-
为了验证提出的基于镜面辅助的FPP测量方法用于不同景深范围内的多个被测物体测量的有效性、准确性和实用性,文中设计了两组实验:(1)利用传统FPP测量系统对6层石膏标准阶梯模型的三维轮廓测量;(2)传统FPP和提出的镜面辅助的FPP方法对不同景深范围内的两个标准乒乓球的三维轮廓同时测量对比实验。
-
在实验(1)中,运用传统FPP测量系统对一个90 mm×115 mm的6层石膏标准阶梯模型的阶梯表面进行测量,并比较每个阶梯平面的测量精度;在实验(2)中,分别采用传统FPP和镜面辅助的FPP测量系统同时测量不同景深范围内的两个标准乒乓球的表面,并将这两个乒乓球的拟合半径和标准半径进行比较,从而验证其测量精度。传统FPP测量系统和镜面辅助的FPP测量系统实验装置分别如图2(a)和图2(b)所示。测量系统主要包括分辨率为3840×2160 dpi的DLP投影仪(ViewSonic PX701-4 K),带有25 mm标准镜头的CMOS工业摄像机(大恒图像MER2-302-37 GM、分辨率为2048×1536),两片光学反射镜(50 mm×50 mm),两个三棱镜(10 mm×10 mm, 15 mm×15 mm),一台计算机。其中,10 mm×10 mm的三棱镜距离摄像机较近,15 mm×15 mm的三棱镜距离摄像机较远,使得摄像机的视场不被棱镜完全挡住而能够捕获#1号和#2号乒乓球条纹图像。实验中相机的视场范围约为170 mm ×130 mm。
-
为验证景深对传统FPP方法测量精度的影响,用传统FPP测量系统测量一个90 mm×115 mm的6层石膏标准阶梯模型的三维轮廓,标准石膏阶梯模型实物如图3(a)所示,阶梯模型每个台阶长宽高约为90 mm×20 mm×15 mm。摄像机和投影仪均聚焦在阶梯最远平面A处,摄像机捕获的6层石膏标准阶梯模型图像如图3(b)所示,测量的三维轮廓如图3(c)所示。从6层石膏阶梯模型的三维点云数据中提取每个阶梯平面的部分三维点云数据,每个平面的三维点云图如图4所示。
Figure 3. (a) Standard step model; (b) Fringe projection image of the step model ; (c) 3D profile of the standard step model
Figure 4. 3D point cloud of the step plane. (a) Step plane A; (b) Step plane B; (c) Step plane C; (d) Step plane D; (e) Step plane E; (f) Step plane F
采用多元线性回归方法拟合阶梯模型的每一个阶梯平面,由平面方程
$ ax + by + cz + d = 0 $ ,可得到阶梯平面A~F的拟合后的方程分别如下:假设阶梯平面由
$ n $ 个点组成,每个测得的点${P_i}(i = 1,2,\cdots n)$ 的三维坐标为$ ({x_i},{y_i},{z_i}) $ ,则可得到每个点到拟合平面的距离$ {e_i} $ :因此,拟合误差可由集合
$ error = [{e_1},{e_2},\cdots {e_n}] $ 表示,误差的分布直方图如图5所示。Figure 5. Histogram of error distribution of step plane. (a) Step plane A; (b) Step plane B; (c) Step plane C; (d) Step plane D; (e) Step plane E; (f) Step plane F
为进一步分析,可由公式(11)求出每个平面的均方根误差(Root mean square error, RMSE):
根据上述每个点的误差,计算得到的阶梯平面A~F的均方根误差结果如图6所示。另外,利用游标卡尺测量两相邻阶梯平面的平均深度作为真实深度,由拟合方程计算平面中心点之前的距离作为测量深度,其测量误差如表1所示,其中“A-B”表示平面A与B之间深度,其他以此类推。
Plane difference A-B B-C C-D D-E E-F Actual depth 18.70 20.10 19.26 20.36 18.08 Measured depth 17.32 19.84 23.33 27.52 33.88 Absolute error 1.38 0.26 4.07 7.16 15.80 Table 1. Depth error of the step plane (Unit: mm)
-
为验证提出的FPP方法对不同景深物体测量的可行性,将两个编号为#1号、#2号的标准乒乓球分别放置在摄像机不同景深范围内,分别采用传统FPP系统和镜面辅助的FPP系统分别同时测量两个标准乒乓球。利用游标卡尺5次测量的#1号和#2号乒乓球半径平均值分别为20.01、19.99 mm。摄像机和投影仪均聚焦在#1号乒乓球处,距离摄像机约487.5 mm,#2号乒乓球距离摄像机387.5 mm,#1号与#2号乒乓球的垂直距离约为100 mm。
传统FPP系统摄像机捕获的两个标准乒乓球表面的条纹图像如图7(a)所示,求解的绝对相位图如图7(c)所示,其表面的三维轮廓测量结果如图8(a)所示,通过拟合空间三维点云数据得到其拟合半径分别为19.420 mm和13.155 mm,与标准乒乓球半径的绝对误差和相对误差如表2所示。
Figure 7. (a) Fringe image captured by the traditional FPP system; (b) Fringe image captured by the mirror-assisted FPP system; (c) Absolute phase map measured by the traditional FPP system; (d) Absolute phase map measured by the mirror-assisted FPP system
在搭建镜面辅助的FPP测量系统前,首先相机和投影仪均聚焦在距离系统最远处的#1号乒乓球;并将棋盘格在#1号乒乓球附近任意摆放多次对测量系统进行标定;标定结束后通过调整三棱镜、光学反射镜在摄像机前的放置位置使得摄像机能够同时清晰捕获#1号、#2号乒乓球表面。摄像机捕获的两个标准乒乓球表面的条纹图像如图7(b)所示,求解的绝对相位图如图7(d)所示,#1号、#2号乒乓球正表面的三维轮廓测量结果如图8(b)所示,通过三维点云数据拟合得到的半径分别为19.463 mm和18.937 mm,与标准乒乓球半径的绝对误差和相对误差如表2所示。
Figure 8. Measurement results of 3D profile of table tennis surface. (a) Traditional FPP system; (b) Mirror-assisted FPP system
Table tennis Fitting radius/mm Absolute error/mm Relative error Traditional FPP Mirror-assisted FPP Traditional FPP Mirror-assisted FPP Traditional FPP Mirror-assisted FPP #1 19.420 19.463 0.59 0.547 2.9% 2.7% #2 13.155 18.937 6.835 1.053 34.3% 5.3% Table 2. Fitting radius and error of the table tennis
Fringe projection profilometry for 3D measurement of objects with different depth of fields
doi: 10.3788/IRLA20220088
- Received Date: 2022-04-10
- Rev Recd Date: 2022-05-20
- Publish Date: 2022-11-30
-
Key words:
- fringe projection /
- profile measurement /
- different depth of fields /
- mirror-assisted
Abstract: Fringe projection profilometry (FPP) was widely used in defect detection, reverse engineering, computer vision and other fields due to its non-contact and high measurement accuracy. However, traditional FPP can only obtain the 3D profile of the measured object within limited depth of field in a single measurement, and can not achieve accurate measurement of multiple measured objects with different depth simultaneously. A mirror-assisted FPP system was constructed by adding two mirrors and two prisms on the basis of traditional FPP system in this paper. The proposed method can transform the measured objects in different depth of field ranges to the same depth of field range, so as to achieve high-precision measurement of the 3D profiles of multiple measured objects in different depth of field ranges. The effect of depth of field on 3D profile measurement results was verified using a standard six-step gypsum model. Meanwhile, the profiles of two standard table tennis within different depth of field ranges were measured using the traditional FPP system and the proposed FPP system. The table tennis radius as error results were obtained from the measured profile. The relative errors of the focused and the unfocused table tennis measured by the traditional FPP were 2.9% and 34.3%, respectively. And the corresponding measured errors by proposed mirror-assisted FPP were 2.7% and 5.3%, respectively. The results show that the proposed method can compensate the errors caused by the depth of field, and verifies the feasibility of the proposed method in 3D measurement with different depth of field ranges.