明提出了一种融合雷达和CCD摄像机信号的车辆检测方法。 [0006] The prior art for use alone in the presence detection sensor is difficult to fully and accurately detect the vehicle and complete the positioning problems, the present invention provides a method for detecting a fusion vehicle radar signals and the CCD camera.

[0007] 本发明的多传感器包括单线激光雷达和视频传感器。 Multisensor [0007] The present invention includes a single-line laser radar and video sensors. 单线激光雷达可以快速精确地获取智能车前方二维水平面上车辆的距离信息;视频传感器可以提供车辆的二维可见光图像,根据图像可以检测目标车辆的类型信息。 Single line laser radar can quickly and accurately obtain information from the intelligent vehicle before the vehicle Fanger Wei horizontal plane; a two-dimensional sensor may provide a visible light video image of the vehicle, it may be detected according to the image information of the object type of the vehicle.

[0008] 一种融合雷达和CXD摄像机信号的车辆检测方法,包括以下步骤: [0008] A fusion of the radar and camera signal CXD vehicle detection method comprising the steps of:

[0009] 步骤一,输入来自雷达的路面障碍物信息信号和来自CXD摄像机的路平面图像信号。 [0009] Step a, the input information signal and road obstacle signal path from the planar image from the camera CXD radar.

[0010] 步骤二,校正视频传感器摄像头,得到路平面坐标与图像坐标的投影矩阵,将路平面世界坐标转换成图像平面坐标。 [0010] Step two, the sensor calibration video camera, to obtain plane coordinates and image coordinates passage projection matrix, the road plane world coordinates into image plane coordinates. 将单线激光雷达所在的路平面坐标转换成选定的标定板所在的参考系坐标,得到雷达监测到的路平面的车辆坐标对应的图像坐标。 Converting the coordinate plane where the laser radar-line path to a selected reference coordinate system where the calibration plate, the image coordinates of the vehicle to obtain the coordinates of the radar surveillance plane corresponding to the road.

[0011] 步骤三,建立适合车辆H0G(Histogram of Oriented Gradient,梯度方向直方图)特征描述器的正负样本集。 [0011] Step three, the establishment of appropriate positive and negative sample set vehicle H0G (Histogram of Oriented Gradient, gradient direction histogram) of the feature descriptor.

[0012] HOG特征描述器是应用于计算机视觉和图像处理领域,用于目标检测的特征描述器。 [0012] HOG features described are applicable to the field of computer vision and image processing, feature descriptor for target detection. 利用HOG来计算局部图像梯度的方向信息的统计值。 Statistical value calculating direction information of the local image gradients using HOG.

[0013] 步骤四,采用HOG算法对车辆样本集进行批量的特征提取,从而建立HOG特征样本集。 [0013] Step 4 using the HOG algorithm bulk sample set vehicle feature extraction, so as to establish sets of samples HOG features.

[0014] 步骤五,建立线性支持向量机SVM分类模型,使用特征样本集对SVM进行训练。 [0014] Step 5 linear establishing the SVM classification model using the feature of the SVM training sample set.

[0015] 步骤六,提取雷达检测到的障碍物在视频图像中的区域,并进行HOG特征提取,输入训练得到的SVM分类器中进行目标类型判断。 [0015] Step six, to extract the region the radar obstacle detection in the video image, and feature extraction HOG, the target type determination obtained input training SVM classifier.

[0016] 步骤七,输出SVM分类器的目标物体识别结果。 [0016] Step 7 SVM classifier outputs a recognition result of the target object.

[0017] 步骤八,输出单线激光雷达测出的智能车前方判断为车辆的目标的距离。 Forward [0017] Step 8-line output of a laser radar intelligent vehicle determines the distance to the target vehicle.

[0018] 本发明的有益效果主要表现在:利用由视频传感器获得的车辆的图像信息和由单线激光雷达测得的车辆的距离信息进行联合检测,不仅获取了车辆的深度信息,同时也能较好地检测出车辆的轮廓信息,实现优势互补以提高车辆检测、定位的可靠性和精确性。 [0018] Advantageous effects of the present invention mainly in: using the distance information of the image information obtained by the video sensor of the vehicle and by a single line laser radar measured vehicle performs joint detection, only obtaining the depth information of the vehicle, but also to more well detect the contour information of the vehicle, and complement each other to improve vehicle detection, positioning accuracy and reliability.

附图说明 BRIEF DESCRIPTION

[0019] 图1为融合雷达和CXD摄像机信号的车辆检测方法流程图; [0019] FIG. 1 is a vehicle detection fusion CXD radar and the camera signal is a flowchart;

[0020] 图2为世界坐标系到照相机坐标系的变换示意图; [0020] FIG. 2 is a schematic view of the world coordinate transformation system to the camera coordinate system;

[0021] 图3为照相机坐标系到图像坐标系的变换示意图; [0021] FIG. 3 is a schematic view converted image coordinate system to the camera coordinate system;

[0022] 图4为车辆图像水平竖直梯度示意图。 [0022] FIG. 4 is a schematic vertical gradient in horizontal vehicle image.

具体实施方式 detailed description

[0023] 下面结合附图和实施例对本发明做进一步说明。 Drawings and embodiments of the present invention will be further described [0023] below in conjunction.

[0024] 本发明所述的车辆检测方法由存于电脑内的软件实现,电脑安装在车辆后备箱中。 [0024] The vehicle detection method according to the present invention is implemented in software stored in the computer, the computer mounted in the vehicle trunk. 雷达水平固定在汽车前段车牌位置处,CCD摄像机安装在车内后视镜位置。 Automotive radar level is fixed at a position preceding plate, CCD camera mounted in the vehicle mirror position.

[0025] 图1为融合雷达和CXD摄像机信号的车辆检测方法流程图,包括以下步骤: [0025] FIG. 1 is a vehicle detection fusion CXD radar and camera signal flowchart, comprising the steps of:

[0026] 步骤一,输入雷达和CXD摄像机信号。 [0026] Step a, CXD radar and camera signal input.

[0027] 步骤二,进行摄像头校正(包括照相机的内参和外参),得到路平面坐标与图像坐标的投影矩阵,将路平面世界坐标转换成图像平面坐标。 [0027] Step II for camera calibration (including internal reference and an outer camera parameters), the projection matrix to obtain plane coordinates and the image coordinates of the path, converts the path into the image plane of the world coordinate plane coordinates.

[0028] 世界坐标系中的点到图像坐标中的点的投影过程分两步来实现: [0028] projected, the world coordinate system of the image point coordinates of points achieved in two steps:

[0029] (I)将世界坐标系(或参考坐标系)中点的坐标(XW,YW,Zw)变换到照相机坐标系(Xe, Y。,Z。)。 [0029] (I) the world coordinate system (or reference frame) coordinate of the midpoint (XW, YW, Zw) transformed into the camera coordinate system (Xe, Y., Z.). 变换过程如图2所示,变换公式为: Transformation process shown in Figure 2, the conversion formula is:

Figure CN103559791AD00051

[0031] 式中, [0031] In the formula,

Figure CN103559791AD00052

为由世界坐标系到照相机坐标系的旋转矩阵, By the world coordinate system to the camera coordinate system rotation matrix,

Figure CN103559791AD00053

为由世界 By the world

坐标系到照相机坐标系的平移矩阵。 Coordinate system to the camera coordinate translation matrix system.

[0032] (2)由照相机坐标系变换到图像坐标系,变换过程如图3,变换公式为: [0032] (2) converted from the camera coordinate system to the image coordinate system transformation process in FIG. 3, the transform formula is:

Figure CN103559791AD00061

[0034] 式中,fx、fy代表以像素为单位的水平方向和垂直方向焦距,U0, V0分别表示主点(摄像机主轴与图像平面的交点)的横、纵坐标,s为投影参数(是一个过程参数,计算过程中被抵消)。 [0034] wherein, FX, representative of fy pixels in the horizontal direction and the vertical direction, the focal length, U0, V0, respectively, a front point (intersection camera spindle and image plane) of the horizontal and vertical coordinates, s is a projection parameters (a a process parameter calculation process is canceled).

[0035] 世界坐标系中的点到图像坐标中的点的投影公式为: [0035] Projection Formula world coordinate system points to a point in image coordinates is:

Figure CN103559791AD00062

[0037] 选择棋盘格中定义的坐标系作为参考坐标系,每个视角都建立相应的刚体变换,通过给定照相机内参数,得到求解过程的初始值,使求得的照相机的内参尽量使重投影误差最小。 [0037] Select a coordinate system defined in a grid as a reference coordinate system, each view has established the respective rigid body transformation, given by the camera parameters to obtain an initial value of the solution process, so that the internal reference camera is determined as far as possible the weight The minimum projection error. 在标定出相机的内参数时,最后选择地平面上的一张标定板的图片的棋盘格坐标系作为参考坐标系。 When the camera parameters are calibrated, a calibration image last selected on the ground plane plate checkerboard coordinate system as the reference coordinate system.

[0038] 通过将雷达所在路平面坐标系与选定的标定板的坐标系之间进行变换,得到雷达监测到的路平面的车辆坐标与图像坐标的转换矩阵、车辆高度与图像坐标的转换矩阵,从而可以确定出车辆在图像平面上的位置及所在位置的车辆的高度。 [0038] By transforming the radar coordinate system where the path between the plane of the calibration plate coordinate system selected, the conversion matrix to obtain radar monitored vehicle coordinates transformation matrix image coordinates of the road plane, the image coordinates of the vehicle height , which can determine the height position of the vehicle and the vehicle's location on the image plane.

[0039] 步骤三,建立适合车辆HOG特征描述器的正负样本集。 [0039] Step three, the positive and negative sample set for establishing the vehicle's HOG features described.

[0040] 本实施例采用CaltechGraz提供的车辆样本库,车辆的大小都为64X 64批量提取训练样本的HOG特征,HOG的特征数据来自CCD。 [0040] The present embodiment employs the vehicle CaltechGraz provide sample database, the size of the vehicle are batch extract 64X 64 HOG features training samples characteristic data from the HOG CCD.

[0041] 步骤四,采用HOG算法对车辆样本集进行批量特征提取,从而建立HOG特征样本集。 [0041] Step 4 using the HOG algorithm bulk sample set vehicle feature extraction, so as to establish sets of samples HOG features.

[0042] HOG特征是针对矩形区域中的梯度方向上的强度统计。 [0042] HOG wherein the statistic for intensity gradient direction of the rectangular region. 车辆水平竖直梯度示意图如图4所示。 Horizontal vehicle gradient schematic vertical as shown in FIG.

[0043] 采用车辆模板大小为64*64,将模板样本分为16*16大小的block块,设block的高为H,宽为W,本发明采用H:W=1:1块特征提取方法:每个block块分为4个相同的cell单元,每个cell单元的大小为8*8,每个单元的特征是其内部64个像素的特征向量之和。 [0043] The vehicle size is 64 * 64 template, the template sample into a 16 * 16 block size of the block, the block set high is H, width W, the present invention employs H: W = 1: 1 block feature extraction method : each block is divided into four blocks of identical cell units, each cell unit size is 8 * 8, characterized in that a feature vector of each unit of 64 pixels inside and.

[0044] 用I(x,y)表示图像I在(x,y)处像素点的灰度值,按下式计算矩形区域中的梯度方向的强度统计特征: [0044] represents a statistical characteristic intensity gradient direction of the pixel gray value image I in point (x, y), is calculated as follows using the rectangular region I (x, y):

Figure CN103559791AD00063

[0049] 其中,Gx(x,y)、Gy(x,y)分别表示(x,y)处像素点的水平方向和垂直方向的梯度幅值,G(x,y)为(x,y)处像素点的梯度强度,α (x, y)表示(x,y)处像素点的梯度方向。 [0049] wherein, Gx (x, y), Gy (x, y) denote the gradient magnitude of pixels in the horizontal direction at the point (x, y) and the vertical direction, G (x, y) of (x, y ) at pixel gradient magnitude, α (x, y) denotes the gradient direction of the pixel at the point (x, y). [0050] HOG特征将 [0050] HOG features will

Figure CN103559791AD00071

的梯度方向均匀分为9个bin (区间),第k个方向的梯度幅值大小Ak(x,y)为: Gradient direction uniformly divided into nine bin (interval), the gradient magnitude of the amplitude Ak of ​​the k-th directions (x, y) is:

Figure CN103559791AD00072

[0052] 其中,bink(x, y)表示梯度方向的第k个方向区间。 [0052] wherein, bink (x, y) represents the k-th interval directions gradient direction. 这样,(x, y)处像素点的每个方向上的梯度特征可以用一个9维的向量Ak(x,y)表示。 Thus, the gradient direction of each pixel in the feature point at (x, y) can be expressed by a 9-dimensional vector Ak (x, y).

[0053] 为了消除光照等因素影响,对块内的每个单元进行归一化处理: [0053] In order to eliminate the influence of light and other factors, for each cell within a block normalization:

[0054] [0054]

Figure CN103559791AD00073

[0055] 其中,f(cm,k)表示第m个单元cm中的第k个区间的归一化强度,ε是为了避免分母为零设置的一个较小的数。 [0055] where, f (cm, k) denotes the normalized intensity of the m units of cm in the k-th section, ε is a small number in order to avoid setting of zero denominator.

[0056] 由f (C111, k)的表达式可知,每个单元提取的特征向量为9维,每个块的特征为将4个cell单元中的特征级联得到的36维向量。 [0056] apparent from the expression f (C111, k), each feature vector extraction unit 9 for the dimension characteristic of each block is 36-dimensional vector wherein the cascade of four cell units obtained.

[0057] 步骤五,建立线性SVM分类器,使用步骤四中的特征样本集训练SVM分类器。 [0057] Step V. Establishing linear SVM classifier using the feature of the Step 4 SVM classifier training sample set.

[0058] 步骤六,目标类型判断。 [0058] Step six, the target type determination. 将雷达得到的车辆在路平面中的位置信息通过矩阵变换转换成图像坐标信息,将目标区域图像进行HOG特征提取,采用步骤五训练得到的SVM分类器进行预测,判断目标是否是车辆。 The radar vehicle position information obtained in the road plane by a matrix conversion into image coordinates information, comparing the target area image HOG feature extraction step using an SVM classifier is trained to predict five, it determines whether the target is a vehicle.

[0059] 步骤七,输出SVM分类器的目标物体识别结果。 [0059] Step 7 SVM classifier outputs a recognition result of the target object.

[0060] 步骤八,输出单线激光雷达测出的智能车前方判断为车辆的目标的距离。 Forward [0060] Step 8-line output of a laser radar intelligent vehicle determines the distance to the target vehicle.

Claims (1)

  1. 1.一种融合雷达和CXD摄像机信号的车辆检测方法,其特征在于包括以下步骤:步骤一,输入来自雷达的路面障碍物信息信号和来自CCD摄像机的路平面图像信号;步骤二,进行包括照相机的内参和外参的摄像头校正,得到路平面坐标与图像坐标的投影矩阵,将路平面世界坐标转换成图像平面坐标; 世界坐标系中的点到图像坐标中的点的投影过程分两步来实现: (1)将世界坐标系(或参考坐标系)中点的坐标(XW,YW,Zw)变换到照相机坐标系(Xe, Yc, Z。),变换公式为: rIl rI 2 rI 3 ^-W _ rIl 1Il 23 Zc r3i r32 ώ t3 Zw _ I」[OOO lj[ I _'rU rB I 「A 式中,T1 r22 r为由世界坐标系到照相机坐标系的旋转矩阵,ί2为由世界坐标J !、,」 U_系到照相机坐标系的平移矩阵; (2)由照相机坐标系变换到图像坐标系,变换公式为: -叫[fx O 叫「xc_ fy V0 Yc I」[OOIJ[zc 式中,fx、fy代表以像素为单位的水平 1. A method for detecting a fusion vehicle radar and camera CXD signal, comprising the following steps: First, an input signal from the radar road obstacle information and path planar image signals from the CCD camera; step two, a camera comprising internal reference and external camera calibration parameters, the projection matrix to obtain plane coordinates and the image coordinates of the path, the path converting the world coordinates into image plane coordinates plane; world coordinate system to the projection point of the image point in the process of two steps coordinates achieved: (1) the world coordinate system (or reference frame) coordinate of the midpoint (XW, YW, Zw) transformed into the camera coordinate system (Xe, Yc, Z.), converted formula: rIl rI 2 rI 3 ^ -W _ rIl 1Il 23 Zc r3i r32 ώ t3 Zw _ I "[OOO lj [I _'rU rB I" a formula, T1 r22 r rotation matrix by the world coordinate system to the camera coordinate system, ί2 by the world ! coordinate J ,, "U_ translation matrix system to the camera coordinate system; (2) transformed by the camera coordinate system to the image coordinate system transformation formula as follows: - call [fx O called" xc_ fy V0 Yc I "[OOIJ [ zc where, fx, fy pixels representative of the level of 方向和垂直方向焦距,IVVci分别表示摄像机主轴与图像平面的交点的横、纵坐标,s为投影参数; 世界坐标系中的点到图像坐标中的点的投影公式为: 勹「4 0 «OITrll ?;2 rB ! sy ^ O fy V0 r2l r22 r23 t2 ; — I」L。 And vertical focal lengths, IVVci respectively cross-intersection camera spindle and the image plane, and the ordinate, s projection parameter; world coordinate system point projection formula dot image coordinates is: Bao "4 0« OITrll ?; 2 rB sy ^ O fy V0 r2l r22 r23 t2; -! I "L. 0 I JLr- r32 r33 rj j* 选择棋盘格中定义的坐标系作为参考坐标系,每个视角都建立相应的刚体变换,通过给定照相机内参数,得到求解过程的初始值,使求得的照相机的内参尽量使重投影误差最小;在标定出相机的内参数时,最后选择地平面上的一张标定板的图片的棋盘格坐标系作为参考坐标系; 通过将雷达所在路平面坐标系与选定的标定板的坐标系之间进行变换,得到雷达监测到的路平面的车辆坐标与图像坐标的转换矩阵、车辆高度与图像坐标的转换矩阵,从而可以确定出车辆在图像平面上的位置及所在位置的车辆的高度; 步骤三,建立适合车辆HOG特征描述器的正负样本集; 步骤四,采用HOG算法对车辆样本集进行批量特征提取,从而建立HOG特征样本集; HOG特征是针对矩形区域中的梯度方向上的强度统计; 米用行人模板大小为64*128,将行人模板样 0 I JLr- r32 r33 rj j * Select a coordinate system defined in a grid as a reference coordinate system, each view has established the respective rigid body transformation, given by the camera parameters, the initial value of the solution process to obtain the determined the camera internal reference reprojection minimum error as far as possible; when the camera parameters are calibrated, the final selection of an image of the calibration plate tessellated ground plane on the coordinate system as a reference coordinate system; where by the radar coordinate system and the plane of the road transforming the coordinate system of the calibration between the selected plate, to obtain a conversion matrix of radar surveillance and coordinate transformation matrix vehicle road image plane coordinates, the coordinates of the vehicle height of the image, which can determine the position of the vehicle on the image plane and the height of the vehicle's location; step three, the positive and negative sample set for establishing the vehicle's HOG features described; step 4 using the HOG algorithm bulk sample set vehicle feature extraction, so as to establish sets of samples HOG features; wherein HOG for intensity gradient direction of the rectangular region statistics; m pedestrian with template size of 64 * 128, a template-like pedestrian 分为16*16大小的block块,设block的高为H,宽为W,本发明采用H:W=1:1块特征提取方法:每个block块分为4个相同的cell单元,每个cell单元的大小为8*8,每个单元的特征是其内部64个像素的特征向量之和;用I(x,y)表示图像I在(x,y)处像素点的灰度值,按下式计算矩形区域中的梯度方向的强度统计特征: Is divided into 16 * 16 block size block, the block set height H, width W, the present invention employs H: W = 1: 1 block feature extraction method: Each block is divided into four blocks of identical cell units, each a unit cell size is 8 * 8, characterized in that a feature vector of each unit of 64 pixels inside thereof and; I represents a gray value image at the pixel point (x, y) with I (x, y) , wherein the strength of the statistical gradient direction in the rectangular region is calculated as follows:
    Figure CN103559791AC00031
    其中,Gx(x,y)、Gy(x,y)分别表示(x,y)处像素点的水平方向和垂直方向的梯度幅值,G(x, y)为(X,y)处像素点的梯度强度,α (x, y)表示(X,y)处像素点的梯度方向; HOG特征将 Wherein, Gx (x, y), Gy (x, y) represent the gradient magnitude (x, y) at a pixel in the horizontal direction and the vertical direction, G (x, y) of (X, y) at the pixel gradient strength point, α (x, y) represents the (X, y) at the pixel gradient direction; the HOG features will
    Figure CN103559791AC00032
    的梯度方向均匀分为9个bin (区间),第k个方向的梯度幅值大小Ak (X,y)为: Gradient direction uniformly divided into nine bin (interval), the gradient magnitude of the amplitude Ak of ​​the k-th directions (X, y) is:
    Figure CN103559791AC00033
    其中,bink(x,y)表示梯度方向的第k个方向区间;这样,(x, y)处像素点的每个方向上的梯度特征可以用一个9维的向量Ak(x,y)表示; 为了消除光照等因素影响,对块内的每个单元进行归一化处理: Wherein, bink (x, y) represents the k th gradient direction orientation section; thus, the gradient direction of each pixel in the feature point at (x, y) can be a 9-dimensional vector Ak (x, y) represents ; to eliminate the influence of light and other factors, for each cell within a block normalization:
    Figure CN103559791AC00034
    其中,f(cm,k)表示第m个单元cm中的第k个区间的归一化强度,ε是为了避免分母为零设置的一个较小的数; 由f (cm,k)的表达式可知,每个单元提取的特征向量为9维,每个块的特征为将4个cell单元中的特征级联得到的36维向量; 步骤五,建立线性SVM分类器,使用步骤四中的特征样本集训练SVM分类器; 步骤六,目标类型判断;将雷达得到的车辆在路平面中的位置信息通过矩阵变换转换成图像坐标信息,将感兴趣区域图像进行特征提取,采用步骤五训练得到的SVM分类器进行预测,判断目标是否是车辆; 步骤七,输出SVM分类器的目标物体识别结果; 步骤八,利用单线激光雷达测出智能车前方判断为车辆的目标的距离。 Wherein, f (cm, k) denotes the normalized intensity of the m units of cm in the k-th section, ε is a small number in order to avoid setting the denominator is zero; manufactured by f (cm, k) of Expression equation, for each feature vector extraction unit for the dimension 9, wherein each block is a 36-dimensional vector wherein the cascade of four cell units obtained; step five, the establishment of linear SVM classifier, step 6 and using four characterized in SVM classifier training sample set; step 6 determines the target type; the location information of the vehicle obtained in the radar plane path by converting the coordinate information into image matrix, the interest region image feature extraction using training obtained in step five SVM classifier prediction determines whether the target vehicle; step seven, the output of SVM classifier recognition result of a target object; step 8 using a single line laser radar detected forward vehicle is determined that the smart target vehicle distance.
CN 201310530503 2013-10-31 2013-10-31 A fusion ccd camera and a radar detection signal of the vehicle CN103559791B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 201310530503 CN103559791B (en) 2013-10-31 2013-10-31 A fusion ccd camera and a radar detection signal of the vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 201310530503 CN103559791B (en) 2013-10-31 2013-10-31 A fusion ccd camera and a radar detection signal of the vehicle

Publications (2)

Publication Number Publication Date
CN103559791A true true CN103559791A (en) 2014-02-05
CN103559791B CN103559791B (en) 2015-11-18

Family

ID=50014028

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 201310530503 CN103559791B (en) 2013-10-31 2013-10-31 A fusion ccd camera and a radar detection signal of the vehicle

Country Status (1)

Country Link
CN (1) CN103559791B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104464173A (en) * 2014-12-03 2015-03-25 国网吉林省电力有限公司白城供电公司 Power transmission line external damage protection system based on space image three-dimensional measurement
CN104916132A (en) * 2015-05-14 2015-09-16 扬州大学 Method used for determining traffic flow running track of intersection
CN104954663A (en) * 2014-03-24 2015-09-30 东芝阿尔派·汽车技术有限公司 Image processing apparatus and image processing method
CN105206065A (en) * 2015-10-10 2015-12-30 浙江宇视科技有限公司 Vehicle snapshot method and system
CN105574511A (en) * 2015-12-18 2016-05-11 财团法人车辆研究测试中心 Adaptive object classification device having parallel framework and method
CN105809971A (en) * 2014-12-30 2016-07-27 深圳市朗驰欣创科技有限公司 Vehicle tracking method and device
CN105913488A (en) * 2016-04-15 2016-08-31 长安大学 Three-dimensional-mapping-table-based three-dimensional point cloud rapid reconstruction method
CN106296708A (en) * 2016-08-18 2017-01-04 宁波傲视智绘光电科技有限公司 Vehicle tracking method and vehicle tracking device
CN104608692B (en) * 2013-11-05 2017-04-12 现代摩比斯株式会社 Parking assist system and method

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101581575A (en) * 2009-06-19 2009-11-18 南昌航空大学 Three-dimensional rebuilding method based on laser and camera data fusion
JP5047515B2 (en) * 2006-03-20 2012-10-10 アイシン・エィ・ダブリュ株式会社 Image creating system and road image generating method roads, and road image synthesizer
CN103150547A (en) * 2013-01-21 2013-06-12 信帧电子技术(北京)有限公司 Vehicle tracking method and device

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5047515B2 (en) * 2006-03-20 2012-10-10 アイシン・エィ・ダブリュ株式会社 Image creating system and road image generating method roads, and road image synthesizer
CN101581575A (en) * 2009-06-19 2009-11-18 南昌航空大学 Three-dimensional rebuilding method based on laser and camera data fusion
CN103150547A (en) * 2013-01-21 2013-06-12 信帧电子技术(北京)有限公司 Vehicle tracking method and device

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
丁志鹏: "汽车安全系统中的行人检测方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 7, 15 July 2013 (2013-07-15) *
王飞等: "一种雷达和摄像机的空间标定方法", 《计算机测量与控制》, vol. 20, no. 2, 27 April 2012 (2012-04-27) *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104608692B (en) * 2013-11-05 2017-04-12 现代摩比斯株式会社 Parking assist system and method
CN104954663B (en) * 2014-03-24 2018-03-20 东芝阿尔派·汽车技术有限公司 The image processing apparatus and an image processing method
CN104954663A (en) * 2014-03-24 2015-09-30 东芝阿尔派·汽车技术有限公司 Image processing apparatus and image processing method
CN104464173A (en) * 2014-12-03 2015-03-25 国网吉林省电力有限公司白城供电公司 Power transmission line external damage protection system based on space image three-dimensional measurement
CN105809971A (en) * 2014-12-30 2016-07-27 深圳市朗驰欣创科技有限公司 Vehicle tracking method and device
CN104916132A (en) * 2015-05-14 2015-09-16 扬州大学 Method used for determining traffic flow running track of intersection
CN105206065B (en) * 2015-10-10 2018-04-27 浙江宇视科技有限公司 A vehicle capture method and a capture system of the vehicle
CN105206065A (en) * 2015-10-10 2015-12-30 浙江宇视科技有限公司 Vehicle snapshot method and system
CN105574511A (en) * 2015-12-18 2016-05-11 财团法人车辆研究测试中心 Adaptive object classification device having parallel framework and method
CN105913488A (en) * 2016-04-15 2016-08-31 长安大学 Three-dimensional-mapping-table-based three-dimensional point cloud rapid reconstruction method
CN105913488B (en) * 2016-04-15 2018-08-07 长安大学 Kind of three-dimensional point cloud 3D mapping table based on rapid reconstruction
CN106296708A (en) * 2016-08-18 2017-01-04 宁波傲视智绘光电科技有限公司 Vehicle tracking method and vehicle tracking device

Also Published As

Publication number Publication date Type
CN103559791B (en) 2015-11-18 grant

Similar Documents

Publication Publication Date Title
US20090169052A1 (en) Object Detector
US20080253606A1 (en) Plane Detector and Detecting Method
JP2007129560A (en) Object detector
CN102542843A (en) Early warning method for preventing vehicle collision and device
CN102682292A (en) Method based on monocular vision for detecting and roughly positioning edge of road
CN102096803A (en) Safe state recognition system for people on basis of machine vision
US20100259609A1 (en) Pavement marker recognition device, pavement marker recognition method and pavement marker recognition program
US20070211919A1 (en) Vehicle surroundings monitoring apparatus
US20090167844A1 (en) Mobile peripheral monitor
CN102288121A (en) Offset from one kind of distance measuring and warning method based on monocular vision lane
JP2007334751A (en) Vehicle periphery monitoring device
CN101777263A (en) Traffic vehicle flow detection method based on video
CN101137003A (en) Gray associated analysis based sub-pixel fringe extracting method
CN103413313A (en) Binocular vision navigation system and method based on power robot
CN101488222A (en) Camera self-calibration method based on movement target image and movement information
US7925050B2 (en) Vehicle environment monitoring apparatus
CN1582851A (en) Method for determining trace of human movement
CN103868460A (en) Parallax optimization algorithm-based binocular stereo vision automatic measurement method
CN102389361A (en) Blindman outdoor support system based on computer vision
CN104005325A (en) Pavement crack detecting device and method based on depth and gray level images
CN102248947A (en) Object and vehicle detecting and tracking using a 3-D laser rangefinder
CN102496232A (en) Transmission facility monitoring method and system
CN102270301A (en) The method of detecting a road boundary unstructured conjunction with laser radar svm
CN102122390A (en) Method for detecting human body based on range image
CN103390164A (en) Object detection method based on depth image and implementing device thereof

Legal Events

Date Code Title Description
C06 Publication
C14 Grant of patent or utility model