计算机视觉中双目视觉综述.rar_视觉_双目_图形图像处理

<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/6265918c4c65f41259088ff6/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/6265918c4c65f41259088ff6/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0">１５０</div><div class="t m0 x2 h3 y2 ff1 fs1 fc1 sc0 ls1 ws0">科技资讯</div><div class="t m0 x3 h4 y3 ff1 fs2 fc0 sc0 ls2 ws0">科技资讯<span class="ls3 ws1">　<span class="_ _0"></span>ＳＣＩＥＮＣＥ　<span class="_ _0"> </span>＆　<span class="_ _1"> </span>ＴＥＣＨＮＯＬＯＧＹ　<span class="_ _0"> </span>ＩＮＦＯＲＭＡＴＩＯＮ</span></div><div class="t m0 x4 h5 y4 ff1 fs3 fc0 sc0 ls4 ws2">２００７　　<span class="_ _2"></span>ＮＯ．３４</div><div class="t m0 x4 h5 y5 ff1 fs3 fc0 sc0 ls5 ws3">ＳＣＩＥＮＣＥ　<span class="_ _0"></span>＆　<span class="_ _3"></span>ＴＥＣＨＮＯＬＯＧＹ　<span class="_ _3"></span>ＩＮＦＯＲＭＡＴＩＯＮ</div><div class="t m0 x5 h6 y6 ff1 fs4 fc0 sc0 ls1 ws0">学　术　论　坛</div><div class="t m0 x6 h7 y7 ff1 fs5 fc0 sc0 ls6 ws0">双目立体视觉是计算机视觉的一个重要</div><div class="t m0 x2 h7 y8 ff1 fs5 fc0 sc0 ls1 ws0">分支，<span class="ls7">是研究如何利用二维投影图像恢复三维</span></div><div class="t m0 x2 h7 y9 ff1 fs5 fc0 sc0 ls1 ws0">景物世界，<span class="ls8">即由不同位置的两台或者一台摄像</span></div><div class="t m0 x2 h7 ya ff1 fs5 fc0 sc0 ls1 ws0">机<span class="ls9">（ＣＣＤ）<span class="lsa">经过移动或旋转拍摄同一幅场景</span></span>，<span class="_ _2"></span>通</div><div class="t m0 x2 h7 yb ff1 fs5 fc0 sc0 ls8 ws0">过计算空间点在两幅图像中的视差<span class="ls1">，获得该点</span></div><div class="t m0 x2 h7 yc ff1 fs5 fc0 sc0 lsb ws0">的三维坐标值。</div><div class="t m0 x6 h7 yd ff1 fs5 fc0 sc0 lsc ws0">美国麻省理工学院人工智能实验室的</div><div class="t m0 x2 h7 ye ff1 fs5 fc0 sc0 lsd ws0">Ｍａｒｒ<span class="_ _4"> </span><span class="lse">提出了一种视觉计算理论并应用了双目</span></div><div class="t m0 x2 h7 yf ff1 fs5 fc0 sc0 ls1 ws0">匹配上，<span class="lsf">使两张有视差的平面图产生了深度的</span></div><div class="t m0 x2 h7 y10 ff1 fs5 fc0 sc0 ls10 ws0">立体图形<span class="ls1">，<span class="_ _1"> </span><span class="ls11">奠定了双目立体视觉发展理论基</span></span></div><div class="t m0 x2 h7 y11 ff1 fs5 fc0 sc0 ls12 ws0">础。<span class="ls13">双目视觉直接模拟人类双眼处理景物的</span></div><div class="t m0 x2 h7 y12 ff1 fs5 fc0 sc0 ls1 ws0">方式，可靠简便，<span class="ls14">在许多领域均极具应用价值</span>，</div><div class="t m0 x2 h7 y13 ff1 fs5 fc0 sc0 ls15 ws0">如机器人导航<span class="ls1">，<span class="_ _2"></span><span class="ls16">三维测量以及虚拟现实等。</span></span></div><div class="t m0 x6 h7 y14 ff1 fs5 fc0 sc0 ls17 ws0">双目视觉的实现可以分为以下步骤<span class="ls1">：图像</span></div><div class="t m0 x2 h7 y15 ff1 fs5 fc0 sc0 ls18 ws0">的获取、<span class="ls19">摄像机的标定、<span class="_ _5"></span><span class="ls1a">特征提取、<span class="ls1b">图像</span></span></span></div><div class="t m0 x2 h7 y16 ff1 fs5 fc0 sc0 ls1c ws0">匹配和三维重建。</div><div class="t m0 x2 h8 y17 ff1 fs6 fc0 sc0 ls1d ws0">１　<span class="ls1e">双目视觉的基本原理</span></div><div class="t m0 x2 h7 y18 ff1 fs5 fc0 sc0 ls1f ws0">１．１<span class="_ _1"> </span><span class="ls20">线性模型摄像机<span class="ls1">（<span class="_ _2"></span>图<span class="_ _1"> </span><span class="ls21">１）</span></span></span></div><div class="t m0 x7 h7 y19 ff1 fs5 fc0 sc0 ls1 ws0">图<span class="_ _1"> </span><span class="ls21">１　<span class="_ _2"></span><span class="ls22">线性模型摄像机的数学模型</span></span></div><div class="t m0 x8 h7 y1a ff1 fs5 fc0 sc0 ls1 ws4">　　<span class="_ _2"></span>　　<span class="_ _2"></span>　　　<span class="_ _2"></span>　　<span class="_ _2"></span>　　<span class="_ _2"></span>　　<span class="_ _2"></span>　　　<span class="_ _2"></span>　　<span class="_ _2"></span>　<span class="_ _2"></span><span class="ls12 ws0">　　　（１）</span></div><div class="t m0 x9 h7 y7 ff1 fs5 fc0 sc0 ls1 ws0">其中<span class="ls23">（ｕ，ｖ，１）</span>为<span class="_ _1"> </span>ｐ<span class="_ _1"> </span><span class="ls20">在图像齐次坐标<span class="ls24">，（Ｘ，Ｙ，</span></span></div><div class="t m0 xa h7 y8 ff1 fs5 fc0 sc0 ls25 ws0">Ｚ，１）<span class="ls1">为<span class="_ _1"> </span>Ｐ<span class="_ _6"> </span><span class="ls26">点在世界坐标系下的齐次坐标。</span></span></div><div class="t m0 xa h7 y9 ff1 fs5 fc0 sc0 ls12 ws0">１．２　<span class="ls1">摄象机的定标</span></div><div class="t m0 x9 h7 ya ff1 fs5 fc0 sc0 ls6 ws0">摄象机定标需要一个特制的标定参照物</div><div class="t m0 xa h7 yb ff1 fs5 fc0 sc0 ls1 ws0">如图<span class="_ _1"> </span>２<span class="_ _7"> </span>所示：</div><div class="t m0 x9 h7 yc ff1 fs5 fc0 sc0 ls1 ws0">根据公式<span class="ls15">（１）<span class="ls27">标定出摄象机内外参数。</span></span></div><div class="t m0 xa h7 yd ff1 fs5 fc0 sc0 ls12 ws0">１．３　<span class="ls14">双目视觉的基本原理</span></div><div class="t m0 x9 h7 ye ff1 fs5 fc0 sc0 ls12 ws0">空间任意点<span class="_ _1"> </span><span class="ls1">Ｐ<span class="_ _6"> </span></span>在两个摄像机<span class="_ _1"> </span><span class="ls28">Ｃ１<span class="_ _1"> </span><span class="ls1">和<span class="_ _4"> </span></span>Ｃ２<span class="_ _1"> </span><span class="ls1">上</span></span></div><div class="t m0 xa h7 yf ff1 fs5 fc0 sc0 ls2 ws0">的图象点<span class="_ _1"> </span><span class="ls28">ｐ１<span class="_ _1"> </span><span class="ls1">和<span class="_ _4"> </span></span>ｐ２<span class="_ _1"> </span><span class="ls12">分别从两个图像中检测出</span></span></div><div class="t m0 xa h7 y10 ff1 fs5 fc0 sc0 ls1 ws0">来，<span class="_ _3"></span><span class="ls29">由摄象机矩阵分别为<span class="_ _1"> </span></span>Ｍ</div><div class="t m0 xb h9 y1b ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 xc h7 y10 ff1 fs5 fc0 sc0 ls1 ws0">和<span class="_ _1"> </span>Ｍ</div><div class="t m0 xd h9 y1b ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 xe h7 y10 ff1 fs5 fc0 sc0 ls1 ws0">（<span class="_ _3"></span>图<span class="_ _1"> </span><span class="ls1b">３）</span></div><div class="t m0 xf h7 y1c ff1 fs5 fc0 sc0 ls1 ws0">图<span class="_ _4"> </span>３</div><div class="t m0 x9 h7 y1d ff1 fs5 fc0 sc0 ls1 ws0">其中（ｕ</div><div class="t m0 x10 h9 y1e ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x11 h7 y1d ff1 fs5 fc0 sc0 ls1 ws0">，ｖ</div><div class="t m0 x12 h9 y1e ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x13 h7 y1d ff1 fs5 fc0 sc0 ls15 ws0">，１）<span class="ls1">与（ｕ</span></div><div class="t m0 x14 h9 y1e ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x15 h7 y1d ff1 fs5 fc0 sc0 ls1 ws0">，ｖ</div><div class="t m0 xb h9 y1e ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x16 h7 y1d ff1 fs5 fc0 sc0 ls15 ws0">，１）<span class="ls1">分别为<span class="_ _2"></span>ｐ</span></div><div class="t m0 x17 h9 y1e ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x18 h7 y1d ff1 fs5 fc0 sc0 ls1 ws0">，ｐ</div><div class="t m0 x19 h9 y1e ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1a h7 y1d ff1 fs5 fc0 sc0 ls1 ws0">在</div><div class="t m0 xa h7 y1f ff1 fs5 fc0 sc0 ls15 ws0">图像齐次坐标<span class="ls2a">，（Ｘ，Ｙ，Ｚ，１）<span class="ls1">为<span class="_ _1"> </span>Ｐ<span class="_ _7"> </span><span class="ls14">点在世界坐标</span></span></span></div><div class="t m0 xa h7 y20 ff1 fs5 fc0 sc0 ls12 ws0">系下的齐次坐标<span class="ls1">，<span class="_ _0"></span></span>得到<span class="_ _1"> </span><span class="ls2b">Ｘ，Ｙ，Ｚ<span class="_ _4"> </span></span>的方程为<span class="ls1">：</span></div><div class="t m0 x9 h7 y21 ff1 fs5 fc0 sc0 ls1b ws0">（ｕ</div><div class="t m0 x1b h9 y22 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x1c h7 y21 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x1d h9 y23 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x1e h9 y22 ff1 fs7 fc0 sc0 ls12 ws0">３１</div><div class="t m0 x10 h7 y21 ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x1f h9 y23 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x13 h9 y22 ff1 fs7 fc0 sc0 ls12 ws0">１１</div><div class="t m0 x20 h7 y21 ff1 fs5 fc0 sc0 ls1b ws0">）Ｘ＋（ｕ</div><div class="t m0 x14 h9 y23 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x15 h7 y21 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x21 h9 y23 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x22 h9 y22 ff1 fs7 fc0 sc0 ls12 ws0">３２</div><div class="t m0 xc h7 y21 ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x23 h9 y23 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x24 h9 y22 ff1 fs7 fc0 sc0 ls12 ws0">１２</div><div class="t m0 xe h7 y21 ff1 fs5 fc0 sc0 ls2c ws0">）Ｙ＋（ｕ</div><div class="t m0 x18 h9 y22 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x25 h7 y21 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x26 h9 y23 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x27 h9 y22 ff1 fs7 fc0 sc0 ls12 ws0">３３</div><div class="t m0 x28 h7 y21 ff1 fs5 fc0 sc0 ls1 ws0">－</div><div class="t m0 xa h7 y24 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x29 h9 y25 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x2a h9 y26 ff1 fs7 fc0 sc0 ls12 ws0">１３</div><div class="t m0 x2b h7 y24 ff1 fs5 fc0 sc0 ls2d ws0">）Ｚ＝ｍ</div><div class="t m0 x1d h9 y25 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x1e h9 y26 ff1 fs7 fc0 sc0 ls21 ws0">１４</div><div class="t m0 x10 h7 y24 ff1 fs5 fc0 sc0 ls28 ws0">－ｕ</div><div class="t m0 x1f h9 y26 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x13 h7 y24 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x20 h9 y25 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x2c h9 y26 ff1 fs7 fc0 sc0 ls21 ws0">３４</div><div class="t m0 x9 h7 y27 ff1 fs5 fc0 sc0 ls1b ws0">（ｖ</div><div class="t m0 x1b h9 y28 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x1c h7 y27 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x1d h9 y29 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x1e h9 y28 ff1 fs7 fc0 sc0 ls12 ws0">３１</div><div class="t m0 x10 h7 y27 ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x1f h9 y29 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x13 h9 y28 ff1 fs7 fc0 sc0 ls12 ws0">２１</div><div class="t m0 x20 h7 y27 ff1 fs5 fc0 sc0 ls1b ws0">）Ｘ＋（ｖ</div><div class="t m0 x14 h9 y28 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x15 h7 y27 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x21 h9 y29 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x22 h9 y28 ff1 fs7 fc0 sc0 ls12 ws0">３２</div><div class="t m0 xc h7 y27 ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x23 h9 y29 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x24 h9 y28 ff1 fs7 fc0 sc0 ls12 ws0">２２</div><div class="t m0 xe h7 y27 ff1 fs5 fc0 sc0 ls2c ws0">）Ｙ＋（ｖ</div><div class="t m0 x18 h9 y28 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x25 h7 y27 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x26 h9 y29 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x27 h9 y28 ff1 fs7 fc0 sc0 ls12 ws0">３３</div><div class="t m0 x28 h7 y27 ff1 fs5 fc0 sc0 ls1 ws0">－</div><div class="t m0 xa h7 y2a ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x29 h9 y2b ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x2a h9 y2c ff1 fs7 fc0 sc0 ls12 ws0">２３</div><div class="t m0 x2b h7 y2a ff1 fs5 fc0 sc0 ls2d ws0">）Ｚ＝ｍ</div><div class="t m0 x1d h9 y2b ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x1e h9 y2c ff1 fs7 fc0 sc0 ls21 ws0">２４</div><div class="t m0 x10 h7 y2a ff1 fs5 fc0 sc0 ls28 ws0">－ｖ</div><div class="t m0 x1f h9 y2c ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x13 h7 y2a ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x20 h9 y2b ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x2c h9 y2c ff1 fs7 fc0 sc0 ls21 ws0">３４</div><div class="t m0 x9 h7 y2d ff1 fs5 fc0 sc0 ls1b ws0">（ｕ</div><div class="t m0 x1b h9 y2e ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1c h7 y2d ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x1d h9 y2f ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1e h9 y2e ff1 fs7 fc0 sc0 ls12 ws0">３１</div><div class="t m0 x10 h7 y2d ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x1f h9 y2f ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x13 h9 y2e ff1 fs7 fc0 sc0 ls12 ws0">１１</div><div class="t m0 x20 h7 y2d ff1 fs5 fc0 sc0 ls1b ws0">）Ｘ＋（ｕ</div><div class="t m0 x14 h9 y2e ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x15 h7 y2d ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x21 h9 y2f ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x22 h9 y2e ff1 fs7 fc0 sc0 ls12 ws0">３２</div><div class="t m0 xc h7 y2d ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x23 h9 y2f ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x24 h9 y2e ff1 fs7 fc0 sc0 ls12 ws0">１２</div><div class="t m0 xe h7 y2d ff1 fs5 fc0 sc0 ls2c ws0">）Ｙ＋（ｕ</div><div class="t m0 x18 h9 y2f ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x25 h7 y2d ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x26 h9 y2f ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x27 h9 y2e ff1 fs7 fc0 sc0 ls12 ws0">３３</div><div class="t m0 x28 h7 y2d ff1 fs5 fc0 sc0 ls1 ws0">－</div><div class="t m0 xa h7 y30 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x29 h9 y31 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x2a h9 y32 ff1 fs7 fc0 sc0 ls12 ws0">１３</div><div class="t m0 x2b h7 y30 ff1 fs5 fc0 sc0 ls2d ws0">）Ｚ＝ｍ</div><div class="t m0 x1d h9 y31 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1e h9 y32 ff1 fs7 fc0 sc0 ls21 ws0">１４</div><div class="t m0 x10 h7 y30 ff1 fs5 fc0 sc0 ls28 ws0">－ｕ</div><div class="t m0 x1f h9 y31 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x13 h7 y30 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x20 h9 y31 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x2c h9 y32 ff1 fs7 fc0 sc0 ls21 ws0">３４</div><div class="t m0 x9 h7 y33 ff1 fs5 fc0 sc0 ls1b ws0">（ｖ</div><div class="t m0 x1b h9 y34 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1c h7 y33 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x1d h9 y35 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1e h9 y34 ff1 fs7 fc0 sc0 ls12 ws0">３１</div><div class="t m0 x10 h7 y33 ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x1f h9 y35 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x13 h9 y34 ff1 fs7 fc0 sc0 ls12 ws0">２１</div><div class="t m0 x20 h7 y33 ff1 fs5 fc0 sc0 ls1b ws0">）Ｘ＋（ｖ</div><div class="t m0 x14 h9 y34 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x15 h7 y33 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x21 h9 y35 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x22 h9 y34 ff1 fs7 fc0 sc0 ls12 ws0">３２</div><div class="t m0 xc h7 y33 ff1 fs5 fc0 sc0 ls1b ws0">－ｍ</div><div class="t m0 x23 h9 y35 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x24 h9 y34 ff1 fs7 fc0 sc0 ls12 ws0">２２</div><div class="t m0 xe h7 y33 ff1 fs5 fc0 sc0 ls2c ws0">）Ｙ＋（ｖ</div><div class="t m0 x18 h9 y34 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x25 h7 y33 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x26 h9 y35 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x27 h9 y34 ff1 fs7 fc0 sc0 ls12 ws0">３３</div><div class="t m0 x28 h7 y33 ff1 fs5 fc0 sc0 ls1 ws0">－</div><div class="t m0 xa h7 y36 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x29 h9 y37 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x2a h9 y38 ff1 fs7 fc0 sc0 ls12 ws0">２３</div><div class="t m0 x2b h7 y36 ff1 fs5 fc0 sc0 ls2d ws0">）Ｚ＝ｍ</div><div class="t m0 x1d h9 y37 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x1e h9 y38 ff1 fs7 fc0 sc0 ls21 ws0">２４</div><div class="t m0 x10 h7 y36 ff1 fs5 fc0 sc0 ls28 ws0">－ｖ</div><div class="t m0 x1f h9 y38 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x13 h7 y36 ff1 fs5 fc0 sc0 ls1 ws0">ｍ</div><div class="t m0 x20 h9 y37 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x2c h9 y38 ff1 fs7 fc0 sc0 ls21 ws0">３４</div><div class="t m0 x2d h8 y39 ff1 fs6 fc0 sc0 ls1 ws0">２　<span class="_ _5"></span>特征提取</div><div class="t m0 x2e h7 y3a ff1 fs5 fc0 sc0 ls2e ws0">特征提取就是将图像中的角点、<span class="ls12">轮廓线</span></div><div class="t m0 x2d h7 y3b ff1 fs5 fc0 sc0 ls2f ws0">等特征从图像中提取出来<span class="ls1">，<span class="_ _1"> </span><span class="ls30">以便于图像的匹</span></span></div><div class="t m0 x2d h7 y3c ff1 fs5 fc0 sc0 ls1 ws0">配，<span class="_ _3"></span><span class="ls24">具体实现如下</span>：</div><div class="t m0 x2d h7 y3d ff1 fs5 fc0 sc0 ls31 ws5">２．１　<span class="_ _8"></span>Ｈａｒｒｉｓ<span class="ls32 ws0">角点探测</span></div><div class="t m0 x2e h7 y3e ff1 fs5 fc0 sc0 ls14 ws0">Ｈａｒｒｉｓ<span class="_ _4"> </span><span class="ls33">角点探测其处理过程表示如下<span class="ls1">：</span></span></div><div class="t m0 x2e h7 y3f ff1 fs5 fc0 sc0 ls1 ws0">其中<span class="ls1b">，ｇ</span></div><div class="t m0 x2f h9 y40 ff2 fs7 fc0 sc0 ls1 ws0">ｘ</div><div class="t m0 x30 h7 y3f ff1 fs5 fc0 sc0 ls1 ws0">为<span class="_ _1"> </span>ｘ<span class="_ _7"> </span><span class="ls34">方向的梯度<span class="ls1b">，ｇ</span></span></div><div class="t m0 x31 h9 y40 ff2 fs7 fc0 sc0 ls1 ws0">ｙ</div><div class="t m0 x32 h7 y3f ff1 fs5 fc0 sc0 ls1 ws0">为<span class="_ _1"> </span>ｙ<span class="_ _7"> </span><span class="ls15">方向的</span></div><div class="t m0 x2d h7 y41 ff1 fs5 fc0 sc0 ls1d ws0">梯度<span class="ls1">，<span class="_ _9"> </span><span class="ls35">为高斯模板<span class="ls32">，ｄｅｔ<span class="ls36">为矩阵的行列式</span></span></span>，ｔｒ</span></div><div class="t m0 x2d h7 y42 ff1 fs5 fc0 sc0 ls1 ws0">为矩阵直迹，ｋ<span class="_ _0"></span><span class="ls37">为默认常数。<span class="_ _a"></span><span class="ls1">矩阵<span class="_ _3"></span>Ｉ<span class="_ _3"></span><span class="ls17">中每一点的</span></span></span></div><div class="t m0 x2d h7 y43 ff1 fs5 fc0 sc0 ls38 ws0">元素值对应于原图相应点的兴趣值。</div><div class="t m0 x2d h7 y44 ff1 fs5 fc0 sc0 ls12 ws0">２．２　<span class="ls1">边缘提取算子</span></div><div class="t m0 x2e h7 y45 ff1 fs5 fc0 sc0 ls39 ws0">边缘提取算子常有<span class="_ _1"> </span><span class="ls3a">Ｒｏｂｅｒｔ<span class="_ _1"> </span><span class="ls2a">边缘算子、</span></span></div><div class="t m0 x2d h7 y46 ff1 fs5 fc0 sc0 ls3b ws0">Ｓｏｂｅｌ<span class="_ _1"> </span><span class="ls3c">边缘算子、<span class="ls3d">Ｃａｎｎｙ<span class="_ _1"> </span><span class="ls2a">边缘算子、<span class="ls12">以及</span></span></span></span></div><div class="t m0 x2d h7 y1c ff1 fs5 fc0 sc0 ls3e ws0">Ｌａｐｌａｃｉａｎ<span class="_ _1"> </span><span class="ls37">边缘算子等。</span></div><div class="t m0 x2e h7 y47 ff1 fs5 fc0 sc0 ls3f ws0">（１）Ｒｏｂｅｒｔ<span class="_ _3"></span><span class="ls1">边缘算法</span></div><div class="t m0 x2e h7 y48 ff1 fs5 fc0 sc0 ls40 ws0">该算子通常有下列计算公式表示<span class="ls1">：</span></div><div class="t m0 x2e h7 y49 ff1 fs5 fc0 sc0 ls1 ws0">（２）拉普拉斯边缘检测算法</div><div class="t m0 x2e h7 y4a ff1 fs5 fc0 sc0 ls6 ws0">拉普拉斯边缘检测算法是一种二阶的边</div><div class="t m0 x2d h7 y4b ff1 fs5 fc0 sc0 ls41 ws0">缘检测算法<span class="ls1">，<span class="_ _5"></span><span class="ls42">设一连续函数<span class="ls43">ｆ（ｘ，ｙ）<span class="ls1d">它在<span class="ls35">（ｘ，ｙ）<span class="_ _2"></span><span class="ls1">处</span></span></span></span></span></span></div><div class="t m0 x2d h7 y4c ff1 fs5 fc0 sc0 ls44 ws0">的拉普拉斯值定义如下<span class="ls1">：</span></div><div class="t m0 x2e h7 y4d ff1 fs5 fc0 sc0 ls29 ws0">应用差分方程表示为<span class="ls1">：</span></div><div class="t m0 x2d h8 y4e ff1 fs6 fc0 sc0 ls45 ws6">３　<span class="_ _5"></span>　<span class="_ _5"></span>Ｃａｎｎｙ<span class="ls1 ws0">边缘检测算法</span></div><div class="t m0 x2e h7 y4f ff1 fs5 fc0 sc0 ls37 ws0">设原始信号为<span class="ls46">ｆ（ｘ，ｙ），<span class="ls47">平滑滤波脉冲响应</span></span></div><div class="t m0 x2d h7 y50 ff1 fs5 fc0 sc0 ls48 ws0">ｈ（ｘ，ｙ），<span class="ls38">则平滑后的图像为<span class="ls1">：</span></span></div><div class="t m0 x2e h7 y51 ff1 fs5 fc0 sc0 ls12 ws0">ｇ（ｘ，ｙ）＝ｆ（ｘ，ｙ）　<span class="_"> </span><span class="ls49">ｈ（ｘ，ｙ）</span></div><div class="t m0 x2e h7 y52 ff1 fs5 fc0 sc0 ls4a ws0">然后再记得算平滑后的图像的梯度</div><div class="t m0 x2e h7 y53 ff1 fs5 fc0 sc0 ls4b ws0">Ｃａｎｎｙ<span class="_ _1"> </span><span class="ls12">边缘检测的过程可以直接采用原</span></div><div class="t m0 x2d h7 y54 ff1 fs5 fc0 sc0 ls4c ws0">始图像与平滑脉冲响应一阶微分的卷积来实</div><div class="t m0 x2d h7 y55 ff1 fs5 fc0 sc0 ls1d ws0">现。</div><div class="t m0 x2e h7 y56 ff1 fs5 fc0 sc0 ls4d ws0">其检测结果如图<span class="_ _1"> </span><span class="ls1b">４：</span></div><div class="t m0 x33 ha y57 ff1 fs8 fc0 sc0 ls4e ws0">计算机视觉中双目视觉综述</div><div class="t m0 x1f h4 y58 ff1 fs2 fc0 sc0 ls15 ws0">沈洪宇</div><div class="t m0 x34 h9 y59 ff1 fs7 fc0 sc0 ls1 ws0">１</div><div class="t m0 x35 h4 y58 ff1 fs2 fc0 sc0 ls1 ws7">　<span class="_ _0"></span>　<span class="_ _1"> </span><span class="ws0">柴毅</span></div><div class="t m0 x36 h9 y59 ff1 fs7 fc0 sc0 ls1 ws0">２</div><div class="t m0 x37 h4 y5a ff1 fs2 fc0 sc0 ls21 ws8">（１　<span class="_ _2"></span><span class="ls14 ws0">重庆大学自动化学院<span class="ls1 ws9">　<span class="_ _3"></span>　<span class="_ _0"></span><span class="ws0">重庆</span></span></span><span class="wsa">　<span class="_ _2"></span>　<span class="_ _2"></span>４０００４４；　<span class="_ _2"></span>２　<span class="_ _1"> </span><span class="ls0 ws0">重庆三峡学院<span class="_ _5"></span><span class="ls1 ws7">　<span class="_ _0"></span>　<span class="_ _0"></span><span class="ls0 ws0">重庆万州<span class="_ _5"></span><span class="ls4f wsb">　<span class="_ _3"></span>　<span class="_ _2"></span>４０４０００）</span></span></span></span></span></div><div class="t m0 x6 h4 y5b ff1 fs2 fc0 sc0 ls1 ws0">摘<span class="ws8">　<span class="_ _6"> </span></span>要<span class="ls0">：<span class="_ _3"></span><span class="ls50">双目视觉是利用已经标定了的不同位置的两台或者一台摄像机<span class="ls51">（ＣＣＤ）<span class="ls52">经过移动或旋转拍摄同一幅场景得到两幅二维图像<span class="_ _5"></span><span class="ls0">，<span class="_ _3"></span><span class="ls15">应用图</span></span></span></span></span></span></div><div class="t m0 x6 h4 y5c ff1 fs2 fc0 sc0 lsd ws0">像处理、<span class="ls2a">图像匹配、<span class="ls1f">三维重建等方法处理<span class="_ _5"></span><span class="ls0">，<span class="_ _4"> </span><span class="ls1f">实现从二维图像恢复场景的三维图像的一种模拟人的双目视觉的方法</span>，<span class="_ _1"> </span><span class="ls39">本文总结了近来三</span></span></span></span></div><div class="t m0 x6 h4 y5d ff1 fs2 fc0 sc0 ls53 ws0">维重构建模的研究成果和计算方法<span class="ls0">，<span class="_ _1"> </span><span class="lsd">并提出了一些观点<span class="ls54">，　<span class="_ _4"> </span><span class="ls55">对三维重构建模的难点和发展趋势作了说明。</span></span></span></span></div><div class="t m0 x6 h4 y5e ff1 fs2 fc0 sc0 ls15 ws0">关键词<span class="ls0">：<span class="_ _2"></span><span class="ls56">计算机视觉<span class="ls1 ws7">　<span class="_ _3"></span>　<span class="_ _1"> </span></span><span class="lsf">三维重构<span class="ls1 ws7">　<span class="_ _0"></span>　<span class="_ _1"> </span></span>相机标定<span class="ls1 ws7">　<span class="_ _0"></span>　<span class="_ _1"> </span></span>立体匹配</span></span></span></div><div class="t m0 x6 h4 y5f ff1 fs2 fc0 sc0 ls12 ws0">中图分类号<span class="ls57">：ＴＰ３１１<span class="_ _b"> </span><span class="ls21">文献标识码<span class="ls58">：Ａ<span class="_ _c"> </span><span class="ls1">文章编号<span class="ls59">：１６７２－３７９１（２００７）１２（ａ）－０１５０－０２</span></span></span></span></span></div><div class="t m0 x38 h7 y60 ff1 fs5 fc0 sc0 ls1 ws0">图<span class="_ _4"> </span>２</div></div><div class="pi" data-data='{"ctm":[1.613445,0.000000,0.000000,1.613445,0.000000,0.000000]}'></div></div> </body> </html>