JPS61277012A - Method and apparatus for correcting position and posture of camera - Google Patents

Abstract

PURPOSE:To simply obtain the position and posture parameter of a camera at a high speed with high accuracy, by analytically calculating the solution of simultaneous equations consisting of a camera position and posture parameter and an error parameter and performing the correction reducing the error parameter by an analytical process. CONSTITUTION:A mark discrimination processing part 4 obtain the center coordinates of a reference mark from the image data stored in an image input part 3. An approximate value calculation processing part 5 performs a means for analytically calculating a camera position and posture parameter and an error parameter by the matrix operation based on a method of least square and a correction processing part 6 performs a correction means for analytically calculating the camera position and posture parameter by the matrix operation based on a method of least square after performing the preparatory correction of the camera position and posture parameter. When the error parameter is larger than a preliminarily indicated value, the correction means is again performed to make it possible to correct the parameter. When the error parameter became smaller than the indicated value, a processing result is outputted from a result output part.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention provides a method for easily and highly accurately calibrating the position and orientation parameters of a camera, which are necessary when manually obtaining three-dimensional information of an object using a camera. This is related to the device.

[Conventional technology]

Camera position and orientation parameters include the camera lens center position and camera direction, and if the optical axis point and optical axis length are known in advance, the number of unknown parameters is six.

Note that the optical axis point here is the intersection of the perpendicular line from the center of the camera lens to the image plane of the camera and the image plane, and the optical axis length is the length of the perpendicular line. The method of determining the parameters is to observe multiple points (reference markons) on an object whose positions are known in advance with a camera, find the error between that image and the image obtained by the camera with the assumed parameters, and calculate the difference between the two. Conventionally, methods have been proposed for selecting parameter values such that .

The following document 1 proposes a method for finding parameters that simultaneously reduce two error functions that include unknowns.

The procedure is to first give initial values to the unknown parameters, and then
The parameters are corrected little by little so that the two error functions become smaller (depending on the situation).If the initial values are incorrectly given, convergence may not occur or the number of corrections may be extremely large.
There is a problem with practicality. Also, there are many reference marks.

Attempting to improve accuracy has the disadvantage that processing time for correction becomes extremely long.

In Document 2 below, three simultaneous equations including parameters are derived from images of three known straight lines. Since the variables in this equation have constraint conditions between variables, analytical methods cannot be applied, and as in the above-mentioned document 1, it is necessary to find an approximate solution while gradually modifying the parameters. If the initial values are incorrectly set, convergence may not occur or the number of corrections may be increased.

Document 3 below proposes a method of finding a solution using a hill-climbing method in which initial values of parameters are determined using a coordinate system transformation matrix, and then each parameter is gradually changed and calculations are repeated until the error becomes small. In this method, reference 1
Similar to Document 2, it is thought that the processing time is required due to repeated calculations.

Further, as far as the transformation matrices disclosed in this method are concerned, no consideration is given to satisfying the orthogonality condition that the matrices should have.

For this reason, the constraint conditions between matrix elements must be fully considered in the hill-climbing operation that is the correction process (℃ will not be achieved. Therefore, as in Reference 1.2, the amount of processing will be large and the processing time will be long. becomes.

Reference Ishii et al., Institute of Electronics and Communication Engineers Edane SC-84-9
3D Position and Posture Sensors and their Application to Robots” Reference 2) Echigo et al., Information Processing Society of Japan Computer Vision Study Group 34-4 (January 1985, 248) “3D Position Detection Using Striped Bakoon Stereo” Reference 3) Kuno et al., Information Processing Society of Japan Computer Vision Study Group 35-1 (March 12, 1985) [Position and orientation measurement of polyhedrons using binary images*] [Problems to be solved by the invention] As described above, 5K, conventional , the proposed method gradually reduces the error through repeated calculations.The value tends to change depending on how the initial value is given and how the reference mark is given, so it cannot necessarily be used for general purpose. In addition, there is a drawback that the amount of processing is large and the processing time is long.

The present invention analytically solves simultaneous equations consisting of camera position/orientation parameters and error parameters, and then performs a simple correction to reduce the error parameters using an analytical procedure. An object of the present invention is to provide a camera position/orientation calibration method and apparatus for obtaining a camera position/orientation variation.

Means for Solving Problem C] The camera position and orientation calibration method according to the present invention K uses an imaging camera to observe six or more specific points whose spatial coordinate system is known, and calculates the image of the obtained specific point. Using the position, a transformation matrix having orthogonality between the spatial coordinate system and the camera coordinate system and an error parameter are analytically calculated by matrix operations based on the method of least squares to obtain an approximate case for the camera position and orientation parameters, and then Then, using the error parameters, perform matrix calculations based on the least squares method to calculate the transformation matrix and the error parameters (parameters are calculated analytically to obtain the corrected values for the camera position and orientation). The camera position and orientation parameters are obtained by making corrections so that the position and orientation are correct.

Further, the camera position and orientation calibration device according to the present invention includes an image input unit, a mark identification processing unit that identifies a reference mark, and a matrix calculation based on the least squares method to calculate camera position and orientation parameters and error parameters from the output of the image input unit. It is composed of a correction processing section consisting of a hardened matrix calculation module that performs analytical calculations.

[Effect]

In the camera position and orientation correction iVc of this invention, the first step is to express the degree error of the approximate value of the camera position and orientation parameter by the value of the error parameter found by orthogonalizing the transformation matrix, and the second step is to evaluate the accuracy. As a correction for increasing the amount of error, matrix calculation is performed directly on the error correction amount using an analytical procedure to obtain camera position and orientation parameters.

Further, the camera position and orientation correction device of the present invention is as follows.

The approximate value calculation processing unit calculates the camera position/orientation parameters and error parameters using matrix calculations based on the method of least squares.
Cg is extracted, and the correction processing unit analytically calculates camera position and orientation parameters and error parameters by matrix calculation based on the method of least squares.

〔Example〕

An embodiment of the present invention will be explained below by dividing into a first step and a second step.

(The center of the first step camera lens is the object coordinate system (X + 7 + z)
It is assumed that the operation is located at (C+ a e31 c3 ). The relationship between the camera coordinate system (XB g y@ + Z@) and the object coordinate system (X*y*z) is generally expressed by the following equation. Note that the coordinate system is a right-handed system as shown in FIG. 2, and the yo axis is in the direction of the camera optical axis. Also,
1 is a mark board! +7+Z is the spatial coordinate, α,
β, ψ are azimuth angles, 2 is camera, Xe* y@ *
z6 is its axis.

However, M is a matrix element of CM, which is a transformation matrix from the object coordinate system to the camera coordinate system, as shown in the following equation.

M ” (ak4) (k, l=l, 2.3) =
...(21) On the other hand, the distance when a perpendicular line is drawn from the camera center to the camera screen is 11! (The optical axis length is L, and the coordinates of the intersection of the perpendicular line and the screen (optical axis point) are (5olo). ,
In addition, the scale factor (reduction ratio) between the image on the screen and the image in the image memory is set by K in one direction (horizontal direction) of the image.
Let η1 be an engineering ηj in the j direction (vertical direction). The coordinates of the reference mark in the camera coordinate system and its image (i, j) have the following relationship.

In subsequent calculations, the transformation matrix) U is orthogonal, ie.

ΣaL1=l(k=1,2.3)...(5)Σak
J arml = 0 (6) (k
In order to guarantee this, we will renew the vows in formulas (3) and (4) as follows.

δ and (ε2.ε,) are error parameters representing errors in the optical axis length and optical axis point coordinates. By substituting equations (1) and (2) into equation (7), the following equation is obtained.

a'1. lx +a',, i y+a'H-i z
+f, x +f, y+B'1z-B;el-B'
zc=-B'3c3=i...' (8 squared, BJ--Lall-Jalj-η1 1oau-'71
εr aH(J=L2+3)ft-B/A1η
, (1=11213n azJ” atj /A1(ll
"1 + 2+ 3) Similarly, a'21 j x + a'B j y + a'Hj
z +D', x+D'27+I)', zD'r l
D's ea D; e3 :j' - (10) However, using equations (8) and (1o), the coordinates of multiple (n) reference marks in the object coordinate system are expressed as (Xls Fl * Z+
) + ...+(XB +7IIIZ11), and the corresponding images are (i, * j+ )+...
・, (in, Jll), then GN=W ・・
・・・・・・・・・02) However, G is a multidimensional parameter vector, and G= (an+ a; b m's +
B'+ + B"! * B; * D7 * p
;, l1lr31 u; +u'! ) --・a3
) Image point vector W = (ill 121・+ 11 II l!'・, j
...05) Reference mark matrix where m = (1,...,1) ... (17) Also, O□ is an O matrix with m rows and n columns, and 1.1' are 3 rows and 3 columns and n rows and n columns, respectively. Next, calculate G using the least squares method using equation (12) using the unit matrix of columns. That is, G that minimizes the error EE=Σ(ΣG..NPk ``mk )'IIkn is determined.The result is given by the following equation.

G=W·N” (NNT door (18) where NT is the transposed matrix of the reference mark matrix N.

The following relationship is established by Sako and No. (9) 2.

a'Hel +a'B el +a'Hc@
= 1 ・ ・(19) This and the (14th)
) from the camera lens center coordinates (CIHcl+Ca)
is given by the following equation.

Also, the transformation matrix M is orthogonal, that is, the (
5) By transforming using the conditions of Equation (6) and Equation (9), the error parameters ε quantity and εj can be calculated.

ε1 ” −AS (&:@, fl+a'lJ'l+
ajHB'*)-io-...i21)83=
・A: (a'i, DH+ &'**D* +jL'
,,D; n j , ...(22) Here, the th (
Considering Equations 5) and Equations (6), it is possible to use the value calculated by the following equation, where A: is taken.

A:=(a',・+a4."+a',,")・'・
(23) Next, find the error parameter δ. No. 9
) from the formula (L+δ)” 1117 (l=1 + 2+
3) and transform it using equations (5) and (6), the following equation is obtained.

(L+15)”/η1”=A,” (B,”+B’,
"+B',")+(1,+g,)"+2 (+6
+j ) (B?a'21+B'2 ass +B
's &'*s)Ax...(24) Similarly (L+δ') AJ"= kX (07"+D;"+
D;”) + (jo + g J) ”+ 2 (
jg +54 ) (D7 a'z+ +l)'=a'
H-+-o, 4s) A x...(25) Next, from equation (9), a'+t=(B'z771-η+ (io+g,)a
-, l/(L+δ), <l=1.2.3), so a'l 1=(-Bko(i@+gt)a'u)/
l(L+δ line/'?t l-=(1, %, =t o+cosine-j,+gj)a'y)/
((L+δ)/ηJ) −·−(27) Here, the transformation matrix M(akj) obtained above is guaranteed to be orthogonal. The apparent optical axis length is the 24th
) or (25) 2, and the apparent optical axis point is (i0+ε1.j0+εj) calculated by equations (20 and 223), then x*'
;"The rotation angles α, β, and ψ of the camera around ila are α" I
I t n-' (-1% z t a n 9)/I
k'H) 'β= t a n-'(a's+
/a'ss)cp=jan-"(-a'
It is determined from Ba inβ/a's I)K.

If the camera and mark are not tilted, the mark configuration or lighting conditions are optimally adjusted, and the mark depth distribution covers a wide range, the error parameter can be negligibly small. Highly sexual. At this time, the obtained 1, +ε8. j0+εj,L
The value of +δ can be used as an internal parameter of ill Joll.

However, the positional relationship between the camera and the mark and the lighting conditions usually change. Furthermore, errors in reading the center coordinates of marks also occur in actual measurements. Therefore, the calculated apparent optical axis point (i6+ε1.jo+ε) is the optical axis length of the island (L
+δ) & is different from the true value, No, and joLL, which causes an error in the position and orientation of the sword blade. As shown in the solid line, this error often becomes extremely thick (sometimes).To perform highly accurate camera position and orientation parameter calibration, it is necessary to make sure that the apparent optical axis point and apparent optical axis length match the true values.
That is, the error parameter ε1+$3. the second where δ is O
Steps are required.

(Second Step) Generally, it is not possible to separate ε and εj from the apparent optical axis point in °C using the above method, but it is possible to separate the optical axis point (io) separately at °C using ordinary means for measuring lens characteristics.

jo) can be obtained. Regarding the optical axis length LVc, when the reference mark arrangement surface is placed approximately at right angles to the optical axis of the camera and the number of reference marks is large, K can be obtained with high accuracy by the parameter calibration in the first step according to the present invention. Since we are busy, we calibrate the relationship between the lens extension position and L in advance. In this way, by determining the camera-specific parameters (Lo, jo) and L in advance, it is possible to separate the values of C6, C4, and δ.

Next, C3, εj, δ=0, and the transformation ma) IJ
The parameters are corrected so that the lines are orthogonal.

First, find the angle θ between the intersection line of the image plane and the horizontal plane (xy plane) of the object coordinate system and the l axis of the image plane. As θ proof, the following equation is derived.

01g=coa-1(bl, a'o+b goose I
L'+*) However, bt=±(1+(a%+/i;*)
"l-'x ・"--"(29)b2=-a'
ff1l bl / &'xs The above intersection line and ""jo
) and the straight line connecting (io+εl+jll+εj) are given by the following equation including θ in equation C29).

θ= −−′=(e o a−′ (ε+1lo)−0
M + −・”(30)1εJ However, l:=ε−1εj2 The camera direction correction amount is given by using θ.

Also, C1+ C1+ am is expressed as ((!II
et, es) = τ town (cr+cz*cs)...(3
2].

K, using the direction corrected by equation (31), a□
Find □ in r&B*1. Calculate C41°a'211 &'++s'k1h using this value and CI+CR#eM corrected according to No. (32) 2 using the linear equation.

a'zs=au/Axe a'a=az*/A*+ a
'Is=atm/Av '°匈) However, A*:! Lh
e++auC*+ILnem...=<A4) This is the preliminary correction.

Now, if a'2+ 1 &'22 * a'ts are known numbers, then equation (8) can be reduced to the following equation in which the unknown coefficient is included only on the left side.

B', x +B', y +B; z -Bτat-
B2c'z-B'* C3:=i-a+1x-a'B
iy-a', 31z・c't5) Similarly, the following equation is obtained instead of the equation (lO). ,D',
X+D'2 y+D'3 z -D'; cl -n;
C3-I)', c.

”j-&'ajx-a'ujY-a'ujz...(
The coordinates of multiple (nivA) reference marks in the size object coordinate system are (X++ 7i l Z+ L...'...
* CXy, a Val Zm ), and the corresponding images are (Ii++J+L..., (ill, j, ), then the following relational expression is obtained using equations (35) and C36). It will be done.

P・Q=W...1 old...(first proviso, V=(I J)-1"A KtKg-f"A'
KsKJ --... (40) (IJ) = (""""
) A=(an l'!!, a'2S)+ (=(Δ)
jl...jllll. After deriving equation (18), P
is obtained from the following formula % formula % (41) k, from P B'+-B't l n'sI D; +D'2 sD
Z #u'llu'! is required. Using this, the first (
26), (27), and (28), the corrected lateral angles α, β, and ψ are found. The obtained transformation matrix is guaranteed to be orthogonal. Using these values, we can calculate C1at l as by the (20th equation), and also calculate the 0υ, (2
2) Find 61+g31δ from equation (25). In the processing up to this point, in the case of normal reference mark conditions, ε籏,
εj and δ become sufficiently small. However, when the condition of the fiducial marks is poor, such as when there are few fiducial marks (and they are clustered in one place on the image), etc.

ε turbidity, εj, and δ may reach a level that cannot be ignored.

In that case, the VC re-executes the processes of equations (29) to (41). In this way, the camera position and posture parameters can be known with high precision.

FIG. 1 shows a flowchart of processing showing an embodiment of the present invention. Note that (1) to (14) indicate each step.

After inputting the coordinates of the fiducial mark in step (1), step (2), (15), (16) 2, and N.

Set W, step (3) busy, G by step (18) 2, i.e. a-t s &'s snow, a'*s* B; *B
;s Bs mD'>mD-*5D-s +u't
+ u.

Step (4) k, C20) formula, y-th (
25) el+ORr C3, gl, 61. δ and find δj. Next, in step (5) &t, equation (a).

According to the 27th formula, Jl'll r a'l!+ ass I
a'st l a'zs l a'l! Then, in step (6) K, the azimuth angles α, β, and ψ are determined.

The correction routine starts from step (7).

εl* 'If J+δ is ignored, step (8)
I'm busy. In step (8), the 3rd expression, the 322nd
After correcting the azimuth angle and correcting 81+el+03 using 2, in step (9), a'ts document a'*t l a'zs is determined by equations (33rd and 04th), and step 00 ) smell 1, by equations (39) and (4o), set Q and V, and in step (11) 6f, by step (37) 2, P, that is, Bτ # n #
, I B's l D; # DS *
D: Calculate *u'+*u'1. In step (12),! (20), (2υ, (2)n.

(24), (25) Find 6Ilεjl δ using 2, and return to step (7). YES in step (7)
In the case of S, that is, if J, gj, and δ are negligibly small, the process moves to step (13) of the confirmation routine. In step (13), camera azimuths α, β, ψ and camera position 01+e! * A reference mark image is calculated using C3, and in step (14), a deviation between the coordinates of the mark actual riii image and the calculated mark image coordinates is calculated.

FIG. 3 is a block diagram showing an embodiment of the camera position and orientation calibration device of the present invention.

In this figure, 1 is a mark board on which reference marks are written, and is placed in front of the camera 2. The image obtained by the camera 2 is stored in the image input section 3VC.

This image input section 3Vc has a frame memory and a camera drive circuit. Reference numeral 4 denotes a mark identification processing section, which obtains the center coordinates of the reference mark from the image data stored in the image input section 3. In the experiment, a white circle was used as a mark, so the mark identification process first performed labeling, calculated the size and circularity of each labeling area, extracted a valid labeling area, and then assigned a mark number.

Reference numeral 5 denotes an approximate value calculation processing unit, which executes means for analytically determining camera position/orientation parameters and error parameters by matrix calculation based on the method of least squares.

Reference numeral 6 denotes a correction processing unit, which executes a correction means that performs preliminary correction of camera position and orientation parameters and then analytically calculates the camera position and orientation parameters by matrix calculation based on the least squares method VC. If the error parameter is larger than a predetermined value, the parameter can be corrected by executing this correction means again. When the error parameter becomes smaller than the specified value, the result output unit 7 outputs the processing result to the meeting.

Next, the results of the experiment will be explained.

For the image input s3 and the mark identification processing section 4, a Toshiba rear general-purpose image processing device TO8PIX-11 is used, and for the camera 2,
A Sony CCD camera was used. Further, the approximate value calculation processing section 5 and the correction processing s6 are registered as subroutines in the library of the VAXu/'ys (B 2 computer).

Furthermore, the spatial coordinates of the marks are registered in advance in the data file, and are referenced as necessary during processing.

CFr calculation m simulation experiment] η, = 0.02. η7 = 0.02, L = 55
mm, α=-〇, 11111 radians, β=0.0.
9) = 0.04 radian * c,: -100,0
mm+cz=-800,0mmacs:=400.
When the camera position and orientation were determined according to the flowchart described above using 64 reference marks using a camera with a parameter of 0 mm, in the first step, the Ct knee was 108.5 pixels,
εj::147i! ii element, δ=3.8mm+ α=-
0,0722 radian, β ~ 0.001 radian 7.9+
= -0,357 radius 7, cH = -109,1mm+
c, knee s 60.6 mm + cl = 434.1 mm, which is a fairly large error from the actual value.

Due to one correction in the second step correction procedure, εl=
−0.9 pixels, g 5 ==−1.8 m g +
δ=0.4 m m +α:=-0.1.118 Radius 77. cp=-0,40225 diane, β=0.000
2 radians + e1 = 100.3 mm, C2 knee 799.83 mm* cs = 401.47 mm, and further correction of K and 2@ eyes makes εI = 0.2 pixels, ε
1 = -1,8uni element, δ=-0.47mm
eα=-0,1112 radians 77,9)=-0,401 radians, β=0.0007 radians*c+: -1
00,4mm+c2=-800,2mm+e,=4
01.6mm, the azimuth angle is up to 0.001 radian, and the position accuracy is up to 0.4%.

The average mark shift in the confirmation procedure was 0.2 pixels.

[Experiment example using real images]

In the example of 12 reference marks, in the first step, t, = -
8,0 pixels, 17:-35 pixels, δ=-25,2mm
However, after performing the correction procedure four times, ε, = Knee Q, 37 pixels, ε.

= -0,021 pixels, δ = 0.0003 mm.

In addition, the mark in the confirmation procedure was 800 pixels on average in the first step, but was able to be reduced to less than 1 pixel after four corrections. In another example, for the number of reference marks 14, in the first step, 6=3Q, 3-element, εj = 78.
Ojji element, a=-0.14mm, tJ''-1
Corrected Lucia 1 @ “t” 81 = 1.3 j
ji elementary-gl::l 7.1 pixels, δ:O, +12m
m, 2rj: A-th routine, g+ = -0,15
1mj element, '1''-0,017ul element, a = 0
The mark deviation in the confirmation routine was 0.9 pixels in average Jal element in the first step due to the first exposure.

Next, we will discuss processing speed.

In the case of 8 marks using a commercially available general-purpose image processing device and a mini combiator, the process from inputting one pixel to identifying marks takes 10 steps. All processing was completed in a total of 6 seconds, including about 5 seconds including disk overhead, and K1 seconds for determining the final busy parameters and displaying the results after completing the confirmation procedure.

The time required per mark is only 60 m5ec. If you try to obtain the accuracy of this invention using the conventional method, it will take several tens of minutes, and there is a strong possibility that it will take more than an hour depending on the busy schedule. Therefore, according to the method of the present invention, the processing time is short and real-time measurement is possible.

In the above embodiment, a two-dimensional television camera was assumed to be used as the image pickup camera for mark-taking input, but it is also possible to use an image pickup camera that can obtain the image position of only one mark at a time, such as a position sense device manufactured by Hamamatsu Photonics. The present invention is also applicable to this method because the mark image position of each mark can be sequentially changed by time sharing. Also, about the one-dimensional image senna! Since a two-dimensional image of the mark image can be input by scanning the mark image plane with a dimensional sensor, the present invention can also be applied. Also an ultrasound camera.

One-dimensional or two-dimensional image sensors using other information transmission media such as XIR cameras are also applicable.

Further, as the reference mark, a feature point of the object, such as a specific point on a vertex or edge, or a three-dimensionally distributed, easy-to-identify set of marks with a specially designed arrangement is used.

〔Effect of the invention〕

As explained above, the camera position and orientation calibration method of the present invention analytically obtains an approximate solution for the camera position and orientation and three types of error parameters, totaling nine points, and then corrects the error using an analytical procedure. 1 because it is a method that minimizes the parameters
It has the following advantages.

(1), a lotus-like solution can be easily obtained. Furthermore, since the error parameters are calculated, the degree of approximation can be determined.

(2) The correction procedure is based on an analytical method.

It is easy to see, direct, simple, and highly accurate.

(3) In this invention, since the orthogonality of the coordinate system transformation matrix is maintained, the obtained solution has good reliability and stability, and can be used as a general-purpose method.

(4) Even if there are many reference mark points, processing can be performed in several seconds even including input/output overhead, and the maximum error in camera position is 0.4%.

High precision measurement of azimuth angle of about 0.001 radian is possible. Therefore, it is useful as a real-time, highly accurate camera position and orientation calibration method.

Furthermore, since the camera position and orientation calibration device of the present invention uses a hardened matrix calculation module to configure the approximate value calculation processing section and the correction processing section, the calculation speed can be increased, and the number of reference marks can be increased. This can be dealt with by replacing the matrix calculation module depending on the situation, which has the advantage that the device is compact and easy to handle.

[Brief explanation of drawings]

Fig. 1 is a flowchart showing the processing procedure of the camera position and orientation calibration method of the present invention, Fig. 2 is an explanatory diagram showing the arrangement of the spatial coordinate system and the camera coordinate system, and Fig. 3 is the camera position and orientation calibration device of the present invention. FIG. 2 is a block diagram showing one embodiment of the present invention. In the figure, 1 is a mark board, 2 is a camera, 3 is an image input unit, 4
5 is a mark identification processing section, 5 is an approximate value calculation processing section, 6 is a correction processing section, and 7 is a result output section. Figure 1 Figure 2 c

Claims (2)

[Claims] (1) Observe six or more specific points whose positions in the spatial coordinate system are known simultaneously or in chronological order using an imaging camera, and use the positions of the images of the obtained specific points to establish the spatial coordinate system. An approximate value of the camera position/orientation parameter is obtained by analytically calculating a transformation matrix having orthogonality between camera coordinate systems and an error parameter by matrix calculation based on the least squares method, and then using the error parameter. By analytically calculating the transformation matrix and error parameters using matrix calculations based on the least squares method to obtain correction values for camera position and orientation parameters, and repeating the correction so that the error parameters become less than or equal to a specified value. A camera position and orientation calibration method characterized by obtaining camera position and orientation parameters. (2) An image input section that inputs the image of the mark, a mark identification processing section that uses the output of this image input section as input and identifies the reference mark, and a camera position/orientation that uses the output of this mark identification processing section as input. An approximate value calculation processing section consisting of a hardened matrix calculation module that analytically calculates parameters and error parameters by matrix calculation based on the least squares method, and the output of this approximate value calculation processing section is input, and the least squares method is used. What is claimed is: 1. A camera position/orientation calibration device comprising: a correction processing unit comprising a hardened matrix calculation module that executes correction for analytically calculating camera position/orientation parameters and error parameters through matrix calculations based on the above-mentioned matrix calculations.