Documente Academic
Documente Profesional
Documente Cultură
Xiao-kun Hu
School of Computer Science & Technology
Yantai University
Yantai Shandong , China
AbstractAccording to the characters that two cameras having transfer calibration method and single-pole double-board
little public visual in vehicles four-wheel alignment based on 3D calibration to get the extrinsic parameters of R21 and T21.
vision, this paper proposes two new methods that can calibrate
the extrinsic parameters of two cameras. These two methods have A. Traditional Calibration Method
their advantages respectively; they all can get the extrinsic
parameters of two cameras in no-public visual condition. The target board is marked as M, the graph of the target
board is designed according reference [3], and it is put within
Keywords-Vehicles four-wheel alignment; 3D vision; the public visual field of C1 and C2, as shown in Figure 1. R1,
Calibration of two cameras T1, R2, T2, respectively express the rotation and translation
matrix of C1 and C2 s coordinate system relative to M
I. INTRODUCTION coordinate system respectively, which have been calibrated
according the internal parameters of the two cameras,
Calibration method of two cameras in vehicles four-wheel respectively.
alignment based on 3D vision is a new advanced measurement.
Compared with the traditional four-wheel-positioning
technology, it has many advantages such as non-contact, real-
time, easy operation, high accuracy and low maintenance cost. C2
At present, there are several foreign four-wheel alignment C1
measurement techniques have adopted these technology, such
as V3D four-wheel aligners of U.S. JohnBean company[1]. The
main factor of these aligners, which can affect the measurement
accuracy, is the cameras calibration, and the calibration of M
extrinsic parameters of two cameras affects the model four-
wheel alignment, it is the key factor, which decides whether the
aligner is useful. In order to improve the the accuracy of the
camera, the cameras which have small Distortions are often be
adopted, as a result, the pubic visual of the camera is very
narrow. Under the narrow public visual, the traditional Figure 1. A sketch of traditional calibration method
calibration method for extrinsic parameters of two cameras will
Then the relation of R21 and T21 can be expressed as
be affected, the accuracy can not be guaranteed. This paper
following [4]:
proposes two new calibration methods that can overcome these
shortcomings and improve the effect of two-camera calibration. R21= (R1R2-1) , T21= T1 R1R2-1T2 (1)
II. CALIBRATION METHODS OF TWO CAMERAS B. Third Camera Transfer Calibration Method
EXTRINSIC PARAMETERS If the two cameras has no public visual, the two cameras
Calibration method for extrinsic parameters of two camera cant take a photo simultaneously for the target board, we can
is used to determiner the relative position of two camera[2], use the third camera to get it. The third camera is marked C3,
that is to get the rotation matrix and translation vectors which its position is shown as Figure 2. The target board M is placed
can unify the two cameras coordinate systems. The two within the public visual field of C1 and C3, and the public visual
cameras are marked as C1, C2, the extrinsic parameter R21 field of C2 and C3, one after another. R3a, T3a, R3b, T3b
marks the rotation matrix from camera C2s coordinate system respectively express the rotation and translation matrix of C1
to camera C1s coordinate system, and the extrinsic parameters and C2s coordinate system relative to M coordinate system,
T21 express the translation vectors from camera C2s coordinate which have been calibrated according the internal parameters
system to camera C1s coordinate system. In the following, we of the two cameras, respectively. By(1), the transform relations
will use the traditional calibration method, three cameras are shown as follows:
M1
C3 M2
415
two cameras; extrinsic parameter R21 mark the rotating matrix limited greatly, the angle between the target board and the
which is from C2s coordinate system to camera C1s camera is greater, then the error of identifying the center dot is
coordinate system, extrinsic parameters T21 mark translation added, as a result it affect the accuracy of the two camera
vectors which is from camera C2s coordinate system to camera calibration. This comparison shows that we can choose the best
C1s coordinate system. calibration method from above two methods, according the
cameras quality, the calibration accuracy of single-camera,
C1A = R1 M1A + T1 (12) accuracy of identifying the center dot in the target board.
C2B = R2 M2B + T2 (13)
IV. CONCLUSION
C1B =R21 C2B + T21 (14)
According to the characters of 3D vision four-wheel aligner
According to (12), (10), the coordinate of M1B in camera for vehicle, this paper studies the calibration methods of two
C1s coordinate system, marked as C1B, can be expressed as cameras in vehicles four-wheel aligner based on 3D vision.
follows: Through the comparison to the traditional calibration method,
C1B = R1 M1B + T1 (15) we find that though in theory the accuracy of the two
calibration methods is lower than the traditional calibration
By (13), (15) and (9), we can get the follows: method, in product reach and development there are better
prospects.
C1B = R1 RM21R2-1 C2B + (R1 TM21 +T1 R1 RM21R2-1 T2)
Then REFERENCES
-1
R21 =R1 RM21R2 , T21 = R1 TM21 +T1 R1 RM21R2-1 T2 (16) [1] Gao li-tao,Tu ya-qing,Wang de-chao. A novel method of wheel
alignment parameters measurement based on computer vision[J].
By (11) and (16), the extrinsic parameters R21, T21 can be Automation & Instrumentation, 2009,143(3):74-77.
calculated. [2] Zhang hui, Zhang li-yan, Cheng jiang, and Zhao zhuan-ping. Field
Calibration of Binocular Stereo System Based on Planar Template and
Free Snapping, Journal, Acta Aeronautica Et Astronautica Sinica, 2007,
III. COMPARING OF DIFFERENT METHODS 28(3): 695-701.
In these three methods, the traditional method has few [3] Hu yong, Wang cong-jun, and Huang shu-huai. Calibration Automation
steps, few errors, it is the most accurate methods; but it has of the Binocular Vision Sensor, Journal, Journal of Wu han University
of Technology, 2005, 27(10): 67-69.
high requirement for the place of the cameras in the four-wheel
[4] Zhang guang-jun. Machine Vision. Beijng: Science Press, 2005.
aligner, if we want use this method to get the public visual, we
need use a long beam, as a result the practical value is reduced
greatly. However, the two new methods have their advantages
respectively, as shown in Table 1.
416