-
本文中的研究对象为一款自制的激光雷达测量系统,其主要用途是通过采集并处理3维点云数据实现工业现场的大型零件的尺寸测量。由于商业3维激光扫描仪器价格极其昂贵,所以本测量系统采用更经济的3维激光测距传感器结合外部的旋转运动以实现3维点云的采集功能。该测量系统采用德国SICK公司生产的LMS111 2维激光雷达传感器,该传感器基于飞行时间测量原理,获取扫描平面内各点到雷达光心的距离值,其最大扫描范围及最大测量距离分别为270°和20m,最高分辨率达0.25°。系统中所用电机是日本松下MINAS的型号为MQMA P022P1的伺服电机,每转产生104个脉冲。如图 1所示,2维激光雷达安装在伺服电机的主轴上,并通过减速器(日本SHIMPO,VRSF-S9C-200)联接。如此,通过控制电机在指定角度范围内旋转,2维激光雷达就可以绕电机转轴进行相应角度的摆动,同时结合雷达自身的2维平面扫描即可实现3维空间的数据采集。将数据实时传入计算机进行数据处理,便可获取感兴趣的尺寸测量信息。
但是,由于原始点云数据是在雷达扫描平面的坐标系下采集的,所以首先需要建立一个固定的笛卡尔参考坐标系,并将原始点云数据向该参考坐标系下映射,才能得到被采集点云数据在3维空间的真实坐标。两坐标系之间的映射转换关系如图 2所示。
Figure 2. Mapping relationship between global reference coordinate system and radar scanning plane coordinate system
图 2中,Og-xgygzg代表全局笛卡尔参考坐标系,yg设为电机转轴轴线方向,zg为竖直向上方向。Or-xryr代表雷达扫描平面坐标系,坐标原点Or为雷达光心位置。xr轴位于竖直平面Og-xgzg内,yr轴与yg同向,两坐标原点间距p为系统内部标定常数,代表雷达光心到电机转轴轴线的距离。任一被测点M在雷达扫描平面坐标系中的位置可用距离l和扫描角α表示。β表示电机的转角,β=0设为雷达的起始位置。故l, α和β可唯一确定空间被测点M。经坐标变换,可得点M在参考坐标系中的坐标:
$\left\{ \begin{array}{l} x = l\sin\alpha\cos\beta + p\cos\beta \\ y = l\cos\alpha \\ z = l\sin\alpha\sin\beta + p\sin\beta \end{array} \right. $
(1) 式中,l, α, β均为测量系统的直接采集数据,经由上式变换即可得到点M在3维空间中的真实表述。
-
然而,由于l, α, β在各自的采集过程中均存在一定的不确定度,点M在参考坐标系中的位置也存在一定误差,进而影响目标尺寸测量的精度。为评价点M在空间中的不确定度,可建立其点位误差椭球模型。参照参考文献[15],可知该激光雷达测量系统在空间任意目标点M(x, y, z)的点位误差不确定性矩阵为:
${\mathit{\boldsymbol{D}}_{x,y,z}} = \mathit{\boldsymbol{K}}{\mathit{\boldsymbol{D}}_{l,\alpha ,\beta }}{\mathit{\boldsymbol{K}}^{\rm{T}}} $
(2) 式中, K为误差传播系数矩阵,由(1)式经过Jacobian矩阵变换而来:
$\mathit{\boldsymbol{K}} = \left[ {\begin{array}{*{20}{c}} { - \sin \alpha \cos \beta }&{ - l\cos \alpha \cos \beta }&{(l\sin \alpha + p)\sin \beta }\\ { - \cos \alpha }&{l\sin \alpha }&0\\ { - \sin \alpha \sin \beta }&{ - l\cos \alpha \sin \beta }&{ - (l\sin \alpha + p)\cos \beta } \end{array}} \right] $
(3) 式中,Dl, α, β为变量[l, α, β]T的协方差矩阵:
${\mathit{\boldsymbol{D}}_{l,\alpha ,\beta }} = \left[ {\begin{array}{*{20}{c}} {\sigma _l^2}&0&0\\ 0&{\sigma _\alpha ^2}&0\\ 0&0&{\sigma _\beta ^2} \end{array}} \right]\ $
(4) 其对角线上的3个元素分别表示l, α, β的方差值。通过先前大量的实验数据表明[8],对于该激光雷达测量系统,当测量距离l接近3m时,l测量值的方差为σl2=1.8225mm2(鉴于本文中的测量距离为3m左右,故此处取该值);根据LMS111激光雷达的最高分辨角0.25°,且工程中以作为3σ极限误差实现分辨率进行计算,有3σα=0.25°/2=0.125°,则σα=0.125°/3=0.042°;伺服电机每转一周产生104个脉冲,则同样以3σ极限误差实现分辨率计算,可得σβ=±360°/(2×104×3)=±0.006°。
设目标点M的真实坐标为(x*, y*, z*),测量坐标为(x, y, z),则测量误差可表示为向量r=[x-x*,y-y*,z-z*]T。由误差椭球理论可知,该目标点处的误差椭球方程为:
${\mathit{\boldsymbol{r}}^{\rm{T}}}{\mathit{\boldsymbol{ \boldsymbol{\varSigma} }}^{ - 1}}\mathit{\boldsymbol{r}} = {k^2} $
(5) 式中, Σ=Dx, y, z,k为放大系数。
至此,联立方程(2)式~(5)式,即可确立该激光雷达测量系统测量范围内任意点相应k值的点位误差椭球模型。图 3所示为任一点M(l, α, β)(中心位置实心五角星)在不同k下误差椭球及基于上面给定(σl, σα, σβ)的2000次随机仿真结果。可见,随机仿真测点(实心圆点)几乎都覆盖在k=3(97.77%置信区间)的椭球范围之内,且越靠近中心的位置,测量点出现的概率越大,符合3维正态分布规律[16]。
为更直观了解空间中各测点的误差分布情况,在激光雷达测量范围内的一立方区域中均匀选取150个目标测点,并分别绘制各点的误差椭球模型,由此得到该区域内空间点位的误差分布,如图 4所示。误差椭球的长轴均朝向激光的光心(为便于图形表达,雷达光心在图中未予以显示)方向,可见测距l的误差为主要误差源。
-
上述误差椭球模型虽然可以评价单一测点的点位不确定度,然而却不能直接应用到3维尺寸的测量评估上。从点云中提取尺寸信息时,常需要对点云平面进行拟合定位,进而实现与该平面关联的尺寸提取。因此,有必要建立点云拟合平面的不确定度模型。假设拟合平面与真实平面的微小偏角误差可以忽略,即认为两平面近似平行,则拟合平面与真实平面的间距可视为拟合平面的误差。目前最常用的点云平面拟合算法是随机抽样一致性算法(random sample consensus, RANSAC),其核心思想在于寻找平面模型去拟合点云数据[17]。
图 5所示为z=-1m的矩形平面区域内30个点的误差椭球分布。由空间点位误差分布规律可知,该平面内必存在一个关键点,其误差椭球在拟合平面的法线方向上的投影距离最小。而RANSAC算法要求拟合平面在一定阈值范围内包含最多的点,于是可以判定拟合平面应与所有误差椭球均相交,故关键点的误差椭球决定了拟合平面的两个极限位置,如图 6所示。
设关键点的真实坐标为(xc, yc, zc),测量值为(x, y, z),则由(5)式可得该点的误差椭球,如图 7所示。设真实平面和拟合平面的法向量为nv=(n1, n2, n3)代表向量。则拟合平面的误差可表示为:
$\begin{array}{l} d = \left| {\mathit{\boldsymbol{r}} \cdot {\mathit{\boldsymbol{n}}_v}} \right| = \left| {\left( {x - {x_c},y - {y_c},z - {z_c}} \right) \cdot {\mathit{\boldsymbol{n}}_v}} \right| = \\ \sqrt {{{\left[ {\left( {x - {x_{\rm{c}}}} \right){n_1} + \left( {y - {y_{\rm{c}}}} \right){n_2} + \left( {z - {z_{\rm{c}}}} \right){n_3}} \right]}^2}} \end{array} $
(6) 易知,椭球上存在一点P使d有最大值dmax。于是,设目标方程为:
$\begin{array}{l} \;\;\;\;\;\;\;\;\;\;\;\;\;\;f(x,y,z) = d = \\ \sqrt {{{\left[ {\left( {x - {x_{\rm{c}}}} \right){n_1} + \left( {y - {y_{\rm{c}}}} \right){n_2} + \left( {z - {z_{\rm{c}}}} \right){n_3}} \right]}^2}} \end{array} $
(7) 则求dmax转化为求有约束条件下的最优解,即:
$\begin{array}{l} \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\max f(x,y,z) = \\ \sqrt {{{\left[ {\left( {x - {x_{\rm{c}}}} \right){n_1} + \left( {y - {y_{\rm{c}}}} \right){n_2} + \left( {z - {z_{\rm{c}}}} \right){n_3}} \right]}^2}} \\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\left( {{\mathit{\boldsymbol{r}}^{\rm{T}}}{\mathit{\boldsymbol{ \boldsymbol{\varSigma} }}^{ - 1}}\mathit{\boldsymbol{r}} - {k^2}≤0} \right) \end{array} $
(8) 将(8)式的结果代入目标方程(7)式即可求得dmax。当k=1时, dmax视为拟合平面的不确定度。
基于误差椭球的激光测量系统的不确定度分析
Uncertainty analysis of laser radar measurement system based on error ellipsoid theory
-
摘要: 为了对3维激光扫描技术的测量精度做出评估,以激光雷达测量系统为研究对象,基于误差椭球理论建立了测量系统的点位误差模型;依据点云平面误差椭球的分布特性,提出了点云拟合平面的不确定度模型,用于评估与拟合平面关联的尺寸测量精度;通过对箱体类物体高度的测量实验,获得了实际测量不确定度,并与模型仿真结果进行了对比。结果表明,该模型可较准确地估算出高度的测量不确定度,从而验证了其有效性及实际意义。Abstract: In order to provide measurement accuracy evaluation for 3-D laser scanning devices, a laser radar measurement system (LRMS) was choosed as the research object and uncertainty model of point cloud fitting plane was proposed. After experimental verification, a single-point error ellipsoid model for the LRMS based on error ellipsoid theory was established. According to the distribution characteristics of plane error ellipsoid of point cloud, uncertainty model of point cloud fitting plane was proposed to evaluate and fit the dimension measurement accuracy associated with plane. Verification experiments were performed by using the LRMS to measure the height of a cubic object, and the measurement accuracy results were evaluated and compared with the simulated results using the proposed uncertainty model. The results show that the model can accurately estimate the uncertainty of height measurement. The research has the validity and practical significance.
-
Key words:
- measurement and metrology /
- uncertainty /
- error ellipsoid theory /
- laser radar /
- point cloud
-
-
[1] LIANG X L, KANKARE V, YU X W, et al. Automated stem curve measurement using terrestrial laser scanning[J].IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(3):1739-1748. doi: 10.1109/TGRS.2013.2253783 [2] CHEN P. Study on three-dimensional measurement of rotary type workpieces[D].Nanjing: Dongnan University, 2015: 15-20(in Chinese). [3] ANIL E B, TANG P B, AKINCI B, et al. Deviation analysis method for the assessment of the quality of the as-is building information models generated from point cloud data[J]. Automation in Construction, 2013, 35(1):507-516. [4] JOSÉ-IGNACIO R, SANCHEZ A, GLEZ-DE-RIVERA G, et al. Environment mapping using a 3-D laser scanner for unmanned ground vehicles[J]. Microprocessors and Microsystems, 2015, 39(8):939-949. doi: 10.1016/j.micpro.2015.10.003 [5] PONZ A, RODRÍGUEZ-GARAVITO C H, GARCÍA F, et al. Laser scanner and camera fusion for automatic obstacle classification in ADAS application[C]//Smart Cities, Green Technologies, and Intelligent Transport Systems: 4th International Conference, SMARTGREENS 2015, and 1st International Conference VEHITS 2015.Basel, Switzerland: Springer International Publishing, 2015: 237-249. [6] ZHOU S, GUO Y C, GAO Ch, et al. Rapid length measuring system for mobile and large scale cylinder workpieces based on 3-D laser scanning[J]. Optical Precision Engineering, 2014, 22(6):1524-1530(in Chinese). doi: 10.3788/OPE. [7] ZHANG Y C, HAN J X, FU X B, et al. Measurement and control technology of the size for large hot forgings[J]. Measurement, 2014, 49(1):52-59. [8] TIAN Zh S, GAO F, JIN Zh L, et al. Dimension measurement of hot large forgings with a novel time-of-flight system[J]. The International Journal of Advanced Manufacturing Technology, 2009, 44(1/2):125-132. [9] HE J, GAO F, WU Sh F, et al. Measure dimension of rotating large hot steel shell using pulse laser on PRRR robot[J]. Measurement, 2012, 45(7):1814-1823. doi: 10.1016/j.measurement.2012.04.001 [10] ZHENG D H, SHEN Y Zh, LIU Ch. 3-D laser scanner and its effect factor analysis of surveying error[J]. Engineering of Surveying and Mapping, 2005, 14(2):32-35(in Chinese). [11] CANG Y, JOHANN B. Characterization of a 2-D laser scanner for mobile robot obstacle negotiation[C]//Proceedings of the 2002 IEEE International Conference on Robotics and Automation.New York, USA: IEEE, 2002: 2512-2518. [12] PARK J H, SHIN Y D, BAE J H, et al. Spatial uncertainty model for visual features using a KinectTM sensor[J]. Sensors, 2012, 12(7):8640-8662. doi: 10.3390/s120708640 [13] WEI Y F. Surface reconstruction and quality evaluation based on laser radar system[D]. Shanghai: Shanghai Jiaotong University, 2009: 35-60(in Chinese). [14] SHI W Zh. Models and algorithms for three dimensional spatial information system[M].Beijing:Publishing House of Electronics Industry, 2007:227-230(in Chinese). [15] DU Zh Ch, WEI Y F, YAO Zh Q. Accuracy analysis of laser radar measurement system based on error ellipsoid model[J]. Journal of Shanghai Jiaotong University, 2009, 43(12):1882-1885(in Chinese). [16] WANG B, SHI W Zh, MIAO Z L. Confidence analysis of standard deviational ellipse and Its extension into higher dimensional euclidean space[J]. Public Library of Science, 2015, 10(3):1-17. [17] FISCHLER M A, BOLLES R C. Random sample consensus:a paradigm for model fitting with applications to image analysis and automated cartography[J].Communications of the ACM, 1981, 24(6):381-395. doi: 10.1145/358669.358692