关于图像处理中特征点描述算子的一点总结

合集下载
  1. 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
  2. 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
  3. 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。

关于图像处理中特征点描述算子的一点总结
1.SIFt算子SIFT算子是David
G Lowe在2004提出的,即尺度不变特征变换(Scale Invariant Feature Transform)。

它是以尺度空间的构造为基础的局部特征描述算子,对于图像的缩放、旋转和放射变换等具有不变性。

SIFT算子在构建好的尺度空间的基础上搜索尺度空间中的极值点(特征点),然后确定极值点的尺度信息以及位置,再确定极值点的方向(其邻域梯度的主方向),最终可以得到具有鲁棒性的128维(4*4*8)的特征描述子。

2.surf特征
SURF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">(Speed-Up Robust Features<span
style="mso-bidi-font-size:12.0pt;line-height:125%;
font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">)算子<span
style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">选取图像在尺度空间上的极值点作为候选特征点。

与SIFT<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">算子不同的是SURF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">算子采用Hessian<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">矩阵行列式近似值来构造金字塔。

提取SURF<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征点需要4<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">个步骤:提取SURF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征,对于特征点进行定位,赋予主方向,生成特征点描述符。

3. BRIEF特征
BRIEF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征(binary robust independent
elementary features)<span
style="mso-bidi-font-size:12.0pt;line-height:
125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">是Calonder<span
style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">等<span style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">在2010<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">年提出来的,他采用二进制字符串作为特征点描述符,因而在速度和性能上都有着卓越的表现。

其主要思路是:在特征点附近随机的选取若干点对,将这些点对的灰度值大小组合成一个长为256<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">的二进制字串,并将这个二进制字串作为该特征点的特征描述子。

由于其描述子利用二进制(“0”<span style="mso-bidi-font-size: 12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">和“1”<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">)编码,因此在特征匹配时只需计算2<span style="mso-bidi-font-size: 12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">个特征点描述子的Hamming<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">距离。

大量实验表明,不匹配特征点的描述子的
Hamming<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">距离在128<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">左右,匹配点对描述子的Hamming<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami
ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">距离则远小于128<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">。

由于BRIEF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">的匹配速度远高于SURF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">和
SIFT<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">,因此应用较为广泛。

BRIEF<span
style="mso-bidi-font-size:12.0pt;line-height:125%;
font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">特征描述符是一个bit<span style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">串,有若干个二值
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">测试组成,即影像块
(<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">尺寸为
)<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">经过平滑后再进行若干
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">测试。

<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">其中,影像块
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">在x<span
style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">处的灰度值定义为
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">,特征描述符为一个长度为
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">的二进制串:
<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;
mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:"Times New Roman";
mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:
FR">这里有两点要注意:1<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:
宋体;mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:"Times New Roman";
mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language: FR">)如何选取内核对图形进行平滑处理(预处理);2<span style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">)如何选择测试点对
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">。

这里有多种描述形式进行描述测试点对的分布类型,其中效果比较明显的是与选取的影像块
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">成高斯分布,再生成72<span style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">维的特征点描述符。

在做平滑处理时有中值滤波、均值滤波和最小方差滤波等可以使用,一般高斯滤波比较常用。

<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;
mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:"Times New Roman";
mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language: FR">为了保证描述符的旋转不变性,需要先对
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">点测试坐标进行旋转,使其旋转到特征点的方向上,接下来再利用BRIEF<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征描述符
进行计算:
<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;
mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:"Times New Roman";
mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language: FR">其中以
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">角形成的旋转矩阵定义为:
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">,再让测试点对与旋转矩阵进行计算来进行旋转:
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">,那么旋转后,计算特征描述符可以采用如下计算方式:
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">。

<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami
ly:宋体;
mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:"Times New Roman";
mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:
FR">然而在进行坐标旋转后,坐标一般就变为了浮点型,此时我们要对图像进行重采样,可以使用的方法有最近邻方法、双线性内插方法、立体卷积方法等。

其中最近邻方法计算简单,它采用待采样点周围4<span style="mso-bidi-font-size: 12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">个相邻的像
素点中距离最近点的灰度值作为该点的灰度,这样仅仅考虑到了一个点的灰度值,没有考虑到其相邻区域内的像素点的影响,所以重采样后图像灰度值就会有明显的不连续现象。

另外立体卷积的方法计算量过大,在实时性方面有些困难。

对于待采样点的灰度值,其采用周围的4<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">个临近点的灰度值,然后在两个方向上做线性内插,从而得到带采样点的灰度值4 ORB特征
ORB<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">算子(Oriented FAST and Rotated BRIEF)<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">是Rublee<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">等在
2011<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">年提出来的,是建立在改进的FAST<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征和改进
的rBRIEF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征之上,
由于FAST<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征和BRIEF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征的运算速度都很快,则ORB<span style="mso-bidi-font-size: 12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">在运算速度上比SIFT<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">和
SURF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">等拥有绝对优势。

<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;
mso-ascii-font-family:"Times New
Roman";mso-hansi-font-family:"Times New Roman";
mso-bidi-font-family:"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language: FR">但由于BRIEF<span
style="mso-bidi-font-size:12.0pt;line-height:125%;
font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">不具有旋转不变性,因此很难用于旋转剧烈的场景,不过上文中我们已经给出了解决办法。


了使BRIEF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">具有旋转不变性,Rublee<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">等在
2011<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">年提出了ORB(oriented brief)<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">特征,在尺度变化不太大的环境中,ORB<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">特征在SLAM<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">系统中取得了了大量的应用,有时候我们的实际场景中比如带有机器的室内外场景中,这些机器可能其景深变化非常剧烈,这种特征描述子就会受到一定程度的限制。

也就是ORB<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">解决了BRIEF<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">的不具有旋转不变性和对噪声敏感的缺点,但ORB<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">没有解决尺度不变性。

ORB<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">算法的大致
步骤如图3-12<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">所示:BRIEF<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">特征向量的每个比特<span
style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New Roman";mso-font-kerning:0pt">(bit<span style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt">)<span
style="mso-bidi-font-size:12.0pt;
line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">具有的方差
和的均值接近0.5<span style="mso-bidi-font-size:12.0pt; line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">。

而加入特
征点的主方向后,特征向量的比特位分布就变得比较平均了,故其方差变小了,这导致特征之间的差异性变小,区分性能变差,因此,主方向引导后BRIEF<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
mso-font-kerning:0pt;mso-ansi-language:FR">后的性能有所下降。

为此要从矩阵
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">中筛选出具有高方差和高不相关的点对
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;
mso-ansi-language:FR">。

具体方法是:对
PASCAL2006<span style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
mso-font-kerning:0pt;mso-ansi-language:FR">图库<span style="mso-bidi-font-size:12.0pt;line-height:125%;font-fami ly:宋体;mso-ascii-font-family:
"Times New Roman";mso-hansi-font-family:"Times New Roman";mso-bidi-font-family:
"Times New
Roman";mso-font-kerning:0pt;mso-ansi-language:FR">上的图片,建立一个包含300k<span
style="mso-bidi-font-size:
12.0pt;line-height:125%;font-family:宋
体;mso-ascii-font-family:"Times New Roman";
mso-hansi-font-family:"Times New
Roman";mso-bidi-font-family:"Times New Roman";
mso-font-kerning:0pt;mso-ansi-language:FR">个关键点的训练集,每个
<span style="mso-bidi-font-size:12.0pt;line-height: 125%;font-family:宋体;mso-ascii-font-family:"Times New Roman";mso-hansi-font-family:
"Times New Roman";mso-bidi-font-family:"Times New Roman";mso-font-kerning:0pt;。

相关文档
最新文档