MATLAB车道线检测.zip_matlab_matlab例程_数值算法/人工智能下载-pudn.com

<html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css"> <link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css"> <link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/6259e6ec92dc900e62cc01c2/raw.css"> <script src="https://static.pudn.com/base/js/compatibility.min.js"></script> <script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6259e6ec92dc900e62cc01c2/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">汽车偏离车道线的图像分析与视觉计算方法</div><div class="t m0 x1 h3 y2 ff2 fs1 fc0 sc0 ls0 ws0">Abstract ：Focusing on the problems of lane d etection and r unn ing de flection for au tonomous land vehicles, a reliable approach for</div><div class="t m0 x1 h4 y3 ff3 fs1 fc1 sc0 ls0 ws0">fast visio n comp uting is propos ed in this paper. Two 5×5 template operators of directio nal filters are used to extract the edges of</div><div class="t m0 x1 h4 y4 ff3 fs1 fc1 sc0 ls0 ws0">the image. The image is binarized with an autom atic threshold derived from Otsu algorithm. To red uce the computat ion al work, the</div><div class="t m0 x1 h4 y5 ff3 fs1 fc1 sc0 ls0 ws0">position characters of the lanes in the image are concerned to remove the horizontal cont our points. Meanwhile, the constraint con-</div><div class="t m0 x1 h4 y6 ff3 fs1 fc1 sc0 ls0 ws0">ditions based on the geometr ic cha racteristi cs are also formulated t o eliminate the abun dant points. Af ter that, the lanes are detected</div><div class="t m0 x1 h4 y7 ff3 fs1 fc1 sc0 ls0 ws0">with Hough transform. With a coordinate system of 3D space, the pin hole camera model is adopte d to calculate the devia tion angles</div><div class="t m0 x1 h4 y8 ff3 fs1 fc1 sc0 ls0 ws0">and the vertical dista nces of t he vehicle to the lanes. These parameters can be applied to estimate the lane departure time with the</div><div class="t m0 x1 h4 y9 ff3 fs1 fc1 sc0 ls0 ws0">instant speed of the vehicle, which will be valuable for intelligent control and sec urity early warning of the autonomous land vehicles.</div><div class="t m0 x1 h3 ya ff2 fs1 fc0 sc0 ls0 ws0">Key words：computer vis io n; lane detection; edge de tectio n; binarizatio n; Hough transform</div><div class="t m0 x1 h5 yb ff1 fs1 fc0 sc0 ls0 ws0">摘 要：针对汽车自主驾驶技术的车道检测和跑偏告警问题，提出了一种快速可靠的视觉计算方法。利用方向滤波算子</div><div class="t m0 x1 h5 yc ff4 fs1 fc0 sc0 ls0 ws0">对路面图像进行 5×5 模板运算，得到边缘图像；采用 Otsu 自动阈值算法对图像进行二值化处理，并根据车道在图像中的位</div><div class="t m0 x1 h5 yd ff4 fs1 fc0 sc0 ls0 ws0">置特性对边沿图像细化去点，减少后续处理运算量；在此基础上，根据车道几何特性引入约束条件，去除干扰点，并采用</div><div class="t m0 x1 h5 ye ff3 fs1 fc0 sc0 ls0 ws0">Hough 变换检测出车道线；依据针孔摄像机模型建立空间坐标系，用于计算汽车相对于车道线的偏转角和垂直距离，估计</div><div class="t m0 x1 h5 yf ff4 fs1 fc0 sc0 ls0 ws0">驶离车道的时间，为汽车自主驾驶中的安全预警及智能控制提供信息支撑。</div><div class="t m0 x1 h5 y10 ff1 fs1 fc0 sc0 ls0 ws0">关键词：计算机视觉；车道线检测；边缘检测；二值化；Hough 变换</div><div class="t m0 x1 h6 y11 ff2 fs2 fc0 sc0 ls0 ws0">1 引言</div><div class="t m0 x2 h7 y12 ff5 fs1 fc1 sc0 ls0 ws0">近十年来，汽车自主驾驶技术</div><div class="t m0 x3 h8 y13 ff3 fs3 fc1 sc0 ls0 ws0">[1-3]</div><div class="t m0 x4 h7 y12 ff5 fs1 fc1 sc0 ls0 ws0">得到广泛关注并取得</div><div class="t m0 x1 h7 y14 ff5 fs1 fc1 sc0 ls0 ws0">突破性进展，并将在军事、智能交通等方面有广阔的应用</div><div class="t m0 x1 h7 y15 ff5 fs1 fc1 sc0 ls0 ws0">前景。汽车自主驾驶需采用多传感器、多信息融合方式，</div><div class="t m0 x1 h7 y16 ff5 fs1 fc1 sc0 ls0 ws0">将 GPS 定 位信 息、GIS 道 路 信 息与 即 时 路 面 信 息 终 合 分</div><div class="t m0 x1 h7 y17 ff5 fs1 fc1 sc0 ls0 ws0">析处理，形成人、车、路结合的智能控制系统。汽车自主驾</div><div class="t m0 x1 h7 y18 ff5 fs1 fc1 sc0 ls0 ws0">驶的核心技术包括汽车跑 偏 告 警 、前方 车 辆 行 人 等 障碍</div><div class="t m0 x1 h7 y19 ff5 fs1 fc1 sc0 ls0 ws0">物识别</div><div class="t m0 x5 h8 y1a ff3 fs3 fc1 sc0 ls0 ws0">[4]</div><div class="t m0 x6 h7 y19 ff5 fs1 fc1 sc0 ls0 ws0">、自动泊车、周围环境感知等技术。其中，车道线</div><div class="t m0 x1 h7 y1b ff5 fs1 fc1 sc0 ls0 ws0">检测作为汽车跑偏预测的 关 键 技 术 ，成为 学 者 们 的 研究</div><div class="t m0 x1 h7 y1c ff5 fs1 fc1 sc0 ls0 ws0">重点之一。</div><div class="t m0 x2 h7 y1d ff5 fs1 fc1 sc0 ls0 ws0">车道线检测的核心是图像分析与视觉计算</div><div class="t m0 x7 h8 y1e ff3 fs3 fc1 sc0 ls0 ws0">[5-6]</div><div class="t m0 x8 h7 y1d ff5 fs1 fc1 sc0 ls0 ws0">。图像分</div><div class="t m0 x1 h7 y1f ff5 fs1 fc1 sc0 ls0 ws0">析的目的是获取车道线特征点，并要求满足实时性、鲁棒性、</div><div class="t m0 x2 h7 y20 ff5 fs1 fc1 sc0 ls0 ws0">性。 于车道线识 的图 分 术一 包括 化、</div><div class="t m0 x9 h7 y21 ff5 fs1 fc1 sc0 ls0 ws0">图像滤波、边沿增强、二值化等预处理步骤。考虑到在实际的</div><div class="t m0 x9 h7 y22 ff5 fs1 fc1 sc0 ls0 ws0">汽车行驶环境中，光照变化、车道线不规则、雨水覆盖、阴影干</div><div class="t m0 x9 h7 y23 ff5 fs1 fc1 sc0 ls0 ws0">扰以及相邻车辆的遮挡等影响因素，近五年来，学者们提出很</div><div class="t m0 x9 h7 y24 ff5 fs1 fc1 sc0 ls0 ws0">多解决的方案及算法</div><div class="t m0 xa h8 y25 ff3 fs3 fc1 sc0 ls0 ws0">[7-9]</div><div class="t m0 xb h7 y24 ff5 fs1 fc1 sc0 ls0 ws0">，如尺度不变特征转换（SIFT）</div><div class="t m0 xc h8 y25 ff3 fs3 fc1 sc0 ls0 ws0">[10 ]</div><div class="t m0 xd h7 y24 ff5 fs1 fc1 sc0 ls0 ws0">、边界</div><div class="t m0 x9 h7 y26 ff5 fs1 fc1 sc0 ls0 ws0">点分类</div><div class="t m0 xe h8 y27 ff3 fs3 fc1 sc0 ls0 ws0">[11-12]</div><div class="t m0 xf h7 y26 ff5 fs1 fc1 sc0 ls0 ws0">、反透视变换（IPM）</div><div class="t m0 x10 h9 y27 ff3 fs3 fc1 sc0 ls0 ws0">[1，13]</div><div class="t m0 x11 h7 y26 ff5 fs1 fc1 sc0 ls0 ws0">、颜色空间转换（CSC）</div><div class="t m0 x12 h8 y27 ff3 fs3 fc1 sc0 ls0 ws0">[11]</div><div class="t m0 x13 h7 y26 ff5 fs1 fc1 sc0 ls0 ws0">等。</div><div class="t m0 x14 h7 y28 ff5 fs1 fc1 sc0 ls0 ws0">车道线检测通用方法是在局部区域内将车道线视为</div><div class="t m0 x9 h7 y29 ff5 fs1 fc1 sc0 ls0 ws0">直线。在获得车道特征信息后，采用相关算法获得车道线</div><div class="t m0 x9 h7 y2a ff5 fs1 fc1 sc0 ls0 ws0">的直线方程。其中被广泛采用的有 Hough</div><div class="t m0 x15 h9 y2b ff3 fs3 fc1 sc0 ls0 ws0">[4，8，13]</div><div class="t m0 x16 h7 y2a ff5 fs1 fc1 sc0 ls0 ws0">变换和随机</div><div class="t m0 x9 h7 y2c ff5 fs1 fc1 sc0 ls0 ws0">抽样一致性算法（R ANSAC）</div><div class="t m0 x10 h9 y2d ff3 fs3 fc1 sc0 ls0 ws0">[5，11，14]</div><div class="t m0 x17 h7 y2c ff5 fs1 fc1 sc0 ls0 ws0">。近年来，一些学者也在</div><div class="t m0 x9 h7 y2e ff5 fs1 fc1 sc0 ls0 ws0">尝试新的算法，如模糊算法（fuzzy logic）</div><div class="t m0 x18 h9 y2f ff3 fs3 fc1 sc0 ls0 ws0">[2，15]</div><div class="t m0 x19 h7 y2e ff5 fs1 fc1 sc0 ls0 ws0">、基于几何模</div><div class="t m0 x9 h7 y30 ff5 fs1 fc1 sc0 ls0 ws0">型与 Gabor 滤波器的算法</div><div class="t m0 x1a h8 y31 ff3 fs3 fc1 sc0 ls0 ws0">[16 ]</div><div class="t m0 x1b h7 y30 ff5 fs1 fc1 sc0 ls0 ws0">、约束 De launay 三角化</div><div class="t m0 x1c h8 y31 ff3 fs3 fc1 sc0 ls0 ws0">[3]</div><div class="t m0 x1d h7 y30 ff5 fs1 fc1 sc0 ls0 ws0">等。对</div><div class="t m0 x9 h7 y32 ff5 fs1 fc1 sc0 ls0 ws0">车道线检测性能的判定包括实时性和准确性因素，Hough</div><div class="t m0 x9 h7 y33 ff5 fs1 fc1 sc0 ls0 ws0">在曲线检测方面具有良好性能，但由于严重依赖于参与投</div></div><div class="pi" data-data='{"ctm":[1.525526,0.000000,0.000000,1.525526,0.000000,0.000000]}'></div></div> </body> </html>

<div id="pf2" class="pf w0 h0" data-page-no="2"><div class="pc pc2 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6259e6ec92dc900e62cc01c2/bg2.jpg"><div class="t m0 x1 h7 y34 ff5 fs1 fc1 sc0 ls0 ws0">票的特征点数量，因此具有挑战性的工作是在减少干扰点</div><div class="t m0 x1 h7 y35 ff5 fs1 fc1 sc0 ls0 ws0">数量同时又能保持有效信息。许多学者采用改进的 Hough</div><div class="t m0 x1 h7 y36 ff5 fs1 fc1 sc0 ls0 ws0">变换</div><div class="t m0 x1e h8 y37 ff3 fs3 fc1 sc0 ls0 ws0">[17 -18]</div><div class="t m0 x1f h7 y36 ff5 fs1 fc1 sc0 ls0 ws0">，针对其运算量大，对环境依赖性强的特点，对特</div><div class="t m0 x1 h7 y38 ff5 fs1 fc1 sc0 ls0 ws0">征点进行删减，但同时容易丢失有用信息，影响了车道线</div><div class="t m0 x1 h7 y39 ff5 fs1 fc1 sc0 ls0 ws0">检测的准确性。目前大部分车道检测算法的测试环境都</div><div class="t m0 x1 h7 y3a ff5 fs1 fc1 sc0 ls0 ws0">选在高速公路或较空旷的城市道路，反应了在抗干扰能力</div><div class="t m0 x1 h7 y3b ff5 fs1 fc1 sc0 ls0 ws0">方面存在不足。</div><div class="t m0 x2 h7 y3c ff5 fs1 fc1 sc0 ls0 ws0">文中提出了一种车道线检测及汽车偏离角度和距离</div><div class="t m0 x1 h7 y3d ff5 fs1 fc1 sc0 ls0 ws0">的计算方法。首先采用 5×5 模板得到边沿图像，并对其采</div><div class="t m0 x1 h7 y3e ff5 fs1 fc1 sc0 ls0 ws0">用 Otsu 自动阈值方法作二值变换；然后，根据车道线在图</div><div class="t m0 x1 h7 y3f ff5 fs1 fc1 sc0 ls0 ws0">像中的分布特性做细化处理，并由几何特性制定多项约束</div><div class="t m0 x1 h7 y40 ff5 fs1 fc1 sc0 ls0 ws0">条件，以去除干扰点，适应复杂的行驶环境，采用 Hough 变</div><div class="t m0 x1 h7 y41 ff5 fs1 fc1 sc0 ls0 ws0">换得到车道线方程。以摄像头为原点建立三维坐标系，记</div><div class="t m0 x1 h7 y42 ff5 fs1 fc1 sc0 ls0 ws0">录摄像头的安装高度和俯角，并依据针孔摄像机模型计算</div><div class="t m0 x1 h7 y43 ff5 fs1 fc1 sc0 ls0 ws0">汽车相对车道线的偏转角和垂直距离，获得汽车行驶中的</div><div class="t m0 x1 h7 y44 ff5 fs1 fc1 sc0 ls0 ws0">安全预警及智能控制信息。</div><div class="t m0 x1 h6 y45 ff2 fs2 fc0 sc0 ls0 ws0">2 车道检测算法</div><div class="t m0 x1 h6 y46 ff2 fs2 fc0 sc0 ls0 ws0">2.1 图像的边沿检测</div><div class="t m0 x2 h7 y47 ff5 fs1 fc1 sc0 ls0 ws0">由于车道图像中车道线具有明显的边界存在，边沿检</div><div class="t m0 x1 h7 y48 ff5 fs1 fc1 sc0 ls0 ws0">测成为车道线特征提取的重要手段。典型的图像边沿检</div><div class="t m0 x1 h7 y49 ff5 fs1 fc1 sc0 ls0 ws0">测算法包括 Sob el 算子、Canny 算子</div><div class="t m0 x3 h8 y4a ff3 fs3 fc1 sc0 ls0 ws0">[7]</div><div class="t m0 x20 h7 y49 ff5 fs1 fc1 sc0 ls0 ws0">等，前者是在以像素点</div><div class="t m0 x1 h7 y4b ff5 fs1 fc1 sc0 ls0 ws0">为中心的</div><div class="t m0 x21 ha y4c ff3 fs4 fc0 sc0 ls0 ws0">3 ´ 3</div><div class="t m0 x22 h7 y4b ff5 fs1 fc1 sc0 ls0 ws0">邻域内做水平和垂直的梯度运算，后者采用</div><div class="t m0 x1 h7 y4d ff5 fs1 fc1 sc0 ls0 ws0">二维高斯函数任意方向上的一阶方向导数作为噪声滤波</div><div class="t m0 x1 h7 y4e ff5 fs1 fc1 sc0 ls0 ws0">器，通过与图像卷积进行滤波，然后对滤波后的图像寻找</div><div class="t m0 x1 h7 y4f ff5 fs1 fc1 sc0 ls0 ws0">图像梯度的局部极大值，以此来确定图像边缘。上述边沿</div><div class="t m0 x1 h7 y50 ff5 fs1 fc1 sc0 ls0 ws0">检测算法主要考虑水平和垂直边沿，对斜向边沿检测效果</div><div class="t m0 x1 h7 y51 ff5 fs1 fc1 sc0 ls0 ws0">不理想，不太适合用于车道线检测，文中提出一种能保留</div><div class="t m0 x1 h7 y52 ff5 fs1 fc1 sc0 ls0 ws0">斜向边沿的图像滤波方法。</div><div class="t m0 x2 h7 y53 ff5 fs1 fc1 sc0 ls0 ws0">车道图像是路面、车道线、人车等障碍物、各种交通标</div><div class="t m0 x1 h7 y54 ff5 fs1 fc1 sc0 ls0 ws0">志和路边微环境的组合，容易受到各种噪声干扰。考虑到</div><div class="t m0 x1 h7 y55 ff5 fs1 fc1 sc0 ls0 ws0">噪声普遍具有离散、无规律的分布特性，而在图像边沿处，</div><div class="t m0 x1 h7 y56 ff5 fs1 fc1 sc0 ls0 ws0">临近点之间的梯度具有一致性，为此采用临近点像素梯度</div><div class="t m0 x1 h7 y57 ff5 fs1 fc1 sc0 ls0 ws0">的矢量叠加法提取边沿信息。</div><div class="t m0 x2 h7 y58 ff5 fs1 fc1 sc0 ls0 ws0">设图 1 中</div><div class="t m0 x23 hb y59 ff7 fs4 fc0 sc0 ls0 ws0">x</div><div class="t m0 x24 h7 y58 ff5 fs1 fc1 sc0 ls0 ws0">为待考察像素点，</div><div class="t m0 x25 hc y59 ff7 fs4 fc0 sc0 ls0 ws0">a~h</div><div class="t m0 x26 h7 y58 ff5 fs1 fc1 sc0 ls0 ws0">为</div><div class="t m0 x27 hb y59 ff7 fs4 fc0 sc0 ls0 ws0">x</div><div class="t m0 x28 h7 y58 ff5 fs1 fc1 sc0 ls0 ws0">点的 8 个临近</div><div class="t m0 x1 h7 y5a ff5 fs1 fc1 sc0 ls0 ws0">点，</div><div class="t m0 x29 hb y5b ff7 fs4 fc0 sc0 ls0 ws0">I</div><div class="t m0 x2a hd y5c ff7 fs5 fc0 sc0 ls0 ws0">a</div><div class="t m0 x2b hc y5b ff3 fs4 fc0 sc0 ls0 ws0">~I</div><div class="t m0 x2c hd y5c ff7 fs5 fc0 sc0 ls0 ws0">h</div><div class="t m0 x2d h7 y5a ff5 fs1 fc1 sc0 ls0 ws0">为这些点对应的灰度值，G</div><div class="t m0 x4 hd y5d ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m0 x4 hd y5e ff7 fs5 fc0 sc0 ls0 ws0">H</div><div class="t m0 x2e h7 y5a ff5 fs1 fc1 sc0 ls0 ws0">和 G</div><div class="t m0 x2f hd y5d ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m0 x30 hd y5e ff7 fs5 fc0 sc0 ls0 ws0">V</div><div class="t m0 x31 h7 y5a ff5 fs1 fc1 sc0 ls0 ws0">分别定义为</div><div class="t m0 x32 hb y5b ff7 fs4 fc0 sc0 ls0 ws0">x</div><div class="t m0 x1 h7 y5f ff5 fs1 fc1 sc0 ls0 ws0">点的水平和垂直方向的梯度，该梯度的计算需考虑斜向灰</div><div class="t m0 x1 h7 y60 ff5 fs1 fc1 sc0 ls0 ws0">度差值所产生的分量，见式（1）。</div><div class="t m0 x14 ha y61 ff6 fs4 fc0 sc0 ls0 ws0">ì</div><div class="t m0 x14 ha y62 ff6 fs4 fc0 sc0 ls0 ws0">í</div><div class="t m0 x14 ha y63 ff6 fs4 fc0 sc0 ls0 ws0">î</div><div class="t m0 x14 ha y64 ff6 fs4 fc0 sc0 ls0 ws0">ï</div><div class="t m0 x14 ha y65 ff6 fs4 fc0 sc0 ls0 ws0">ï</div><div class="t m0 x14 ha y66 ff6 fs4 fc0 sc0 ls0 ws0">ï</div><div class="t m0 x14 ha y67 ff6 fs4 fc0 sc0 ls0 ws0">ï</div><div class="t m0 x33 hb y68 ff7 fs4 fc0 sc0 ls0 ws0">G</div><div class="t m0 x34 hd y69 ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m0 x35 hd y6a ff7 fs5 fc0 sc0 ls0 ws0">H</div><div class="t m0 x36 ha y68 ff6 fs4 fc0 sc0 ls0 ws0">= I</div><div class="t m0 x37 hd y6a ff7 fs5 fc0 sc0 ls0 ws0">a</div><div class="t m0 x38 ha y68 ff6 fs4 fc0 sc0 ls0 ws0">- I</div><div class="t m0 x39 hd y6a ff7 fs5 fc0 sc0 ls0 ws0">e</div><div class="t m0 x3a ha y68 ff6 fs4 fc0 sc0 ls0 ws0">+</div><div class="t m0 xa hc y6b ff3 fs4 fc0 sc0 ls0 ws0">2</div><div class="t m0 x3b hc y6c ff3 fs4 fc0 sc0 ls0 ws0">2</div><div class="t m0 x3c hc y68 ff3 fs4 fc0 sc0 ls0 ws0">(I</div><div class="t m0 x3d hd y6a ff7 fs5 fc0 sc0 ls0 ws0">b</div><div class="t m0 x3e ha y68 ff6 fs4 fc0 sc0 ls0 ws0">+ I</div><div class="t m0 x1b hd y6a ff7 fs5 fc0 sc0 ls0 ws0">h</div><div class="t m0 x10 ha y68 ff6 fs4 fc0 sc0 ls0 ws0">- I</div><div class="t m0 x3f hd y6a ff7 fs5 fc0 sc0 ls0 ws0">f</div><div class="t m0 x40 ha y68 ff6 fs4 fc0 sc0 ls0 ws0">- I</div><div class="t m0 x41 hd y6a ff7 fs5 fc0 sc0 ls0 ws0">d</div><div class="t m0 x42 hc y68 ff3 fs4 fc0 sc0 ls0 ws0">)</div><div class="t m0 x33 hb y6d ff7 fs4 fc0 sc0 ls0 ws0">G</div><div class="t m0 x34 hd y6e ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m0 x43 hd y6f ff7 fs5 fc0 sc0 ls0 ws0">V</div><div class="t m0 x44 ha y6d ff6 fs4 fc0 sc0 ls0 ws0">= I</div><div class="t m0 x45 hd y6f ff7 fs5 fc0 sc0 ls0 ws0">c</div><div class="t m0 x46 ha y6d ff6 fs4 fc0 sc0 ls0 ws0">- I</div><div class="t m0 x47 hd y6f ff7 fs5 fc0 sc0 ls0 ws0">g</div><div class="t m0 x48 ha y6d ff6 fs4 fc0 sc0 ls0 ws0">+</div><div class="t m0 x49 hc y70 ff3 fs4 fc0 sc0 ls0 ws0">2</div><div class="t m0 x4a hc y71 ff3 fs4 fc0 sc0 ls0 ws0">2</div><div class="t m0 x4b hc y6d ff3 fs4 fc0 sc0 ls0 ws0">(I</div><div class="t m0 x4c hd y6f ff7 fs5 fc0 sc0 ls0 ws0">b</div><div class="t m0 x4d ha y6d ff6 fs4 fc0 sc0 ls0 ws0">+ I</div><div class="t m0 x1b hd y6f ff7 fs5 fc0 sc0 ls0 ws0">d</div><div class="t m0 x10 ha y6d ff6 fs4 fc0 sc0 ls0 ws0">- I</div><div class="t m0 x3f hd y6f ff7 fs5 fc0 sc0 ls0 ws0">f</div><div class="t m0 x40 ha y6d ff6 fs4 fc0 sc0 ls0 ws0">- I</div><div class="t m0 x41 hd y6f ff7 fs5 fc0 sc0 ls0 ws0">h</div><div class="t m0 x4e hc y6d ff3 fs4 fc0 sc0 ls0 ws0">)</div><div class="t m0 x4f h7 y72 ff5 fs1 fc1 sc0 ls0 ws0">（1）</div><div class="t m0 x14 h7 y73 ff5 fs1 fc1 sc0 ls0 ws0">考虑到边沿的梯度一致性，再将 3×3 邻域点的梯度求</div><div class="t m0 x9 h7 y74 ff5 fs1 fc1 sc0 ls0 ws0">矢量平均，作为中心点的像素矢量：</div><div class="t m0 x50 he y75 ff6 fs6 fc0 sc0 ls0 ws0">ì</div><div class="t m0 x50 he y76 ff6 fs6 fc0 sc0 ls0 ws0">í</div><div class="t m0 x50 he y77 ff6 fs6 fc0 sc0 ls0 ws0">î</div><div class="t m1 x50 hf y78 ff6 fs7 fc0 sc0 ls0 ws0">ï</div><div class="t m1 x50 hf y79 ff6 fs7 fc0 sc0 ls0 ws0">ï</div><div class="t m2 x51 h10 y7a ff7 fs6 fc0 sc0 ls0 ws0">G</div><div class="t m2 x52 h11 y7b ff8 fs6 fc0 sc0 ls0 ws0">ˉ</div><div class="t m2 x53 h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">x</div><div class="t m2 x54 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x55 he y7a ff6 fs6 fc0 sc0 ls0 ws0">= (G</div><div class="t m2 x56 h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">a</div><div class="t m2 x56 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x37 he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x57 h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">b</div><div class="t m2 x57 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x48 he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x49 h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">c</div><div class="t m2 x49 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x4b he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x58 h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">d</div><div class="t m2 x3e h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x59 he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x5a h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">e</div><div class="t m2 x5a h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x5b he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x5c h12 y7e ff7 fs8 fc0 sc0 ls0 ws0">f</div><div class="t m2 x41 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x5d he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x5e h12 y7e ff7 fs8 fc0 sc0 ls0 ws0">g</div><div class="t m2 x5e h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x18 he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x5f h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">h</div><div class="t m2 x19 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x60 he y7a ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x61 h12 y7c ff7 fs8 fc0 sc0 ls0 ws0">x</div><div class="t m2 x62 h12 y7d ff7 fs8 fc0 sc0 ls0 ws0">H</div><div class="t m2 x1d h13 y7a ff3 fs6 fc0 sc0 ls0 ws0">)/9</div><div class="t m2 x51 h10 y7f ff7 fs6 fc0 sc0 ls0 ws0">G</div><div class="t m2 x52 h11 y80 ff8 fs6 fc0 sc0 ls0 ws0">ˉ</div><div class="t m2 x53 h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">x</div><div class="t m2 x63 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x64 he y7f ff6 fs6 fc0 sc0 ls0 ws0">= (G</div><div class="t m2 x65 h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">a</div><div class="t m2 x66 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x67 he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x68 h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">b</div><div class="t m2 x69 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x6a he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x6b h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">c</div><div class="t m2 x6c h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 xa he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x6d h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">d</div><div class="t m2 x3d h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x1a he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x10 h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">e</div><div class="t m2 x6e h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x5a he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x6f h12 y83 ff7 fs8 fc0 sc0 ls0 ws0">f</div><div class="t m2 x70 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x41 he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x71 h12 y83 ff7 fs8 fc0 sc0 ls0 ws0">g</div><div class="t m2 x72 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x73 he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x74 h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">h</div><div class="t m2 x75 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x76 he y7f ff6 fs6 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x77 h12 y81 ff7 fs8 fc0 sc0 ls0 ws0">x</div><div class="t m2 x78 h12 y82 ff7 fs8 fc0 sc0 ls0 ws0">V</div><div class="t m2 x79 h13 y7f ff3 fs6 fc0 sc0 ls0 ws0">)/9</div><div class="t m2 x4f h7 y84 ff5 fs1 fc1 sc0 ls0 ws0">（2）</div><div class="t m2 x14 h7 y85 ff5 fs1 fc1 sc0 ls0 ws0">最后，定义 x 点处的灰度值为 I</div><div class="t m2 x7a h14 y86 ff8 fs4 fc0 sc0 ls0 ws0">′</div><div class="t m2 x5d hd y87 ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m2 x72 ha y85 ff6 fs4 fc0 sc0 ls0 ws0">= G</div><div class="t m2 x15 h14 y88 ff8 fs4 fc0 sc0 ls0 ws0">ˉ</div><div class="t m2 x74 hd y89 ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m2 x75 hd y87 ff7 fs5 fc0 sc0 ls0 ws0">H</div><div class="t m2 x7b ha y85 ff6 fs4 fc0 sc0 ls0 ws0">+ G</div><div class="t m2 x7c h14 y88 ff8 fs4 fc0 sc0 ls0 ws0">ˉ</div><div class="t m2 x7d hd y89 ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m2 x77 hd y87 ff7 fs5 fc0 sc0 ls0 ws0">V</div><div class="t m2 x1c h7 y85 ff5 fs1 fc1 sc0 ls0 ws0">，从而得</div><div class="t m2 x9 h7 y8a ff5 fs1 fc1 sc0 ls0 ws0">到边沿图像。通过分析不难知道该边沿检测方法可等效</div><div class="t m2 x9 h7 y8b ff5 fs1 fc1 sc0 ls0 ws0">为两个 5×5 的模板运算，其中水平和竖直方向边沿检测模</div><div class="t m2 x9 h7 y8c ff5 fs1 fc1 sc0 ls0 ws0">板分别由式（3）和（4）定义。</div><div class="t m2 x53 ha y8d ff6 fs4 fc0 sc0 ls0 ws0">æ</div><div class="t m2 x53 ha y8e ff6 fs4 fc0 sc0 ls0 ws0">è</div><div class="t m2 x53 ha y8f ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y90 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y91 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y92 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y93 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y94 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y95 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y96 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y97 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha y98 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x7e ha y8d ff6 fs4 fc0 sc0 ls0 ws0">ö</div><div class="t m2 x7e ha y8e ff6 fs4 fc0 sc0 ls0 ws0">ø</div><div class="t m2 x7e ha y8f ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y90 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y91 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y92 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y93 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y94 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y95 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y96 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y97 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7e ha y98 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x7f ha y99 ff3 fs4 fc0 sc0 ls0 ws0">2 2 1 + 2 /2 1 + 2 1 + 2 /2 2 2</div><div class="t m2 x7f ha y9a ff3 fs4 fc0 sc0 ls0 ws0">2 2 1 + 2 /2 1 + 2 1 + 2 /2 2 2</div><div class="t m2 x80 hc y9b ff3 fs4 fc0 sc0 ls0 ws0">0 0 0 0 0</div><div class="t m2 x33 ha y9c ff6 fs4 fc0 sc0 ls0 ws0">- 2 /2 -1 - 2 /2 -1 - 2 -1 - 2 /2 - 2 /2</div><div class="t m2 x33 ha y9d ff6 fs4 fc0 sc0 ls0 ws0">- 2 /2 -1 - 2 /2 -1 - 2 -1 - 2 /2 - 2 /2</div><div class="t m2 x4f h7 y9e ff5 fs1 fc1 sc0 ls0 ws0">（3）</div><div class="t m2 x53 ha y9f ff6 fs4 fc0 sc0 ls0 ws0">æ</div><div class="t m2 x53 ha ya0 ff6 fs4 fc0 sc0 ls0 ws0">è</div><div class="t m2 x53 ha ya1 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya2 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya3 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya4 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya5 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya6 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya7 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya8 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha ya9 ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha yaa ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha yab ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 x53 ha yac ff6 fs4 fc0 sc0 ls0 ws0">ç</div><div class="t m2 xc ha y9f ff6 fs4 fc0 sc0 ls0 ws0">ö</div><div class="t m2 xc ha ya0 ff6 fs4 fc0 sc0 ls0 ws0">ø</div><div class="t m2 xc ha ya1 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya2 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya3 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya4 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya5 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya6 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya7 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya8 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha ya9 ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha yaa ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha yab ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 xc ha yac ff6 fs4 fc0 sc0 ls0 ws0">÷</div><div class="t m2 x43 ha yad ff6 fs4 fc0 sc0 ls0 ws0">- 2 /2 - 2 /2 0 2 2 2 2</div><div class="t m2 x33 ha yae ff6 fs4 fc0 sc0 ls0 ws0">-1 - 2 /2 -1 - 2 /2 0 1 + 2 /2 1 + 2 /2</div><div class="t m2 xe ha yaf ff6 fs4 fc0 sc0 ls0 ws0">-1 - 2 -1 - 2 0 1 + 2 1 + 2</div><div class="t m2 x33 ha yb0 ff6 fs4 fc0 sc0 ls0 ws0">-1 - 2 /2 -1 - 2 /2 0 1 + 2 /2 1 + 2 /2</div><div class="t m2 x43 ha yb1 ff6 fs4 fc0 sc0 ls0 ws0">- 2 /2 - 2 /2 0 2 2 2 2</div><div class="t m2 x4f h7 yb2 ff5 fs1 fc1 sc0 ls0 ws0">（4）</div><div class="t m2 x14 h7 yb3 ff5 fs1 fc1 sc0 ls0 ws0">利用上面模板分别对图像进行乘—加运算，得到对应</div><div class="t m2 x9 h7 yb4 ff5 fs1 fc1 sc0 ls0 ws0">于 水 平 和 垂 直 方 向 的 两 幅 梯 度 图 像</div><div class="t m2 x81 hb yb5 ff7 fs4 fc0 sc0 ls0 ws0">I</div><div class="t m2 x82 hd yb6 ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m2 x83 h7 yb4 ff5 fs1 fc1 sc0 ls0 ws0">和</div><div class="t m2 x7b hb yb5 ff7 fs4 fc0 sc0 ls0 ws0">I</div><div class="t m2 x84 hd yb6 ff7 fs5 fc0 sc0 ls0 ws0">y</div><div class="t m2 x60 h7 yb4 ff5 fs1 fc1 sc0 ls0 ws0">，再 由 公 式</div><div class="t m2 x9 ha yb7 ff7 fs4 fc0 sc0 ls0 ws0">I = I</div><div class="t m2 x64 h15 yb8 ff3 fs5 fc0 sc0 ls0 ws0">2</div><div class="t m2 x33 hd yb9 ff7 fs5 fc0 sc0 ls0 ws0">x</div><div class="t m2 x43 ha yb7 ff6 fs4 fc0 sc0 ls0 ws0">+ I</div><div class="t m2 x56 h15 yb8 ff3 fs5 fc0 sc0 ls0 ws0">2</div><div class="t m2 x65 hd yb9 ff7 fs5 fc0 sc0 ls0 ws0">y</div><div class="t m2 x45 h7 yb7 ff5 fs1 fc1 sc0 ls0 ws0">得到边沿图像</div><div class="t m2 x85 hb yba ff7 fs4 fc0 sc0 ls0 ws0">I</div><div class="t m2 x86 h7 yb7 ff5 fs1 fc1 sc0 ls0 ws0">。边沿图像保留了图像的轮廓</div><div class="t m2 x9 h7 ybb ff5 fs1 fc1 sc0 ls0 ws0">信息，尤其是车道线的边沿信息。</div><div class="t m2 x14 h7 ybc ff5 fs1 fc1 sc0 ls0 ws0">在获取边沿图像后，需要对图像进行二值化处理。图</div><div class="t m2 x9 h7 ybd ff5 fs1 fc1 sc0 ls0 ws0">像二值化的关键在于阈值计算，文中采用最大类间方差法</div><div class="t m2 x87 h7 ybe ff5 fs1 fc1 sc0 ls0 ws0">（Otsu 算 法）计 算出自适应阈值，用 于 对 车 道 边 沿 图 像 的</div><div class="t m2 x9 h7 ybf ff5 fs1 fc1 sc0 ls0 ws0">二值分割。图 2 描述了车道图像几种边沿检测算法的对比</div><div class="t m2 x9 h7 yc0 ff5 fs1 fc1 sc0 ls0 ws0">结果。</div><div class="t m2 x68 h16 yc1 ff1 fs9 fc1 sc0 ls0 ws0">图 1 考察点 x 及其临近点定义</div><div class="t m2 x88 h17 yc2 ff7 fsa fc0 sc0 ls0 ws0">a</div><div class="t m2 x89 h17 yc3 ff7 fsa fc0 sc0 ls0 ws0">ji hgf</div><div class="t m2 x8a h17 yc2 ff7 fsa fc0 sc0 ls0 ws0">e</div><div class="t m2 x8a h17 yc4 ff7 fsa fc0 sc0 ls0 ws0">d c b</div><div class="t m2 x8b h18 yc5 ff3 fsa fc0 sc0 ls0 ws0">x</div><div class="t m2 x8c h19 yc6 ff5 fsb fc0 sc0 ls0 ws0">（a）车道图像</div><div class="t m2 x8d h19 yc7 ff5 fsb fc0 sc0 ls0 ws0">（b）Sobel 算子结果 （c）Canny 算子结果 （d）检测结果</div><div class="t m2 x8e h16 yc8 ff1 fs9 fc1 sc0 ls0 ws0">图 2 车道图像的边沿检测效果</div></div><div class="pi" data-data='{"ctm":[1.525526,0.000000,0.000000,1.525526,0.000000,0.000000]}'></div></div>