-
为验证该标定方法的准确性,设计了静态测量重复性精度评定实验和绝对测量精度评定实验,实验设备及硬件参数如表1所示。其中,俯仰角和滚转角的测量精度是由倾角仪的精度决定,已知倾角仪出厂精度为0.01°,所以只对方位角进行精度分析即可。
Name Parameter Camera Model:Basler aca1300-60 gm Resolution:1280×1024 pixel Pixel size:5.3 μm×5.3 μm Frame rate:60 fps Total station Model:Leica TS12 Ranging accuracy:2 mm±2 ppm Angle accuracy:2″ Repeatability precision:0.1 mm Inclinometer Angle accuracy:0.01° One dimensional turntable Angle accuracy:2″ Table 1. Parameters of hardware
标定过程中,全站仪置于距离双屏视觉标靶9 m位置,利用调平旋钮将双屏视觉标靶调至与水平面平行,分别对前、后感光成像屏进行标定。根据上述标靶结构,全站仪测量网格角点坐标时,由于前屏遮挡后屏,导致全站仪无法直接测量后屏网格角点坐标,则利用全站仪先测量后屏网格角点坐标,再将前屏安装牢固,并测量前屏网格角点坐标,进而完成标定。其中,安装前屏导致的误差可通过调平旋钮和标靶内部倾角仪进行补偿。由于标定网格数量影响索引坐标的准确度和标定效率,所以前、后屏的标定网格分布为50行、50列,从而保证精度和标定效率。实验环境如图6所示。
-
为验证网格索引方法的静态测量重复性精度,该实验利用全站仪发射一束指示激光指向双屏视觉标靶,根据网格索引方法对前屏实时光斑三维坐标进行500次重复测量,测量结果如图7(a)所示;对双屏视觉标靶方位角进行500次重复测量,测量结果如图7(b)所示。
Figure 7. (a) Repeatability accuracy of static measurement of coordinates; (b) Repeatability accuracy of static measurement of heading angle
根据实验数据,测得
$x_{{f}}^{{s}}$ 坐标变化的峰峰值${x_{{{p - p}}}}$ 为0.10 mm,平均值$\overline {x_{{f}}^{{s}}}$ 为0.00 mm,标准差${s_{{x}}}$ 为0.02 mm;$y_{{f}}^{{s}}$ 坐标变化的峰峰值${y_{{{p - p}}}}$ 为0.00 mm,平均值$\overline {y_{{f}}^{{s}}}$ 为0.00 mm,标准差${s_{{y}}}$ 为0.00 mm;$z_{{f}}^{{s}}$ 坐标变化的峰峰值${z_{{{p - p}}}}$ 为0.13 mm,平均值$\overline {z_{{f}}^{{s}}}$ 为0.00 mm,标准差${s_{{z}}}$ 为0.03 mm;方位角测量重复性精度优于0.01°。 -
根据标定原理,该标定方法的精度直接影响标靶方位角测量的精度,而标靶方位角测量的关键在于精确测量实时光斑的三维坐标。因此,为验证该标定方法反算坐标的绝对测量精度,通过分析全站仪测量坐标与双屏视觉标靶反算坐标在X、Y、Z轴方向的偏差,对坐标的绝对测量精度进行评价。为验证该标定方法反算方位角的绝对测量精度,通过分析一维转台实际转动角度与双屏视觉标靶反算方位角变化量的差值,对方位角绝对测量精度进行评价。
-
该实验利用全站仪测量实时光斑在t系下的三维坐标
$P_{\text{1}}^{{t}}$ ,并通过三公共点坐标系转换算法求得s系下的三维坐标$P_{\text{1}}^{{s}}$ ,将其与网格索引实时光斑坐标$P_{\text{0}}^{{s}}$ 求差,则该差值为网格索引方法测量坐标的绝对测量精度。坐标精度评价实验数据如表2所示,表中${d_{{X}}}$ 、${d_{{Y}}}$ 、${d_{{Z}}}$ 表示三维坐标分别在X、Y、Z轴方向的偏差。No. X/mm Y/mm Z/mm ${d_{{X} } }$/mm ${d_{{Y} } }$/mm ${d_{{Z} } }$/mm 1 $P_{\text{0} }^{{s} }$ 24.48 −25.31 79.78 −0.33 0.26 −0.12 $P_{\text{1} }^{{s} }$ 24.15 −25.05 79.66 2 $P_{\text{0} }^{{s} }$ 69.87 −25.30 50.30 −0.94 −0.63 −0.73 $P_{\text{1} }^{{s} }$ 68.93 −25.93 49.57 3 $P_{\text{0} }^{{s} }$ 11.12 −25.69 31.06 −0.43 −0.77 0.51 $P_{\text{1} }^{{s} }$ 10.69 −26.46 31.57 4 $P_{\text{0} }^{{s} }$ −60.41 −25.76 8.43 −0.24 0.18 −0.14 $P_{\text{1} }^{{s} }$ −60.65 −25.58 8.29 5 $P_{\text{0} }^{{s} }$ 53.42 −26.47 −66.79 −0.83 −0.72 −0.34 $P_{\text{1} }^{{s} }$ 52.59 −27.19 −67.13 6 $P_{\text{0} }^{{s} }$ 64.26 −25.73 −43.86 −0.27 −0.17 −0.18 $P_{\text{1} }^{{s} }$ 63.99 −25.9 −44.04 Table 2. Experiment data of coordinate accuracy evaluation
通过坐标精度评价实验数据可知,该方法反算坐标的绝对测量精度优于1 mm。
-
该实验将双屏视觉标靶安装在高精度一维转台上,二者为刚性固连关系,将一维转台的实际转动角度与标靶解算的方位角变化角度作差,即可获得双屏视觉标靶方位角的绝对测量精度。无论双屏视觉标靶与一维转台的旋转中心是否重合,二者旋转角度均一致,并利用调平旋钮和倾角仪可以保证俯仰角和滚转角的测量误差小于0.01°。因此,方位角测量精度评价实验部分不需要考虑一维转台与双屏视觉标靶的初始配准问题。方位角精度评价实验原理如图8所示。
基于上述实验原理,对方位角的绝对测量精度进行评定,数据如图9所示。
实验结果表明,文中提出的网格标定方法能准确地标定前、后感光成像屏的位姿关系,坐标的绝对测量精度优于1.00 mm,姿态角的绝对测量精度优于0.05°。因此,该标定方法能够对前、后屏位姿关系进行准确快速地标定,并具有较高的精度和稳定性。
-
上述实验利用全站仪、一维转台、相机和倾角仪对双屏视觉标靶感光成像屏的位姿进行标定,并完成了精度测量。在测量过程中,影响系统精度的误差主要来源如表3所示。
Error type Error source Calibration errors of dual screen visual target The error of image recognition The error of total station measurement The error of coordinate system conversion Measurement errors of attitude angles Calculating error of light vector The error of inclinometer measurement Table 3. Errors analysis
(1)图像识别误差。光斑图像由Basler ACA 1300-60 gm工业相机进行采集,分辨率为1280×1024 pixel。若不考虑环境亮度、可见度等原因,图像识别误差约为0.5 pixel。
(2)全站仪测量误差。标定实验采用Leica TS12全站仪,免棱镜坐标测量精度约为2 mm±2 ppm。
(3)坐标系转换误差。该算法经过多次坐标系转换,根据实验可得,坐标系转换误差约为0.01 mm。
(4)激光光矢量解算误差。该算法求解的激光光矢量无法与物理空间绝对一致,则光矢量解算过程会引入误差。根据实验可得,光矢量解算精度约为0.01 mm。
(5)倾角仪测角误差。倾角仪型号为SST-460-30,测角分辨率0.0001°,测角精度为约0.01°。
根据上述误差来源的分析,双屏视觉标靶标定受到图像识别误差、全站仪测量误差、坐标系转换误差的影响,引入该误差源后,双屏视觉标靶坐标测量误差小于0.1 mm。由于全站仪测量误差、光矢量解算误差、倾角仪测量误差等因素造成姿态角的测量误差,在系统误差中加入该误差源后,双屏视觉标靶方位角测量误差小于0.01°。因此,上述误差源对标靶位姿解算影响较小,满足系统精度要求。
Calibration method and accuracy evaluation of visual target with double screen for straight pipe jacking machine
doi: 10.3788/IRLA20210933
- Received Date: 2022-03-20
- Rev Recd Date: 2022-04-15
- Publish Date: 2022-09-28
-
Key words:
- vision measurement /
- guidance of pipe jacking machine /
- coordinate point cloud /
- grid calibration
Abstract: Aiming at the problem that it was difficult to calibrate the position and attitude relationship between the front and rear photosensitive imaging screens of visual target with dual-screen, a method for calibrating the position and attitude of the photosensitive imaging screen based on points-cloud was proposed. The front and rear photosensitive imaging screens were respectively divided into grid arrays of n rows and n columns. Combining the image 2D coordinates which were obtained in real time by industrial cameras and the spatial 3D coordinates which were measured by a total station to obtain the 2D-3D mapping relationship of each grid corner point on the photosensitive imaging screen, the coordinate point cloud data were acquired. Next, the 3D coordinates of the coordinate point cloud data are converted to the target coordinate frame according to the three common point coordinate frame transformation algorithm, which can determine the positional relationship between the camera and front/rear photosensitive imaging screen. And then the positional relationship between the front and rear photosensitive imaging screens was solved by the grid indexing method. In order to evaluate the accuracy of target attitude measurement, the static repeatability and the absolute measurement accuracy evaluation experiments were designed. The experimental results show that the static repeatability accuracy of the coordinates is 0.13 mm, the absolute accuracy of the coordinates is 0.93 mm, the static repeatability accuracy of the heading angle is 0.01°, and the absolute accuracy of the heading angle is 0.05°. Therefore, the calibration method can realize the accurate calibration of the pose of two spatial planes, which has the characteristics of simple operation and high precision, and can be used for calibration of visual target with double screen.