|
发布时间: 2018-04-16 |
图像处理和编码 |
|
|
收稿日期: 2017-02-16; 修回日期: 2017-09-21
基金项目: 江苏省自然科学基金项目(BK20171249)
第一作者简介:
吴霄(1986-), 男, 苏州大学电子与通信工程专业硕士研究生, 主要研究方向为图像处理。E-mail:568983818@qq.com.
中图法分类号: TP301.6
文献标识码: A
文章编号: 1006-8961(2018)04-0478-12
|
摘要
目的 直接基于点云数据本身的拼合算法对点云模型的位置和重叠度有着较高的要求。为了克服这种缺陷,提出一种针对散乱点云的分步拼合算法。方法 不同于大多数已有的基于曲率信息的拼合算法,本文算法包含了一个序贯式的匹配点对筛选过程和一个基于霍夫变换的坐标变换参数估计过程。在筛选过程中,首先利用曲率相似度确定点云数据之间的初始匹配关系,然后利用刚体不变量特征邻域标识相似度以及持续特征直方图相似度对初始匹配点对进行连续两次筛选以便得到更为精确的匹配点对集。在参数估计阶段,通过对匹配点对的旋转矩阵和平移矢量的参数化处理,利用霍夫变换消除错误匹配点对对坐标变换参数估计的影响,从而得到更加准确的坐标变换参数,实现点云的3维拼合。结果 利用本文算法对两片部分重叠的点云数据进行了拼接实验。实验结果表明,本文算法能很好地实现对部分重叠点云的拼合。由于霍夫变换的引入,本文算法相较于经典的Ransac算法具有更高的正确率、稳定性以及抗噪性,在运行速度上也具有一定的优越性。结论 本文算法不仅能适用于任何具有任意初始相对位置的部分重叠点云的拼接,而且可以取得很高的拼合精度和很好的噪声鲁棒性。
关键词
点云拼合; 不变量特征; 霍夫变换; 持续特征直方图(PFH); 匹配; 法向量调整; 曲率
Abstract
Objective
Typical cloud data registration methods based on cloud data have high demands on the locations and overlapping degree of the two cloud datasets to be registered. For instance, the iterative closest point (ICP) algorithm establishes the corresponding relationship by searching the closest points, usually measured in Euclidean distance, between two point sets. Poor initial positions of two point cloud sets commonly lead to many erroneous correspondences, and eventually only local optimal solutions can be obtained rather than the global optimal solution. In addition, the ICP algorithm has high requirements on the overlapping degree between two clouds to be registered. To tackle this problem, we propose a
Key words
point clouds registration; invariant signature; hough transform; persistence feature histogram; match; normal vector adjustment; curvature
0 引言
最近3维(
对于大多数商用级的
直接的拼合方式,就是将不同视角的点云直接配准在一起。一种典型的配准方式就是最近点迭代(
另一种拼合方式是基于特征点的,特征点可以是洞、边沿、拐角、法向量等。
最近,在这两类方法的研究上取得了重要进展。
在基于特征的拼合方法中,通常需要用到微分信息。但是这类方法对噪声非常敏感,容易产生错误的匹配点对,同时对重叠率有较高要求。为此,提出一种逐层筛选的匹配点对计算策略,并利用霍夫变换确定最优的坐标变换参数,最终实现点云拼合。首先利用点云的曲率相似度确定初始匹配点对的集合,然后利用刚体不变量特征相似度对初始匹配点对进行筛选。由于曲率和刚体不变特征都是基于局部求导的,对噪声很敏感,所以匹配点对数组中会存在很多错误点对。为了减小噪声对匹配点对的影响,利用持续特征直方图(
1 点云拼接
1.1 几何信息的计算
对于规则排列的数据点可以进行二次参数曲面[25]拟合来估算法向量和曲率。由于很难对散乱无序的点云数据进行参数化,因此采用局部切平面参数化[26]来解决该问题,即用投影到局部切平面的点代替原始数据点来进行参数化。然而,该方法计算出来的法向量方向往往是不一致的,所以需要对法向量方向进行调整。为了解决这一问题,
1.2 来自不同点云的数据匹配
1.2.1 刚体不变量的定义
对点云
$ {\mathit{\boldsymbol{G}}_r}\left( {{p_i}} \right) = \left\{ {\mathit{\boldsymbol{N}}\left( q \right)\left| {q \in \mathit{\boldsymbol{R}}\left( {{p_i}} \right)} \right.} \right\} $ | (1) |
式中,
$ \mathit{\boldsymbol{D}}\left( {{p_i}} \right) = \left[ {\begin{array}{*{20}{c}} {\mathit{\boldsymbol{N}}\left( {{q_1}} \right) - g\left( {{p_i}} \right)}\\ {\mathit{\boldsymbol{N}}\left( {{q_2}} \right) - g\left( {{p_i}} \right)}\\ \vdots \\ {\mathit{\boldsymbol{N}}\left( {{q_K}} \right) - g\left( {{p_i}} \right)} \end{array}} \right] $ | (2) |
式中,
$ f\left( {{p_i}} \right) = \left\{ \begin{array}{l} {\left[ {\frac{{{\sigma _1}}}{{{\sigma _3}}},\frac{{{\sigma _2}}}{{{\sigma _3}}}} \right]^{\rm{T}}}\;\;\;\;\;{\sigma _3} \ne 0\\ {\left[ {1,1} \right]^{\rm{T}}}\;\;\;\;\;\;\;\;\;\;\;\;{\sigma _3} = 0 \end{array} \right. $ | (3) |
这里,利用奇异值的比值来定义不变特征标识,不仅仅使得它与坐标系无关,而且也可以消除点数量的影响。它可以很好地体现出曲面的特性,而且减少噪声的影响,增强鲁棒性。
1.2.2 基于曲率相似度的数据配准
假设目标点云
$ \begin{array}{l} {\mathop{\rm sgn}} \left( {{C_g}\left( {{p_i}} \right)} \right) = {\mathop{\rm sgn}} \left( {{C_g}\left( {{q_j}} \right)} \right)\\ {\mathop{\rm sgn}} \left( {{C_a}\left( {{p_i}} \right)} \right) = {\mathop{\rm sgn}} \left( {{C_a}\left( {{q_j}} \right)} \right) \end{array} $ | (4) |
式中,
$ \begin{array}{l} \left| {{C_a}\left( {{p_i}} \right) \cdot {C_g}\left( {{p_i}} \right)} \right| > {\varepsilon _1}\\ \left| {{C_a}\left( {{q_j}} \right) \cdot {C_g}\left( {{q_j}} \right)} \right| > {\varepsilon _1} \end{array} $ | (5) |
为了进一步剔除错误的匹配点对,本文引入点曲率相似度过滤对应点对,即
$ \begin{array}{l} \left| {\frac{{{K_1}\left( {{p_i}} \right) - {K_1}\left( {{q_j}} \right)}}{{{K_1}\left( {{p_i}} \right) + {K_1}\left( {{q_j}} \right)}}} \right| \le {\varepsilon _2}\\ \left| {\frac{{{K_2}\left( {{p_i}} \right) - {K_2}\left( {{q_j}} \right)}}{{{K_2}\left( {{p_i}} \right) + {K_2}\left( {{q_j}} \right)}}} \right| \le {\varepsilon _2} \end{array} $ | (6) |
式中,
1.2.3 基于刚体不变量的邻域标识相似度过滤
从理论上来说,如果两个点是一对正确的匹配点对,那么它们不仅仅在曲率上有相似性,而且其各自邻域内对应点的不变量特征标识也应该相似。利用点云特征不变量标识间的归一化互相关系数(
$ \begin{array}{*{20}{c}} {Ncc\left( {{p_i},{q_j}} \right) = }\\ {\frac{{\sum\limits_{l = 1}^k {\left( {f\left( {{p_l}} \right) - {c_{f{p_i}}}} \right) \cdot \left( {f\left( {{q_l}} \right) - {c_{f{q_j}}}} \right)} }}{{\sqrt {\sum\limits_{l = 1}^k {{{\left\| {\left( {f\left( {{p_l}} \right) - {c_{f{p_i}}}} \right)} \right\|}^2}} } \sqrt {\sum\limits_{l = 1}^k {{{\left\| {\left( {f\left( {{q_l}} \right) - {c_{f{q_j}}}} \right)} \right\|}^2}} } }}} \end{array} $ | (7) |
式中,
1.2.4 基于持续特征直方图(${\rm PFH}$ )的筛选
使用持续特征直方图(
$ \left( {{p_s},{p_t}} \right) = \left\{ \begin{array}{l} \left( {{p_i},{p_j}} \right)\;\;\;\;\alpha \le \beta \\ \left( {{p_j},{p_i}} \right)\;\;\;\;\alpha > \beta \end{array} \right. $ | (8) |
式中,
图 1中,
$ \mathit{\boldsymbol{u}} = {\mathit{\boldsymbol{n}}_s},\mathit{\boldsymbol{v}} = \left( {{\mathit{\boldsymbol{p}}_t} - {\mathit{\boldsymbol{p}}_s}} \right) \times \mathit{\boldsymbol{u}},\mathit{\boldsymbol{w}} = \mathit{\boldsymbol{u}} \times \mathit{\boldsymbol{v}} $ | (9) |
利用上述达布架构,计算点对
$ {p_f} = \sum\limits_{i = 1}^4 {step\left( {{s_i},{f_i}} \right) \cdot {2^{i - 1}}} $ | (10) |
式中
$ step\left( {{s_i},{f_i}} \right) = \left\{ \begin{array}{l} 0\;\;\;\;{f_i} < {s_i}\\ 1\;\;\;\;{f_i} > {s_i} \end{array} \right. $ | (11) |
$ \begin{array}{*{20}{c}} {{f_1} = \left\langle {\mathit{\boldsymbol{v}},{\mathit{\boldsymbol{n}}_t}} \right\rangle }\\ {{f_2} = \left\| {{p_t} - {p_s}} \right\|}\\ {{f_3} = \left\langle {\mathit{\boldsymbol{u}},{p_t} - {p_s}} \right\rangle /{f_2}}\\ {{f_4} = a\tan \left( {\left\langle {\mathit{\boldsymbol{w}},{\mathit{\boldsymbol{n}}_t}} \right\rangle ,\left\langle {\mathit{\boldsymbol{u}},{\mathit{\boldsymbol{n}}_t}} \right\rangle } \right)} \end{array} $ | (12) |
因为
显然,通过式(10)可以将点对
利用
1.3 基于霍夫变换的聚类
经过基于曲率相似度、邻域标识相似度、持续直方图的筛选之后,可以得到一个初步的候选匹配点对集合
$ \begin{array}{*{20}{c}} {\mathit{\boldsymbol{Match}} = \left\{ {\left( {{{p'}_i},{{p'}_i}} \right)\left| {{{p'}_i} \in \mathit{\boldsymbol{P}},{{p'}_i} \in \mathit{\boldsymbol{Q}}} \right.,} \right.}\\ {\left. {i = 1,2, \cdots ,{N_m}} \right\}} \end{array} $ | (13) |
式中,
一个匹配点对对应的是一组坐标系的变换参数,
1.3.1 局部坐标系统的建立
通过偏差矩阵
$ \begin{array}{*{20}{c}} {\left| {{\mathit{\boldsymbol{z}}_{{p_i}}} \cdot \mathit{\boldsymbol{N}}\left( {{p_i}} \right)} \right| = \max \left\{ {\left| {{\mathit{\boldsymbol{v}}_1} \cdot \mathit{\boldsymbol{N}}\left( {{p_i}} \right)} \right|,} \right.}\\ {\left. {\left| {{\mathit{\boldsymbol{v}}_2} \cdot \mathit{\boldsymbol{N}}\left( {{p_i}} \right)} \right|,\left| {{\mathit{\boldsymbol{v}}_3} \cdot \mathit{\boldsymbol{N}}\left( {{p_i}} \right)} \right|} \right\}} \end{array} $ | (14) |
式中,
$ \left\{ \begin{array}{l} \left( {g\left( {{p_i}} \right) - {p_i}} \right) \cdot {\mathit{\boldsymbol{x}}_{{p_i}}} \ge 0\\ \left( {g\left( {{p_i}} \right) - {p_i}} \right) \cdot {\mathit{\boldsymbol{z}}_{{p_i}}} \ge 0 \end{array} \right. $ | (15) |
式中,
1.3.2 坐标系间变换的参数化
假设
$ \left\{ \begin{array}{l} \mathit{\boldsymbol{R}} = \left( {{\mathit{\boldsymbol{x}}_p},{\mathit{\boldsymbol{y}}_p},{\mathit{\boldsymbol{z}}_p}} \right){\left( {{\mathit{\boldsymbol{x}}_q},{\mathit{\boldsymbol{y}}_q},{\mathit{\boldsymbol{z}}_q}} \right)^{ - 1}}\\ \mathit{\boldsymbol{t}} = \mathit{\boldsymbol{p}} - \mathit{\boldsymbol{Rq}} \end{array} \right. $ | (16) |
式中,
$ {\mathit{\boldsymbol{Q}}_r} = \left( {{q_0},v} \right) = {q_0} + {q_1}\mathit{\boldsymbol{i}} + {q_2}\mathit{\boldsymbol{j}} + {q_3}\mathit{\boldsymbol{k}} $ | (17) |
式中,
$ \mathit{\boldsymbol{R}} = \left[ {\begin{array}{*{20}{c}} \begin{array}{l} q_0^2 + q_1^2 - q_2^2 - q_3^2\\ 2\left( {{q_1}{q_2} + {q_0}{q_3}} \right)\\ 2\left( {{q_1}{q_3} - {q_0}{q_2}} \right) \end{array}&\begin{array}{l} 2\left( {{q_1}{q_2} - {q_0}{q_3}} \right)\\ q_0^2 - q_1^2 + q_2^2 - q_3^2\\ 2\left( {{q_2}{q_3} + {q_0}{q_1}} \right) \end{array}&\begin{array}{l} 2\left( {{q_1}{q_3} + {q_0}{q_2}} \right)\\ 2\left( {{q_2}{q_3} - {q_0}{q_1}} \right)\\ q_0^2 - q_1^2 - q_2^2 + q_3^2 \end{array} \end{array}} \right] $ | (18) |
结合式(16)(18)可以得到参数
$ {\mathit{\boldsymbol{Q}}_r} = \cos \left( {\frac{\theta }{2}} \right) + \omega \sin \left( {\frac{\theta }{2}} \right) $ | (19) |
旋转参数
$ \theta = 2{\cos ^{ - 1}}{q_0},\;\;\;\;\omega = \left\{ \begin{array}{l} \frac{v}{{\sin \left( {\theta /2} \right)}}\;\;\;\;\;\;\theta \ne 0\\ 0\;\;\;\;\;\;\;\;\;\;\;\;\;\;其他 \end{array} \right. $ | (20) |
利用上述公式可以得到集合
1.3.3 参数空间的划分
如前所述,通过对参数空间中不同单元内点的计数来确定点在参数空间中点的密度分布。因此,密度分布的精度取决于单元的大小以及空间划分的均匀性。实际上,需要划分两类参数空间:一类是球体表面的划分(即:相应于旋转轴
本文使用一种基于正多面体的方法对球体表面进行划分。这种方式的优点是可以对单元无限细分而不改变它们的形状。而且,在划分过程中该方法可以确保均匀性。与其他多面体相比,正二十面体具有以下优点:在所有已知的正多面体中,它拥有最多面的,它的面都是正三角形,4个细分以后的子三角形具有更相似的形状。基于以上考虑,本文采用与球体内接的正二十面体对球体表面进行划分,并通过细分正二十面体的表面实现对球面的进一步细分(如图 3所示)。图 3(a)给出了初始三角形以及对它们进行细分的示意图。由该图可以看出,在细分过程中,该方法利用三角形各边的中点将其划分成4个子三角形。图 3(b)给出了为子三角形分配索引的方式。
假设将被划分的球体中心与坐标原点
$ \mathit{\boldsymbol{a}} \cdot \mathit{\boldsymbol{x}} > 0,\mathit{\boldsymbol{b}} \cdot \mathit{\boldsymbol{x}} > 0,\mathit{\boldsymbol{c}} \cdot \mathit{\boldsymbol{x}} > 0 $ | (21) |
式中,
$ {\mathit{\boldsymbol{J}}_i} = \mathit{\boldsymbol{J}}{\mathit{\boldsymbol{M}}_i}\;\;\;\left( {i = 0,1,2,3} \right) $ | (22) |
式中
$ {\mathit{\boldsymbol{M}}_0} = \left[ {\begin{array}{*{20}{c}} 1&{ - 1}&0\\ 0&1&0\\ 0&{ - 1}&1 \end{array}} \right]\;\;\;\;\;{\mathit{\boldsymbol{M}}_1} = \left[ {\begin{array}{*{20}{c}} 1&0&{ - 1}\\ 0&1&{ - 1}\\ 0&0&1 \end{array}} \right] $ |
$ {\mathit{\boldsymbol{M}}_2} = \left[ {\begin{array}{*{20}{c}} 1&0&0\\ { - 1}&1&0\\ { - 1}&0&1 \end{array}} \right]\;\;\;\;{\mathit{\boldsymbol{M}}_3} = \left[ {\begin{array}{*{20}{c}} { - 1}&1&1\\ 1&{ - 1}&1\\ 1&1&{ - 1} \end{array}} \right] $ |
如上所述,本文对球面的划分始于内接正二十面体,并以如图 3所示的递归的方式进行细分,直到满足收敛条件为止。如果细分次数是
可以将旋转角
1.3.4 变换参数的估算
通过对
2 实验结果和讨论
为了验证本文算法,利用本文算法对来自于斯坦福大学官网的点云数据进行了拼接实验。该点云数据共有35 947个点。将整份点云数据分割为部分重叠的两份点云数据,图 4给出了两份部分重叠的点云数据,左图是目标点云数据
如前所述,在进行微分信息计算之前,法向量需要调整为一致向外。图 5给出了对图 4中目标点云
由曲率相似度确定的初始匹配点对的筛选是基于刚体不变特征的。因此,该算法的性能取决于不同点云的特征值的不变性。图 6给出的是两片点云数据的刚体不变特征和曲率特征的示意图,图 6(a)给出的是两片不同点云的刚体不变量特征值的分布情况,其中横轴和纵轴分别为
由于基于刚体不变特征得到的匹配点对中还存在大量错误的匹配点对,为了进一步提高匹配点对的正确率,利用了PFH算法对匹配点对进行进一步的剔除。为了验证该方法的合理性,图 7给出了
为了验证本文算法对噪声的鲁棒性,对目标点云中点的3个坐标值分别添加3组相互独立,且均值为零,标准方差为点平均间距的1%5%高斯噪声。图 8给出了在曲率相似度匹配基础上经过刚体不变量筛选(蓝线)以及在此基础上进一步经过PFH筛选(红线)后的正确率随噪声大小的变化曲线。这里,匹配正确率定义为
经过上面的PFH过滤后得到了一组新的匹配点对,下面需要使用霍夫变换,求取两片点云数据间的旋转和平移参数。为了利用霍夫变换进行参数检测,需要对参数空间进行划分。本文将旋转角
为了研究霍夫变换对提升算法抗噪性能的效果,将其与经典的Ransac[30]算法进行了比较。其中,对球面空间的匹配参数设定随机采样次数
表 1
Ransac和Hough在不同噪声下的匹配点对正确率统计
Table 1
Statistics of Ransac and Hough matching point accuracies at different noise
算法 | 噪声 | ||||
1% | 2% | 3% | 4% | 5% | |
Ransac/% (10次运行结果) |
87.47 | 74.63 | 60.87 | 40.01 | 18.39 |
88.28 | 75.05 | 64.08 | 37.86 | 17.89 | |
88.33 | 74.58 | 62.55 | 32.31 | 20.73 | |
88.72 | 75.69 | 59.84 | 41.86 | 22.67 | |
87.91 | 76.55 | 60.89 | 34.93 | 18.89 | |
88.31 | 75.16 | 59.14 | 39.55 | 17.17 | |
88.52 | 74.68 | 65.98 | 41.54 | 17.39 | |
88.28 | 74.79 | 58.73 | 40.31 | 15.38 | |
88.63 | 75.32 | 62.21 | 43.44 | 17.28 | |
88.14 | 77.07 | 60.41 | 36.69 | 18.84 | |
Hough/% | 92.67 | 87.5 | 71.88 | 58.82 | 50 |
表 2
Ransac和Hough运行时间统计
Table 2
Statistics of Ransac and Hough runtime
噪声 | |||||
1% | 2% | 3% | 4% | 5% | |
Ransac/s | 50.02 | 42.31 | 38.25 | 35.571 | 30.39 |
Hough/s | 22.63 | 22.83 | 22.25 | 21.56 | 21.73 |
图 11给出了不同方法对含2%噪声的点云数据进行拼接的结果。显然使用霍夫变换的方法能得到更佳的拼合结果。从图 10和图 11也可以看出, 参数聚类在点云拼接中的重要性和有效性。
3 结论
提出了一种针对散乱点云数据的拼合算法。该算法首先利用曲率相似度建立初始匹配点对。然后,分别利用刚体不变特征以及持续特征直方图(PFH)对初始匹配点对进行连续两次筛选形成新的匹配点对。通过基于PFH的二次筛选,匹配点对的正确率提升2倍左右。为了避免误匹配对点云拼接的影响,本文提出对坐标变换进行参数化处理并利用霍夫变换对坐标变换参数进行最优估计,从而最终实现点云的拼合。实验结果表明, 该方法可以应用在对任何部分重叠或者全部重叠的具有任意相对位置的点云数据的拼接中,并且保持着较高的精度和较强的鲁棒性。由于本文算法需要分步实现,比较耗时,下一步将研究点云数据精简策略及其相应的拼合方法,以进一步提高算法的效率。
参考文献
-
[1] Chen J, Wu X J, Wang M Y, et al. Human body shape and motion tracking by hierarchical weighted ICP[C]//International Symposium on Visual Computing. Las Vegas, NV, USA: Springer-Verlag, 2011: 408-417. [DOI:10.1007/978-3-642-24031-7_41]
-
[2] Gutiérrez-Heredia L, D'Helft C, Reynaud E G. Simple methods for interactive 3D modeling, measurements, and digital databases of coral skeletons[J]. Limnology and Oceanography:Methods, 2015, 13(4): 178–193. [DOI:10.1002/lom3.10017]
-
[3] Vasiliauskas A, Šidlauskas A, Šaferis V, et al. Applications of 3D maxillary dental arch scanning for mathematical prediction of orthodontic treatment need for complete unilateral cleft lip and palate patients[J]. Electronics and Electrical Engineering, 2010, 100(4): 107–112.
-
[4] Haghighipanah M, Miyasaka M, Li Y M, et al. Unscented Kalman Filter and 3D vision to improve cable driven surgical robot joint angle estimation[C]//Proceedings of 2016 IEEE International Conference on Robotics and Automation. Stockholm, Sweden: IEEE, 2016: 4135-4142. [DOI:10.1109/ICRA.2016.7487606]
-
[5] Husstedt H, Ausserlechner U, Kaltenbacher M. Precise alignment of a magnetic sensor in a coordinate measuring machine[J]. IEEE Sensors Journal, 2010, 10(5): 984–990. [DOI:10.1109/JSEN.2009.2037235]
-
[6] Zhu S P, Gao Y. Noncontact 3-D coordinate measurement of cross-cutting feature points on the surface of a large-scale workpiece based on the machine vision method[J]. IEEE Transactions on Instrumentation and Measurement, 2010, 59(7): 1874–1887. [DOI:10.1109/TIM.2009.2030875]
-
[7] Chen W Y, Tung C K, Wang C M, et al. The non-contact human-height measurement scheme[C]//Proceedings of 2011 International Conference on Machine Learning and Cybernetics. Guilin, China: IEEE, 2011: 572-575. [DOI:10.1109/ICMLC.2011.6016821]
-
[8] Mao Y X, Guo J P, Liang Y M, et al. Analysis of noise characteristics in an optical coherence tomographic system[J]. Acta Optica Sinica, 2005, 25(3): 324–330. [毛幼馨, 郭建平, 梁艳梅, 等. 低相干光断层扫描系统的噪声分析与研究[J]. 光学学报, 2005, 25(3): 324–330. ] [DOI:10.3321/j.issn:0253-2239.2005.03.008]
-
[9] Mohammadzade H, Hatzinakos D. Iterative closest normal point for 3D face recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(2): 381–397. [DOI:10.1109/TPAMI.2012.107]
-
[10] Gómez-García-Bermejo J, Zalama E, Feliz R. Automated registration of 3D scans using geometric features and normalized color data[J]. Computer-Aided Civil and Infrastructure Engineering, 2013, 28(2): 98–111. [DOI:10.1111/j.1467-8667.2012.00785.x]
-
[11] Bucksch A, Khoshelham K. Localized registration of point clouds of botanic trees[J]. IEEE Geoscience and Remote Sensing Letters, 2013, 10(3): 631–635. [DOI:10.1109/LGRS.2012.2216251]
-
[12] Besl P J, Mckay N D. A method for registration of 3-D shapes[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1992, 14(2): 239–256. [DOI:10.1109/34.121791]
-
[13] Chen Y, Medioni G. Object modelling by registration of multiple range images[J]. Image and Vision Computing, 1992, 10(3): 145–155. [DOI:10.1016/0262-8856(92)90066-C]
-
[14] Masuda T, Yokoya N. A robust method for registration and segmentation of multiple range images[C]//Proceedings of the 2nd Cad-Based Vision Workshop. Champion, PA, USA: IEEE, 1994: 106-113. [DOI:10.1109/CADVIS.1994.284510]
-
[15] Jiang J, Cheng J, Chen X L. Registration for 3-D point cloud using angular-invariant feature[J]. Neurocomputing, 2009, 72(16-18): 3839–3844. [DOI:10.1016/j.neucom.2009.05.013]
-
[16] Xin W, Pu J X. Point cloud integration base on distances between points and their neighborhood centroids[J]. Journal of Image and Graphics, 2011, 16(5): 886–891. [辛伟, 普杰信. 点到邻域重心距离特征的点云拼接[J]. 中国图象图形学报, 2011, 16(5): 886–891. ] [DOI:10.11834/jig.20110515]
-
[17] Chen H, Bhanu B. 3D free-form object recognition in range images using local surface patches[J]. Pattern Recognition Letters, 2007, 28(10): 1252–1262. [DOI:10.1016/j.patrec.2007.02.009]
-
[18] Zhu Y J, Zhou L S, Zhang L Y. Registration of scattered cloud data[J]. Journal of Computer-Aided Design & Computer Graphics, 2006, 18(4): 475–481. [朱延娟, 周来水, 张丽艳. 散乱点云数据配准算法[J]. 计算机辅助设计与图形学学报, 2006, 18(4): 475–481. ] [DOI:10.3321/j.issn:1003-9775.2006.04.001]
-
[19] Akca D. Matching of 3D surfaces and their intensities[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2007, 62(2): 112–121. [DOI:10.1016/j.isprsjprs.2006.06.001]
-
[20] Chen J, Wu X J, Wang M Y, et al. 3D shape modeling using a self-developed hand-held 3D laser scanner and an efficient HT-ICP point cloud registration algorithm[J]. Optics & Laser Technology, 2013, 45: 414–423. [DOI:10.1016/j.optlastec.2012.06.015]
-
[21] He B W, Lin Z M, Li Y F. An automatic registration algorithm for the scattered point clouds based on the curvature feature[J]. Optics & Laser Technology, 2013, 46: 53–60. [DOI:10.1016/j.optlastec.2012.04.027]
-
[22] Papazov C, Burschka D. Stochastic global optimization for robust point set registration[J]. Computer Vision and Image Understanding, 2011, 115(12): 1598–1609. [DOI:10.1016/j.cviu.2011.05.008]
-
[23] Harada T, Kuniyoshi Y. Graphical Gaussian vector for image categorization[C]//Proceedings of the 26th Annual Conference on Neural Information Processing Systems. Lake Tahoe: NIPS, 2012: 1547-1555. http://www.mendeley.com/catalog/graphical-gaussian-vector-image-categorization/
-
[24] Rusu R B, Blodow N, Marton Z C, et al. Aligning point cloud views using persistent feature histograms[C]//Proceedings of 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems. Nice, France: IEEE, 2008: 3384-3391. [DOI:10.1109/IROS.2008.4650967]
-
[25] Yan D M, Wang W P, Liu Y, et al. Variational mesh segmentation via quadric surface fitting[J]. Computer-Aided Design, 2012, 44(11): 1072–1082. [DOI:10.1016/j.cad.2012.04.005]
-
[26] He M F, Zhou L S, Shen H C. Curvature estimation of scattered-point cloud data and its application[J]. Journal of Nanjing University of Aeronautics & Astronautics, 2005, 37(4): 515–519. [贺美芳, 周来水, 神会存. 散乱点云数据的曲率估算及应用[J]. 南京航空航天大学学报, 2005, 37(4): 515–519. ] [DOI:10.3969/j.issn.1005-2615.2005.04.024]
-
[27] Hoppe H, Derose T, Duchamp T, et al. Surface reconstruction from unorganized points[J]. ACM SIGGRAPH Computer Graphics, 1992, 26(2): 71–78. [DOI:10.1145/142920.134011]
-
[28] Eigenstetter A, Ommer B. Visual recognition using embedded feature selection for curvature self-similarity[C]//Proceedings of the Conference on Advances in Neural Information Processing Systems. Cambridge: MIT Press, 2012: 377-385.
-
[29] Perumal L. Quaternion and its application in rotation using sets of regions[J]. International Journal of Engineering and Technology Innovation, 2011, 1(1): 35–52.
-
[30] Fischler M A, Bolles R C. Random sample consensus:a paradigm for model fitting with applications to image analysis and automated cartography[J]. Readings in Computer Vision, 1987: 726–740. [DOI:10.1016/B978-0-08-051581-6.50070-2]