JP2002340923A - Method and program executing the method for optimized exclusive binarization correlation measurement - Google Patents

Abstract

PROBLEM TO BE SOLVED: To execute an exclusive binarization correlation measurement in the method and program executing the method for optimized exclusive binarization correlation measurement, by setting an allowable error span δfor judging whether the magnitudes of a reference signal and a measured signal are equal or not, at an optimum value. SOLUTION: Pixel values of each image of the reference signal and the measured signal are compared and exclusively binarized according to whether or not the values of the reference signal and the measured signal are equal. For the basis of judging whether or not the values of the reference signal and the measured signal are equal, the allowable error span is determined. According to the relation among the reference signal, the measured signal and the allowable error span, the exclusive binarization is applied to the measured signal, and the correlation of the reference signal and the measured signal is measured.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optimized exclusive binary correlation measurement method, an optimized exclusive binary measurement apparatus, and a program for executing the method.
In particular, it measures the correlation between a measurement signal that is a moving measurement target signal and a reference signal. It enables observation of the moving speed, moving direction, and the like of an object moving at high speed, and can also be used for observing fluctuations of the earth's atmosphere.

[0002] The fluttering of a star is caused by the light from the star being disturbed in the wavefront (phase) by the fluctuation of the refractive index in the terrestrial atmosphere. By observing a random pattern generated by the blinking of the star on the surface of the earth, the wind direction in the fluctuation layer,
Wind speed, altitude of fluctuation layer, intensity of fluctuation, etc. can be measured. Since the random pattern moves at high speed, high-speed image processing is required.

[0003]

2. Description of the Related Art In order to observe such a random pattern, a part of an image signal of a two-dimensional image of a random pattern is acquired as a reference signal at a high speed, and the two-dimensional image of the random pattern is used as a measurement signal to measure the two signals. The random pattern is measured by taking the correlation.

FIG. 12A shows a measurement signal, and FIG.
(B) represents a reference signal. A is the measurement signal to be measured, B
Is a reference signal.

[0005] The correlation between the measurement signal A and the reference signal B is obtained, and it is determined whether or not both are the same signal. Measurement signal A
And the reference signal B can be represented by A = <A> + α and B = <B> + β.

[0006]

(Equation 1)

[0007]

(Equation 2)

Where <A> and <B> represent the average values of the respective signal intensities. α represents the variation of the measurement signal A from the average value, and β represents the variation of the reference signal B from the average value.

The correlation C between the measurement signal A and the reference signal B is calculated by the following equation.

C = <A × B> = <(<A> + α>) × (<B> + β)> = <<<<A> × <B> + α <B> + β <A> + αβ> = <A> × <B> + <α> × <B> + <β> × <A> +
<Αβ> Here, <> represents the average of A and B. If A and B fluctuate randomly, <α> = 0, <β> = 0
It is.

When the measurement signal A and the reference signal B are not the same (A ≠ B) <αβ> = 0 When the measurement signal A and the reference signal B are the same (when A = B) <αβ> = σ ² . σ ² is the variance of the measurement signal A and the reference signal.

Therefore, C = <A> ² (when A ≠ B) C = <A> ² + σ ² (when A = B)

For example, it is assumed that a moving speed of a pattern which fluctuates randomly in the apparatus shown in FIG. 1A without changing the pattern shape is measured (the details of FIG. 1A will be described later).

Array element photodetector 5 and single element photodetector 6
Acquires a plurality (i = 1, 2,..., N) of data pairs in synchronization.

The measurement signal A is a signal A in which the random pattern 1 is observed at any one pixel of the array element photodetector 5.
It is a set of _i . The reference signal B is a signal observed by the single-element photodetector 6 via the pinhole 7, and the signal B acquired with a delay time τ from the data acquisition time of the paired array element photodetector 5 It is a set of _i .

In the apparatus shown in FIG. 1A, a general correlation C between the reference signal and the measurement signal is represented by the above equation.

If the random pattern 1 moves without changing its shape based on the result, the direction and speed at which the random pattern 1 moves can be obtained by calculating the correlation C.

In contrast to the general calculation of the correlation C, before calculating the correlation, the i-th element of the measurement signal A and the reference signal B
By comparing the i-th element of, if they are different, the values of both elements are set to zero, and then calculating the exclusive correlation C ′ by C ′ = <A × B>, C ′ = 0 (A ≠ B) C ′ = <A> ² + σ ² (when A = B) By reducing the value of the uncorrelated portion as in (20), the peak can be made clearer. As a result, the correlation can be calculated with a small number of signal elements.

[0019]

As will be described later, when the measurement signal, the reference signal, or both include measurement noise, the signal of A = B is replaced with the signal of A = B.
= B + ε (ε is measurement noise), and if applied as it is, the peak height shown by the above equation (20) will be greatly reduced. Therefore, when it is in the range of A = B ± δ, it is necessary to introduce an allowable error width δ that assumes that A = B. How to set the value of the allowable error width δ is very important as described below.

If this δ is not set properly, for example,
If δ is too small, the signal originally A = B
The probability of being determined as ≠ B increases, and the peak of the correlation decreases. If δ is too large, the probability that a signal originally A ≠ B is determined to be A = B increases, the value of a portion having no correlation increases, and a peak cannot be clearly determined. In view of the above, the present invention determines an optimal allowable error width δ in consideration of the signal fluctuation width and measurement noise, optimizes the exclusive binarization, and performs correlation measurement. An object of the present invention is to provide an exclusive binary measurement device and a program for executing the method.

[0021]

SUMMARY OF THE INVENTION The present invention provides an optimized exclusive binary measurement method and an optimized exclusive binary measurement device for measuring the correlation between a reference signal and a measurement signal that is a moving measurement target signal. And the program that executes the method, compares the signal values of the reference signal and the measurement signal,
This is to binarize the measurement signal exclusively according to whether the value of the reference signal and the measurement signal are the same, and optimized as a criterion for determining whether the value of the reference signal and the measurement signal are the same. An allowable error width is provided, the measurement signal is exclusively binarized based on the relationship between the reference signal, the measurement signal, and the allowable error width, and the correlation between the reference signal and the measurement signal is measured.

The allowable error width is determined according to the measurement environment. The allowable error width is determined according to the measurement noise in the measurement environment, and exclusive binarization is performed according to the magnitude relationship between the difference between the reference signal and the measurement signal and the optimized allowable error width. .

Furthermore, when the difference between the expected value of the correlation peak and the expected value of the portion having no correlation is divided by the magnitude of the fluctuation of the expected value, the SN ratio is maximized. This is to optimize the allowable error width.

Here, the SN ratio is a value obtained by dividing the difference between the expected value of the correlation peak and the expected value of a portion having no correlation by the magnitude of the fluctuation of the expected value. It shows that it is clear.

According to the present invention, the determination as to whether or not the signal strength of the measurement signal of the two-dimensional image to be correlated and the signal strength of the reference signal are the same is performed based on the allowable error width determined so that the SN ratio is as good as possible. Since the correlation between the measurement signal and the reference signal is obtained by binarization, a correlation with a good SN ratio can be obtained. Further, the pixel value of a portion having a different signal strength is set to, for example, 0, and the pixel value of a portion having the same signal strength is set to, for example, 0.
Since the correlation is obtained after the binarization to be 1, the amount of stored data can be reduced, and the correlation processing can be performed at high speed with a simple calculation.

[0026]

FIG. 1 is a diagram showing a first embodiment of the present invention. FIG. 1A shows an observation device, and FIG.
Is an explanatory diagram of an observation method.

In FIG. 1A, reference numeral 1 denotes a light pattern to be measured, which is a random pattern. The random pattern 1 is, for example, light radiated from an object moving at high speed, or a random pattern formed by stars on the ground surface. Reference numeral 3 denotes a condensing lens that condenses light emitted from the random pattern 1. The condenser lens 3 may be a telescope. Reference numeral 4 denotes a beam splitter, which separates light emitted from the random pattern 1 into measurement light 8 and reference light 9. Reference numeral 5 denotes an array element photodetector, for example, a two-dimensional photodetector such as a CCD. Reference numeral 6 denotes a single-element photodetector for detecting light having a narrow visual field passing through the pinhole 7. The single element photodetector 6 can sample at a higher speed than the array element photodetector 5 (for example, using a photomultiplier tube). 7 is a pinhole. Reference numeral 8 denotes measurement light, which is light incident on the array element photodetector 5. Reference numeral 9 denotes reference light, which is light incident on the single-element photodetector 6.

In the configuration shown in FIG. 1A, the light emitted from the random pattern 1 is condensed by the condenser lens 3 and is incident on the array element photodetector 5. A part of the emitted light is split by the beam splitter 4, the field of view is narrowed by the pinhole 7, and is incident on the single-element photodetector 6.

The array element photodetector 5 has a shutter, and can perform exposure for a short time sufficient for the moving speed of the random pattern 1. The position of the pinhole 7 is adjusted so that the portion of the random pattern 1 that is observed by the array element at the center of the array element photodetector 5 is also observed by the single element photodetector. Array element (array element of array element photodetector 5) and single element (single element photodetector 6)
) Acquire data in synchronization. With one exposure, a frame image of an array element for each time and a set of data pairs of a single element having a different delay time τ are formed.

Referring to FIG. 1B, a method of observing the random pattern 1 by the observation device of FIG. 1A will be described. FIG. 1B shows a pair of data acquired by synchronizing the array element and the single element. In FIG. 1B,
Reference numeral 15 denotes a frame image obtained by one exposure of the array element photodetector 5, and reference numeral 16 denotes a data string of the single element photodetector acquired in synchronization with the frame image 15.

With respect to the frame image 15 shown in FIG.
The image actually projected onto the two-dimensional array until the next frame image is acquired constantly changes with time. The single-element photodetector 6 continues to record the change in light intensity at the center of the frame image during that period, and acquires a series of data strings. FIG. 1B shows the relationship between a set of data pairs and a data sequence of a single-element photodetector acquired in synchronization with the frame image 15.

Reference numeral 25 denotes an image A ₀ of the measurement light, and a part R ₀ of the random pattern 1 is projected on the array element photodetector 5.

R ₀ moves after the delay time τ has elapsed, and is detected as the reference light B ₀ by the single-element photodetector 6 via the pinhole 7.

X _vp is a vector indicating the position of the array element measured from the center array on the array element photodetector 5 (the subscript “v” indicates that the reference symbol indicates the vector quantity).

Frame image 15 of array element photodetector
Is an image with a delay time τ = 0. The data B ₀ (light intensity) of the single-element photodetector 6 at the delay time τ and the delay time 0
When the correlation with each pixel of each array of the array element photodetector 5 is obtained, it is assumed that the light intensity of the data B ₀ and the image A ₀ have the highest correlation. This means that the image A ₀ (coordinate position x _vp ) of the measurement light moves to the position of the pinhole, that is, the array element at the center of the array element photodetector 5 after τ seconds. Therefore, if the random pattern 1 moves without changing the shape, the image A ₀ of the measurement light in the frame image 15 and the data B ₀ of the reference light in the delay time τ of the data train 16 have the same value. Is recorded. Therefore,
When a large number of such data pairs are acquired and correlation is calculated,
Correlation peak position of the image A ₀ of the measuring light in the frame image 15 appears.

As described above, part of the pattern R ₀ is at the coordinate position x _vp when the array element photodetector 5 is exposed,
After the elapse of the delay time τ, it can be observed that it moves to the coordinate position _0v . Therefore, according to the observation method of FIG.
The moving speed and moving direction of the random pattern projected on the array element photodetector 5 can be observed.

In the following, SD (Single De)
detector) is a single-element photodetector 6, TAD (Two
Dimensional Array Detecto
r) represents an array element photodetector 5. It is assumed that the random pattern 1 moves on a plane perpendicular to the line of sight at a speed v _v (a vector of the speed v) without changing its shape. It is assumed that the intensity distribution of the random pattern is a Gaussian distribution (N (m _I , σ _I ² )). Here, m _I is the average and σ _I is the standard deviation. The pattern (the above-mentioned random pattern, the same applies hereinafter) is the TAD
Suppose we are focused on top and are statistically independent of neighboring pixels. Half of the incident light is split by the beam splitter 4 and projected on SD. The field of view of the light incident on SD is limited to be the same as illuminating the area of the central pixel of the TAD.

TAD is M pixels × M pixels, and the width of the pixel is Δd. The maximum measurable displacement is d _max = (M /
2) Δd. The difference in sensitivity between pixels is corrected in advance. Array element photodetector 5 and single element photodetector 6
Is sufficiently short, and the moving distance of the random pattern during the exposure time is set to be equal to or less than the width of one pixel. It is assumed that the frame interval t _max is sufficiently long and the array element photodetector 5 can detect the movement. The SD starts data sampling in synchronization with the start of the TAD exposure, and continues sampling until the start of the next frame. SD is also TA
D also has the resolution required to detect the intensity change of the pattern. Here, the minimum measurement speed and the maximum measurement speed are v _vmin = Δd / t _max and v _vmax = d, respectively.
_max /. The velocity resolution Δv _v / v _v is Δt / t
_max (when v _v <Δd / Δt) and Δd / d _max (v
_v > Δd / Δt).

TAD image frames (hereinafter referred to as frames)
The relationship between the pixel pair and the data pair of the SD data column is
1 (b) is as described above. Reference light (reference signal A
Average) and measurement light (same as measurement signal B)
A sufficient number of reference light and measurement light data pairs are measured.
You. Screen position x_vThe light intensity of I (x_v, T)
(X_vIs a vector). x_vIs the pixel position of TAD
Corresponds to the mark. Coordinate x _vIs the center of TAD.
SD observes the light intensity at the origin of the array of TAD pixels.
You. Coordinate x of i-th frame of TAD_v, Time t_iof
Pixel value is Iⁱ(X _v, T_i). i-th frame
Value of data recorded in SD after τ seconds from the start of
To Iⁱ _SD(Τ). i-th data pair (each measurement light
A data pair consisting of an array element and a reference beam.
The same)ⁱ _SD(Τ) = I (0, t_i+ Τ),
Where t_iIs the start time of the i-th frame.
τ is a delay time from the start of the i-frame. Measurement
As a result, the intensity change of each pixel of TAD is
I_TAD(X_v) = ｛Iⁱ _TAD｝ (I = 1,2, ...
I_TADOf SD), and I in SD_SD= ｛I
ⁱ _SD(Τ)｝ (i = 1,2, ...)
Of delay time τ_SDIs obtained. Position x_vpTo
On the other hand, I_TAD(X_vp) = I_SD(Τ) for some τ
Once found, the velocity v_vIs v_v= -X_v/ Τ. x
_vpI to identify_TAD(X_v) And I_SD(Τ) correlation
Is calculated. In the actual measurement, the required data contains measurement noise.
Is included. For simplicity, each of the detectors SD and TAD
Pixels independently generate noise, with a distribution of mean 0 and variance σ
_E ^TwoGaussian distribution N (0, σ_E ^Two).

[0040] the light intensity of the pixel _{x vp I 0 = I TAD (} x vp), the light intensity of the other pixel _{I X = I TAD (x v} ≠ x vp). It is assumed that the pattern is moving without changing its shape.
If the device is noise-free, I ₀ is equal to the light intensity I _SD (τ) with a delay time τ in the paired SD data sequence. However, due to measurement noise, a difference occurs between I ₀ and the value of SD by TAD. The intensity of the pixel x _vp correlated to I _P. The intensity at the other pixels of the TAD and I _B (background level). The SD of the data of the delay time τ and I _T. The respective measurement noises are E _p , E _B and E
_T. At this time, they are expressed as follows.

I _P = I ₀ + E _p I _B = I _X + E _B (1) I _T = I ₀ + E _{T In} equation (1), the i-th data of each variable is superscript like A ^i. Expressed by subscript.

In the above, there is a difference between the intensity at the correlated coordinates (peak level) and the intensity in the background. Peak levels are correlated I _P and I _T, the background level is the correlation of the I _B and I _T. The fluctuation of the peak level and the background level due to noise is
An error occurs in determining the correlation strength.

With respect to the calculation method of the present invention, a method of obtaining a correlation will be described using a two-dimensional model as an example.

(1) General Method (Conventional Method) The two-dimensional correlation is represented by the following equation.

C (x _v ) = <(I ⁱ _TAD (x _v ) − <I ⁱ _TAD (x _v )>) × (I ⁱ _SD (τ) − <I ⁱ _SD (τ)>)> (2 The peak of the correlation strength is expressed by the following equation.

Here, for simplification of the subsequent calculation,
Before calculating C in the description of the prior art, the average value is subtracted.

[0047] ^{_{C p = <(I i P}} - <I i P>) × (I i T - <I i T>)> (3) E p and E _T is the is the Gaussian distribution of the variance sigma _E ² I do. Further, the random pattern of the measurement target is assumed to be Gaussian distributions of variance sigma _I ^2, since the equation (1) and the noise is a Gaussian distribution, I _P the distribution of I _T is Gaussian distribution N (m _I, sigma _I ² + σ _E ² ). Here, m _I is the average, and σ _I ² + σ _E ² is the variance. The two signals I _P and I _T are statistically dependent on the common term I _{0 in} equation (1). Expected value C _{pE of} C _p and unbiased standard deviation σ
_CP is C _pE = σ _I ² for N data pairs (4)

[0048]

(Equation 3)

Is as follows.

On the other hand, with respect to the background ^{_{level, C B = <(I i}} B - <I i B>) × (I i T - <I i T>)> (6) I B and I _T is because it is independent, the expected value C _bE and an unbiased standard deviation sigma _CB of C _B is: for N data pairs.

C _BE = 0 (7)

[0052]

(Equation 4)

As a result, the S / N ratio of the peak normalized for one data pair is as follows.

[0054]

(Equation 5)

Here, ^* indicates that the data has been standardized.

[0056] When σ _E <σ _I, SNR _G is, as variation in light intensity increases, becomes constant at 1/3 ^1/2 (described later). In order to improve the SN ratio, N needs to be increased. Then, SNR _G is modified to be approximately proportional to SNR _G N ^1/2 . (2) Exclusive binarization method According to the correlation calculation method by the optimized exclusive binarization method (EBM) of the present invention, the limit of the SNR _G of the random pattern in the above general method is exceeded. Becomes possible.
EBM is a pair of SADs in the TAD pixel data.
Exclude data of pixels not equal to the value of D (uncorrelated pixels) (set the pixel value to 0). I ⁱ _TAD (x _v ) in each TAD frame is compared with I ⁱ _SD (τ), which is recording data of a paired SD, and is binarized as follows. That is, when both have the same value, the pixel value is set to 1, and when both values are different, the pixel value is set to 0. Actually, both are not exactly the same because of the influence of noise. Therefore, in the present invention, when the magnitudes of the two are within a certain allowable error range δ, they are equal. Therefore, the exclusive binarized image of the present invention is expressed as follows.

I ⁱ _SD (τ) −δ ≦ I ⁱ _TAD (x _v ) ≦ I
^{When i} _SD (τ) + δ, B ⁱ (x _v ) = 1 (10) Here, assuming that δ = 0, instead of B ⁱ (x _v ) = 1,
If B ⁱ (x _v ) = A ² , it becomes the same as the exclusive correlation C ′ (Equation 20) in the description of the conventional technique described above.

In other cases, B ⁱ (x _v ) = 0 (10) 'In this method, the SN ratio (the ratio of the correlation peak value to the background) changes depending on the allowable error width δ. Therefore, in the present invention, the allowable error width δ is determined so as to be optimal in consideration of the magnitude of the variation of the random pattern and the ratio of the measurement noise as described below.

The two-dimensional correlation strength B (x _v ) is obtained by averaging the exclusive binary image {B ⁱ (x _v )}. That is, B ( _xv ) = < ^Bi ( _xv )> (11) This equation corresponds to equation (2) of the general method. The correlation peak appears as a B ( _xv ) peak.

In the EBM, the peak of the correlation intensity is the intensity I
ⁱ _PIs correctly Iⁱ _TThe probability of being recognized as equal to_P
It is. Excluding noise intensity, Iⁱ _PAnd Iⁱ _Tinclude
Iⁱ ₀Are equal. The error is Eⁱ _pAnd Eⁱ _TIs the tolerance
Occurs when it is greater than the enclosure δ. Therefore, this error occurs
The probability of time is expressed by the following equation.

[0061]

(Equation 6)

[0062] Here, N (0,2σ _E ²⁾ mean 0, a synthetic Gaussian distribution of the dispersed 2 [sigma] _E ^2.

Unlike the general method, P _P has no dependence on σ _I. When δ to σ _E , almost all of the probability density of Expression (12) is covered. At this time, P _P is almost equal to one. The expected value B _PE and the unbiased standard deviation σ _BP are as follows.

B _PE = P _P (13)

[0065]

(Equation 7)

On the other hand, the intensity I ⁱ _B , which should be an uncorrelated signal, appears around I ⁱ ₀ due to noise, and may be processed as a correlated signal. The probability P _B of erroneous recognition is
_{^{i B = (I i X +}} E i B) and _{^{I i T = (I i 0}} +
E ⁱ _T ) is the same as the probability that the difference is within the allowable error range δ. Then, the probabilities are expressed as follows using the composite probability density function of I ₀ , I _X , E _B and E _T.

[0067]

(Equation 8)

Where N (0,2σ _I ² + 2σ _E ² )
Is the composite Gaussian probability density function. Average 0, deviation is 2σ _I ² + 2σ _E ^2. δ ~ σ _E ≪σ _I (that is,
δ≪2σ _I ² + 2σ _E ² ). In equation (15), P _B is a small portion of the probability density function near I = 0. Therefore, it is close to P _B = 0. The expected value B _BE and the unbiased standard deviation σ _BB for N data pairs are as follows:

B _BE = P _B (16)

[0070]

(Equation 9)

As shown in equation (9), the S / N ratio of the peak normalized per data pair is as follows.

[0072]

(Equation 10)

As for the SN ratio for N data pairs, SNR _B is proportional to SNR _B ^* × N ^1/2 as N increases.

[0074] The important point here is that the P _P and P _B is not influenced by the variation sigma _I of the light intensity of the random pattern. These are mainly determined only by the measurement noise σ _E. From equation (12), P _P is independent of σ _E. Although P _B is affected by light intensity fluctuations (Equation 15), the probability can be adjusted by δ. The measurement noise σ _E is generally smaller than the light intensity variation σ _I. Therefore, it is possible to improve the SN ratio by exclusive binarization. Before applying the EBM to an actual device, the allowable error width δ is optimized so as to have the best SN ratio for a given σ _I and σ _E. When considering the SN ratio, the parameter σ _I /
σ _E and δ / σ _E are available. This is because even when the system is improved and σ _E is reduced, by making δ / σ _E constant, P _{P in} equation (12) becomes the same. The optimization can be achieved by numerically calculating Expression (18) using Expressions (12) and (15) so as to increase SNR _B ^* . the optimum value of the parameter [delta] / sigma _E with respect to σ _I / σ _E are listed in the second column of the table of FIG. 2 (a), it is plotted in FIG. 2 (b). The optimal value is weakly dependent on σ _I and σ _E. Σ _I / 3 to 100
If realistic σ _E, δ / σ _E is changed from 2.5 to 4.3.

(3) Comparison between the general method and the exclusive binarization method Graphs of the SN ratios calculated by the exclusive binarization method and the general method are shown in FIGS. This is shown in columns 3 and 4 of table a). In the graph of FIG. 3 (a), EBM is one using the [delta] / sigma _E that SN ratio is maximized with respect to σ _I / σ _E. The optimal SN ratio obtained by EBM increases as σ _I / σ _E increases,
The SN ratio of the general method is constant at σ _I / σ _E > 2 (= 1 /
3 ^1/2 ). Since σ _I / σ _E of a practical device is about 2 to 100, SNR _B ^* increases from 0.6 to 6.0. FIG. 3B shows the ratio of the SN ratio between the exclusive binarization method and the general method. The ratio increases almost in proportion to the square root of σ _I / σ _E but does not depend on the number of data. σ
_{When I} / σ _E ≧ 2 ^1/2 , SNR _B ^* is larger than SNR _G. For example, SNR _B / SNR _G = 3.3 and 10.3
(Σ _I / σ _E = 10 and 100). However, σ _I /
When σ _E ≦ 2 ^1/2 , SNR _B is smaller than SNR _G.
As can be seen from this, it is important for the EBM to reduce measurement noise in order to measure at a good SN ratio.

Example of Simulation In order to show the effectiveness of EBM, a simulation was performed for the case described in the above model. The following parameters are assumed: TAD pixels are 11 × 11 (M
= 11). The size of the pixel is Δd = 1 mm.
The frame interval is t _max = 5 ms. The SD sampling interval is Δt = 1 ms. It is assumed that the light intensity of each pixel is statistically independent and has a Gaussian distribution (average m _I
= 10, standard deviation σ _I = 1, N (10,1)). Statistically independent measurement noise was added to the data obtained by each pixel of TAD and SD. The probability density function is a Gaussian function with mean 0. Two cases are considered for the standard deviation of the measurement noise. σ _E = 0.1 (that is, σ _I / σ _E
= 10) and σ _E = 0.01 in other cases (ie, σ _I / σ
_E = 100). Noise is assumed to be statistically independent of light intensity. EBM parameter δ / σ _E is σ
_{For I} / σ _E = 10 and σ _I / σ _E , FIG.
2 and FIG. 2B, 3.2 and 4.1 are set as optimum values. The random pattern has a velocity v _v =
(V _x , v _y ), v _x = −1 m / s, v _y = −1 m / s
Move with. Twenty-five pairs of TAD frames and SD data were obtained synchronously and used for the following correlation calculations.

FIGS. 4 (a) and 4 (b) show three-dimensional plots in the case of the state of zero delay time by the exclusive binarization method and the case of the general method. FIG. 4A shows the case of the exclusive binarization method, and FIG. 4B shows the case of the general method. This is the case where σ _I / σ _E = 10. The two-dimensional correlation of the delay time 0 is calculated by using the EBM equation (11) and the general method equation (2).
Calculate using These figures are standardized as follows. When there was a correlation, the expected value was normalized to 1, and when there was no correlation, the expected value was normalized to 0.

SN of peak height determined by EBM
The ratio is better than the general method. SNR _B = 11
And SNR _G = 2.9 (the evaluation in the table of FIG. 2A or FIG. 2B is 9.0 and 2.8).

FIGS. 5A and 5B respectively show FIGS.
(A) and (b) are two-dimensional correlation intensities with a delay time τ = 3 ms. The pattern appears to move to the center after 3 ms. These results indicate that the EBM has a reduced background level fluctuation in the two-dimensional correlation. This indicates that the movement of the peak correlation position can be determined more clearly than the general method.

FIGS. 6 and 7 show the same ones as FIGS. 4 and 5, in which σ _I / σ _E = 100. FIG.
FIG. 7A shows an initial state of the exclusive binarization method, and FIG.
(A) shows a state where the delay time τ = 3 ms in the exclusive binarization method. FIG. 6B shows the initial state of the general method, and FIG. 7B shows the state of the general method when the delay time τ = 3 ms. The expected correlation values are B _PE = 1,0 and B _BE = 0.02 (σ _I / σ _E = 100). In the case of EBM, the SN ratio is improved. SNR _B = 29, SNR _G = 2.
8 The evaluation based on the table of FIG. 2A and the graph of FIG. 2B indicates 29 and 2.8.

FIG. 8 is a diagram showing an embodiment of the device configuration of the present invention.

In FIG. 8, reference numeral 5 denotes an array element photodetector (TAD). Reference numeral 6 denotes a single-element photodetector (SD). An input interface 21 is an input interface for the array element photodetector (TAD) 5 and the single element photodetector (SD) 6. 22 is a computer. 2
Reference numeral 3 denotes an input unit such as a keyboard. An output unit 24 is a display, a printer, or the like.

Reference numeral 31 denotes a parameter holding unit for holding various parameters required for the correlation operation. 32
Is a correlation calculation unit for calculating the correlation between the measurement signal A and the reference signal B based on the data of the array element photodetector 5 and the single element photodetector 6 and various parameters. The correlation operation unit 32 includes a binarized correlation buffer that accumulates and holds the pixel values of the binarized frame image for each pixel. 3
Reference numeral 3 denotes a unit for holding data input from the data holding unit, the array element photodetector 5, and the single element photodetector 6. Reference numeral 34 denotes a synchronization control unit, which is an array element photodetector 5
And the data input from the single-element photodetector 6 is synchronized. Reference numeral 35 denotes an optimum parameter calculation unit for obtaining δ / σ _{E at} which the S / N ratio is maximum according to various parameters.

In the configuration shown in FIG. 2, various parameters such as σ _I , σ _E , and optimum δ are input from the input unit 23 and are stored in the parameter storage unit 31. σ _I and σ _E are previously measured and obtained, or are obtained by using the recording data of the SD delay time 0. In this case, the calculation can be performed by the optimum parameter calculation unit 35 based on the recording data in the data holding unit 33. The synchronization control unit 34 synchronizes data input in time series from the array element photodetectors 5 and the single element photodetectors 6 for each frame. The respective data input in synchronization are held in the data holding unit 33. The correlation operation unit 32 calculates the exclusive exclusive function according to the input data from the array element photodetector 5 and the single element photodetector 6 held in the data holding unit 33 and the parameters held in the parameter holding unit 31. Exclusive binarization is performed by a binarization method, and a correlation operation between the reference signal and the measurement signal is performed to obtain a correlation peak. The obtained coordinates of the correlation peak, the delay time, and the like are output to the output unit 24.

Various parameters necessary for obtaining the optimum parameters are input from the input unit 23. σ _I and σ _E are
It is also possible to input a value that has been obtained in advance, or to obtain the optimum parameter calculation unit 35 based on the recording data in the data holding unit 33. The optimum parameter calculation unit 35 calculates δ / σ _E optimized so that the SN ratio becomes maximum with respect to σ _I / σ _E according to various parameters. The calculation result is output to the output unit 24.

FIG. 9 is a flowchart of data recording according to the present invention. FIG. 9 is a flowchart of an operation for synchronously acquiring a TAD frame image and SD data in the configuration of FIGS. 1A and 1B. There is a shutter between the array element photodetector 5 (TAD) and the beam splitter 4 in FIG. 1A, and even after the exposure of the array element photodetector 5 is completed, the single element photodetector 6 (SD) Can continue to acquire data.

At S1, i = 1 is initialized. TAD at S2
Exposure and SD data acquisition are started synchronously. The exposure time is waited in S3 (during this time, the TAD image data is obtained and a series of SD data is obtained). In S4, the exposure time ends. In step S5, the process waits for a predetermined time (take a time longer than the maximum value of τ). During this time, the SD keeps recording a series of data. In S6, the acquisition of a series of SD data ends. S7
Record and save the TAD frame image and the SD data string. In S8, it is determined whether or not the data of the expected number of image frames has been obtained. If the data of the expected number has been obtained, the processing is terminated. If the data of the expected number has not been obtained, i is updated in S9, and S2 and thereafter. Is repeated.

FIG. 10 is a flowchart of the correlation operation according to the present invention.

S1 The pixel values of the binarized correlation value buffer are all initialized to 0 (i = 1).

At S2, the TAD and SD data of the i-frame
Enter a pair. At S3, SD recording data and TAD images
Compare prime values. In S4 and S5, the pixel value is (Iⁱ _SD−δ)
And (Iⁱ _SD+ Δ) is determined. Picture
If the prime value is (Iⁱ _SD−δ) and (I ⁱ _SD+ Δ)
The pixel value is set to 1 in S6. If not, pixel value in S7
To 0. TAD frame image binarized in S8
Is added to the binarized correlation value buffer. S9
Since it is determined whether to add a data pair,
If there is, i = i + 1 is set in S10, and the processing after S2 is repeated.
Return. It should be noted that the present invention does not store the acquired data.
Not only the method of calculating the correlation from
It is also possible to perform a correlation calculation. Therefore, in S9
Addition, the data pair recorded in the data holding unit 33 is read.
Or retrieve a new pair of data.
To taste. If it is determined in step S9 that no additional data is to be added,
At 11, each pixel value of the binarized correlation value buffer is divided by i,
Take the average. In step S12, the correlation result is output, and the process ends.
You.

FIG. 11 is a flowchart for determining the optimum parameters according to the present invention.

In S1, σ _I and σ _E are determined. In S2, δ /
P _B with respect to sigma each [delta] / sigma _E while changing the value of _E (Formula 1
2) and P _B (Equation 15) are numerically calculated. In S3, the SN ratio (Equation 18) is calculated, and δ / σ _E that maximizes the SN ratio is selected for each σ _I / σ _E.

At S4, the optimum parameters are output.

[0094]

According to the present invention, the binarization based on the difference between the magnitude of the reference signal and the measurement signal, the SN ratio indicating the sharpness of the correlation peak is maximized in accordance with the optimized error tolerance. Can be performed as follows. Therefore, the correlation peak can be determined with high accuracy. Therefore, the movement of an object moving at a high speed can be accurately measured at a high speed. Therefore, it can be used for observing fluctuations of the earth's atmosphere, and it is possible to accurately observe the wind speed and direction of the fluctuation layer of the earth's atmosphere.

[Brief description of the drawings]

FIG. 1 is a diagram showing a first embodiment of the present invention.

FIG. 2 is a graph showing a graph 1 showing a result of the present invention.

FIG. 3 is a graph showing a graph 2 showing a result of the present invention.

FIG. 4 is a three-dimensional diagram 1 of the correlation result of the present invention.

FIG. 5 is a three-dimensional diagram 2 of the correlation result of the present invention.

FIG. 6 is a three-dimensional diagram 3 of the correlation result of the present invention.

FIG. 7 is a three-dimensional diagram 4 of the correlation result of the present invention.

FIG. 8 is a diagram showing a second embodiment of the present invention.

FIG. 9 is a diagram showing a flowchart of data recording of the present invention.

FIG. 10 is a diagram showing a flowchart of a correlation operation of the present invention.

FIG. 11 is a diagram showing a flowchart of determining an optimum parameter according to the present invention.

FIG. 12 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

 1: random pattern 3: condensing lens 4: beam splitter 5: array element photodetector 6: single element photodetector 7: pinhole 8: measurement light 9: reference light

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04N 7/18 H04N 7/18 K

Claims (9)

[Claims] In a method for measuring a correlation between a reference signal and a measurement signal which is a moving measurement target signal, the signal values of the reference signal and the measurement signal are compared to determine whether the values of the reference signal and the measurement signal are the same. This is to binarize the measurement signal exclusively according to whether the reference signal and the measurement signal have the same allowable error width as a criterion for judging whether the value of the reference signal and the measurement signal are the same. An optimized exclusive binary correlation measurement method, characterized in that exclusive binarization of a measurement signal is performed based on a relationship between a signal and the allowable error width, and a correlation between the reference signal and the measurement signal is measured. 2. The method according to claim 1, wherein the allowable error width is determined according to a measurement environment, and exclusive binarization is performed according to a magnitude relationship between a difference between a reference signal and a measurement signal and the optimized allowable error width. 2. The optimized exclusive binary correlation measurement method according to claim 1, wherein: 3. An SN ratio obtained by dividing the difference between the expected value of the correlation peak and the expected value of a portion having no correlation by the magnitude of the fluctuation of the expected value, so that the SN ratio is maximized. 3. The method according to claim 1, wherein the allowable error width is optimized.
2. The optimized exclusive binary correlation measurement method described in 1. above. 4. The measurement object is a random pattern that moves without changing its shape, the measurement signal is a signal of a two-dimensional image of the measurement object, and the reference signal is a limited two-dimensional image of the measurement signal. 4. A signal of an image of a region, which is acquired in synchronization with a measurement signal, and measures a movement of a measurement target by correlation between a reference signal and the measurement signal. 3. The optimized exclusive binary correlation measurement method according to item 3. 5. An array element photodetector for receiving measurement light, a single element photodetector for receiving a part of the measurement light, a measurement signal output from the array element photodetector, and a single element photodetector. What is claimed is: 1. An optimized exclusive binary correlation measuring device comprising: a correlation calculating unit for obtaining a correlation between a reference signal to be output; and the correlation calculating unit compares respective signal values of the reference signal and the measurement signal.
This is to binarize the measurement signal exclusively according to whether the value of the reference signal and the measurement signal are the same, and optimized as a criterion for determining whether the value of the reference signal and the measurement signal are the same. An optimization exclusive feature characterized by having an allowable error width, optimizing exclusive binarization of the measurement signal based on a relationship between the reference signal, the measurement signal, and the allowable error width, and measuring a correlation between the reference signal and the measurement signal. Binary correlation measurement device. 6. An allowable error width is determined according to a measurement environment, and performs an exclusive binarization according to a magnitude relation between a difference between a reference signal and a measurement signal and the optimized allowable error width. The optimized exclusive binary correlation measurement apparatus according to claim 5, wherein: 7. When the SN ratio is obtained by dividing the difference between the expected value of the correlation peak and the expected value of a portion having no correlation by the magnitude of the fluctuation of the expected value, the SN ratio is maximized. 7. The method according to claim 5, wherein the allowable error width is optimized.
2. The optimized exclusive binary correlation measurement device according to item 1. 8. A program for measuring a correlation between a reference signal and a measurement signal which is a moving measurement object signal, comparing respective signal values of the reference signal and the measurement signal, and determining whether the values of the reference signal and the measurement signal are the same. This is to binarize the measurement signal exclusively according to whether the reference signal and the measurement signal have the same allowable error width as a criterion for judging whether the value of the reference signal and the measurement signal are the same. Exclusive binarization of the measurement signal based on the relationship between the signal and the permissible error width, and execution of the optimized exclusive binarization correlation measurement by measuring the correlation between the reference signal and the measurement signal. And the program. 9. A program for optimizing an optimized exclusive binary correlation calculation, compares respective signal values of a reference signal and a measurement signal, and determines whether the reference signal and the measurement signal have the same value. This is to binarize the measurement signal exclusively, and to give an allowable error width as a criterion for judging whether the value of the reference signal and the measurement signal are the same, and to determine the relationship between the reference signal, the measurement signal and the allowable error width. Is used to exclusively binarize the measurement signal, and the permissible error width is determined according to the measurement environment, and the difference between the expected value of the correlation peak and the expected value of the uncorrelated portion is calculated by: A program characterized in that when an SN ratio is an amount obtained by dividing the expected value fluctuation, the allowable error width optimized so as to maximize the SN ratio is obtained.