Considering the limited measurement range of a machine vision method for the threedimensional (3D) surface measurement of largescale components, a noncontact and flexible global measurement method combining a multiple field of view (FOV) is proposed in this paper. The measurement system consists of two theodolites and a binocular vision system with a transfer mark. The process of multiple FOV combinations is described, and a new global calibration method is proposed to solve the coordinate system unification issue of different instruments in the measurement system. In addition, a highprecision image acquisition method, which is based on laser stripe scanning and centre line extraction, is discussed to guarantee the measurement efficiency. With the measured 3D data, surface reconstruction of largescale components is accomplished by data integration. Experiments are also conducted to verify the precision and effectiveness of the global measurement method.
In the field of modern manufacturing and assembly, the accuracy of largescale components of dimensions is required to guarantee automatic assembly of highend equipment. To ensure that the components are manufactured as designed, the dimensions of the components have to be accurately measured [
Machine vision measurement has such advantages as noncontact, high efficiency, and high accuracy, so it has been extensively applied to industry [
First, the principle and measurement process of the proposed method are introduced. Second, the calibration method of the measurement system is presented. Third, the algorithms of laser stripe extraction, image matching, and reconstruction are described. Last, the flexibility and precision of the measurement method are verified by experiments, as discussed in Section
The measurement system is a binocular vision system that consists of two theodolites and a transfer mark that is designed for the transition of views. In the measuring process, two theodolites that are placed in the back are employed as the global control station for field combination and data integration. The binocular vision system with the transfer mark is placed in the front to capture an image of the component surface. The global measurement coordinate system is established on the left theodolite to ensure that the binocular vision system can be placed in any position in the measurement range of the theodolites. By transferring the surface information that is captured by cameras in different positions into the global coordinate system, highprecision data integration can be achieved. The measurement principle of the system is shown in Figure
Schematic of the measurement system.
As the measurement system consists of different types of instruments, the measurement coordinate systems, respectively, established on these instruments are different as well. To reconstruct an entire surface, the coordinates that are measured by these instruments must be unified in the global coordinate system. The
The binocular vision system consists of two industry complementary metal oxide semiconductor (CMOS) cameras. The coordinate system of the left camera—
In the process of measurement, the large component to be measured must be divided into multiple fields of view and measured several times; thus, the cameras must be moved to different positions. To connect the mobile camera coordinate systems with the fixed global coordinate system, a transfer mark is taken into use of the binocular vision system. The relative position between the mark and the cameras remains unchanged during the measurement. By measuring the feature points on the mark, two theodolites in the back can obtain the position and the direction of the binocular vision system. The coordinate system of the transfer marks
The measurement process of the proposed system is as follows: first, the system is calibrated, which includes the calibration of two theodolites, the binocular cameras, and their transformation relation which is calibrated using the transfer mark. Second, the surface of the largescale component is artificially divided into several parts and separately measured by the binocular cameras at different positions; based on binocular vision method with assisted laser, the 3D point cloud data of measured part is obtained [
The calibration of the measurement system is the foundation of the 3D measurements. The proposed measurement system consists of different instruments. Thus, the highprecision calibration of all instruments is essential for ensuring the accuracy of large field measurements.
To measure the large surface of a component, the binocular vision system must be moved several times to acquire regional characteristic information. Due to the changed measuring position, the spatial location of the camera coordinate system varies in the meantime. Thus, an accurate transformation relationship between the camera coordinate system and the global coordinate system must be established.
In this paper, the global coordinate system is established on the left theodolite. The twotheodolite system is calibrated according to the twotheodolite spatial threedimensional (3D) measurement model. The binocular cameras are calibrated using Zhang’s calibration method [
The global coordinate system is established on the left theodolite according to the perspective projection model [
Coordinate systems for the two theodolites.
Similarly, the 3D coordinate system
The coordinate vector of point
According to (
By measuring a target of certain length thrice using the two theodolites, the rotation matrix
In the global coordinate system, the coordinate vector of any point
Two industrial cameras with high resolution are employed in the 3D surface measurement system. To measure the surface of the largescale component that is blocked, the binocular cameras must be moved to different positions.
The pixel coordinates vector of point
In the process of measurement, the transfer mark is essential for the transformation from the camera coordinate system to the global coordinate system.
As the transfer mark is not in the opposite direction to the binocular cameras, the feature points on the mark cannot be captured by the cameras. The transformation relation between the camera coordinate system and the transfer mark coordinate system cannot be directly established. As a result, the calibration process must be divided into the following two steps.
Binocularcamera measurement system with the transfer mark.
The coordinate system of the transfer mark
By measuring the feature points on the transfer mark with two theodolites, the transformation matrix
After the calibration, the relative position between the binocular cameras and the transfer mark remains unchanged. In the process of measurement, data transformation from the binocularcamera coordinate system to the global coordinate system is achieved by calibrating
In a single FOV, the feature information of the component is measured using a laser scanning method. The assisted laser stripes that are projected on the surface of the component are captured by a binocular vision measurement system.
Prior to reconstructing the laser stripes, the centres of the stripes are extracted using a grey centroid method [
After matching the feature points of the left and right images, reconstruction of the image can be realized according to the binocular reconstruction principle. The coordinate vector of any point
A binocular vision system is utilized to capture the largescale component in different positions. Measuring the spatial positions of the feature points on the transfer mark with the two theodolites,
The global measurement system for the 3D surface of a largescale component was set up in a laboratory, as shown in Figure
Global measurement system for 3D surface measurement.
According to the calibration method proposed in Section
The calibration results of the binocular cameras are shown as follows:
The intrinsic parameters of the left and right cameras are
The extrinsic parameters of the binocular cameras are
According to the calibration method that was discussed in Section
With
To evaluate the measurement accuracy, a long onedimensional (1D) target with two characteristic points (the distance between the two points is 1225.0214 mm) is considered. The measurement FOV is divided into two parts. The evaluation process is as follows: first, the target is placed in front of the cameras, as shown in Figure
Evaluation of the measurement accuracy (mm).
Position 



Measured value/mm  Standard value/mm  Deviation/% 

1  375.4370  −234.7837  166.1296  1224.63  1225.0214  0.032% 
−185.4221  852.2557  107.0487  
2  −203.9730  −205.5046  131.8487  1223.753  0.103%  
437.1248  834.3868  203.9114  
3  −79.5582  −312.6632  −114.0780  1224.81  0.017%  
283.6261  801.1597  233.7018 
Target measurement experiment. (a) 1D long target with two feature points. (b) First measurement position of the cameras.
Reconstruction of the target at three different positions.
To verify the feasibility of the proposed method, a standard flat part with the size of 600 mm × 800 mm is measured in the laboratory. Before the experiments, the standard flat part was measured using a threecoordinate measuring machine (Zeiss Prismo Navigator) to obtain the measuring reference data. By comparing the reconstruction results with the actual value, the construction accuracy of the global measurement system is validated. The experimental system is shown in Figure
The calibration results of binocular cameras.
Left camera  Right camera  

Intrinsic matrix 


Transformation matrix 

The calibration results of global system.

 

Transformation matrix 


Reconstruction of the target at different positions.
In the experiment, the measurement FOV is divided into four parts based on the size of the plate. Next, the four parts of the board are measured by the respective cameras at different positions. Images of the laser stripes, which are projected on the board, are captured by the cameras. The centres of laser stripes are extracted and matched based on image processing method in Section
The results of feature extraction and transformation in different positions.
Position  Results of feature extraction  Transformation matrix 


Left camera  Right camera  
1 



2 



3 



4 



Reconstruction results of the large plate in the global coordinate system.
To verify the validity of the system, a curved composite part is measured. The part is shown in Figure
Experimental results of a curved composite part in the global coordinate system. (a) The measured curved composite part. (b) Reconstruction results.
In this paper, a global measurement method that is based on a multiple FOV combination was proposed for measuring the 3D surface of a largescale component. Compared with existing methods, the proposed method has the advantage of high efficiency and no requirement for pasting a target mark on the component. As the cameras with the transfer mark can be placed at any position in the measurement range of theodolites, the measurement system is more flexible for largescale component measurement. The experimental results in the laboratory indicate that a maximum accuracy of the measurement system of 0.103% can be attained when the length of the 1D target is approximately 1.225 m. For the large board measurement experiment, the reconstruction accuracy is less than 0.14%. Thus, the proposed method is practicable and suitable for measuring the 3D surface of a largescale component in an industrial site. The measurement system can be employed in the assembly process of largescale industry components. Additional research is suggested to improve the global measurement accuracy by optimizing the calibration process and improving the calibration precision.
The authors declare that there are no conflicts of interest regarding the publication of this paper.
This paper is supported by the National Basic Research Program of China 973 Project (Grant no. 2014CB046504), the National Science Foundation for Outstanding Young Scholars of China (no. 51622501), the National Natural Science Foundation of China (Grant no. 51375075), and the Science Fund for Creative Research Groups (Grant no. 51621064).