JPH0719284B2 - Anomaly detection device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an abnormality detection device that analyzes image information of an object to be measured and detects an abnormality in the object to be measured.

(Prior Art) Conventionally, as this type of detection device, a device to which a feature extraction method is applied, a device to which a pattern matching method is applied, and the like are known.

The feature extraction method extracts various geometric features of an object or a target scene (hereinafter referred to as an object to be measured), and changes the characteristics of the object to be measured, for example, a local feature obtained from a part of an image. It is intended to detect image abnormalities. Further, the pattern matching method is to create a standard pattern of a binary image from an object to be measured and detect an abnormality in the image based on the degree of similarity between the pattern to be measured and the standard pattern. For example, as an abnormality detection method for an image based on grayscale information using this pattern matching method, and an abnormality detection device,
There are inventions related to JP-A-53-50620 and JP-A-53-75735.

(Problems to be solved by the invention) However, in the conventional abnormality detection device to which these methods are applied, for example, in the case of inspecting the coating state of an object (especially in the case of detecting color unevenness), a detection signal is detected. However, the feature extraction method has a disadvantage that the image abnormality cannot be inspected because of the small change. In addition, the simple pattern matching method has a small difference between the normality and the abnormality of the detection signal, so that highly accurate determination cannot be performed. The cause of the decrease in detection accuracy is due to thermal noise included in the image information.

The present invention has been proposed in order to solve the above problems, and even if the difference between the normal signal and the abnormal signal of the image information of the measurement target is small, it is possible to accurately detect the abnormality of the measurement target. An object is to provide a detectable abnormality detection device.

(Means for Solving the Problems) The present invention is directed to an average value deviation, a standard deviation or a maximum value deviation of a grayscale frequency distribution based on thermal noise information included in image information, and an abnormal portion such as thermal noise information and color unevenness. By comparing the average value deviation, standard deviation, or maximum value deviation of the grayscale frequency distribution for the sum of information, it is possible to easily detect even a minute change that could not be detected due to being buried in thermal noise. It is based on knowledge.

That is, the present invention is an imaging means for imaging an object to be measured,
An A / D converter that converts analog image information indicating the light and shade captured by the image capturing unit into digital image information and outputs the digital image information, and a frame that stores and outputs one-time observation information of the digital image information for each pixel. A memory, a standard pattern generation means for sequentially inputting and storing the digital image information, accumulating and averaging the digital image information for each pixel a plurality of times, and outputting the average as a standard pattern, and for each pixel, Standard pattern and above 1
A subtractor that calculates and outputs the difference from the observation information of the number of times and the output of the subtractor for each pixel are used as density values, and the number of pixels with the same density value is used as the frequency to calculate the grayscale frequency distribution for all pixels. , A deviation value calculating means for calculating and outputting an average value deviation, a standard deviation or a maximum value deviation of this frequency distribution, and a difference between the deviation in the normal state and the deviation in the measuring time outputted from the deviation value calculating means. And a determination section that outputs an abnormality detection signal when exceeds a predetermined value.

(Operation) In a normal state, the digital image information of the object to be measured is sequentially input and stored in the standard pattern generating means as normal image information, and integrated (integrated) and averaged a plurality of times for each pixel. Since the thermal noise information included in the digital image information becomes so small that it can be ignored by the integration operation, the information from which the thermal noise information has been removed is output to the subtractor as a standard pattern. On the other hand, the normal image information is stored in the frame memory as one-time observation information, and this observation information is also output to the subtractor.

The subtractor subtracts the normal image information from the standard pattern, thereby canceling the information of the digital image information depending on the measured object itself, and the error mainly depends on one thermal noise in a normal state. Will be things. This difference is output to the deviation value calculating means, and the deviation value calculating means calculates the frequency distribution of the error in the normal condition, and calculates the deviation (mean value deviation, standard deviation or maximum value deviation) of this density distribution. This is output to the determination unit. If the frequency distribution is normalized, it is a frequency distribution of thermal noise information once in a normal state. The determination unit stores the deviation of the density distribution as the reference deviation that serves as a reference for orientation.

By the way, assuming that the image information observed when detecting the presence or absence of abnormality is the detection target image information, this detection target image information is input to the frame memory similarly to the normal image information and then output to the subtractor. It In this case, the subtractor calculates the difference between the detection target image information and the standard pattern.
The deviation value calculation means inputs this difference and calculates the grayscale frequency distribution for the error. The frequency distribution at this time is the frequency distribution of the sum of the thermal noise information included in the digital image information and the information of the abnormal portion, and when there is no abnormal portion, it becomes equal to the frequency distribution of the thermal noise information once in the normal state. Various deviation values are calculated from this density distribution and the calculation result is output to the determination unit.

The determination unit compares the stored standard deviation value in the normal state with the deviation value, and when the difference between these deviation values is a certain value or more, it is determined to be abnormal and an abnormality detection signal is detected. Is output. That is, the determination unit compares the average value deviation of the grayscale frequency distribution due to the thermal noise in the normal state and the average value deviation of the grayscale frequency distribution due to the sum of the thermal noise and the abnormal component at the time of measurement, etc. Determine whether the measured object is normal or abnormal.

Embodiment An embodiment of the present invention will be described below with reference to FIG. First
In the figure, an object to be measured 10 is an image pickup means 1 such as a television camera.
Has been observed by. This image pickup means 1 is an A / D converter 2
This A / D converter 2 converts the analog image information indicating the grayscale imaged by the image pickup means 1 into analog / digital data.
It is a digital conversion.

The output of the A / D converter 2 is input to the frame memory 3 and the standard pattern generation means 4, respectively, and the digital image information input to the frame memory 3 is observed once for each observation and for each pixel. Remembered. On the other hand, as the digital image information input to the standard pattern generation means 4, the digital image information for each pixel is sequentially input and stored, and the stored digital image information is integrated and averaged for each pixel, This averaged information is output as a standard pattern.

The outputs of the frame memory 3 and the standard pattern generating means 4 are input to the subtractor 5, and the subtractor 5 calculates the difference between the standard pattern and the one-time observation information from the frame memory 3 for each pixel. The output of the subtractor 5 is input to the frequency distribution calculation unit 6, and the frequency distribution calculation unit 6 calculates the difference between the standard pattern and the observation information of one time for each pixel based on the output of the subtractor 5. Compute the intensity frequency distribution of all pixels.

The output of the frequency distribution calculation unit 6 is input to the deviation value calculation unit 7, and the deviation value calculation unit 7 calculates the average value deviation, standard deviation or maximum value deviation of the input frequency distribution. Furthermore, the output of the deviation value calculation unit 7 is input to the determination unit 8, and the determination unit 8 stores the average value deviation value, the standard deviation or the maximum value deviation in a normal state as a reference deviation,
An abnormality detection signal is output when the difference between the reference deviation and the average deviation, standard deviation or maximum deviation during measurement exceeds a predetermined value.

The frequency distribution calculation unit 6 and the deviation value calculation unit 7 constitute deviation value calculation means 20.

The operation of the abnormality detection device configured as described above will be described below. Now, the object to be measured 10 is imaged a plurality of times (n times) by the image pickup means 1, and the analog image information of this image pickup means 1
It is converted into digital image information by the A / D converter 2.

Here, if the image information of each pixel is G (x, y), G
(X, y) can be generally expressed by the following equation.

G (x, y) = S (x, y) + C (x, y) + T (x, y) ... (1) (In an abnormal condition, S (x, y) is S (x, y)) + V
It becomes (x, y). ) However, S (x, y): Information to be observed C (x, y): Information unique to the imaging means 1 T (x, y): Information due to thermal noise V (x, y): Information of abnormal portion ( For example, the uneven color part).

By the way, if the normal image information is G ₀ (x, y), (1)
From the formula, G ₀ (x, y) becomes G ₀ (x, y) = S (x, y) + C (x, y) + T (x, y) ... (2).

Therefore, under normal conditions, this G (x, y) is sequentially input and stored in the standard pattern generating means 4 as G ₀ (x, y), and integrated (integrated) n times for each pixel and averaged. And
This averaged information is output to the subtractor 5 as a standard pattern G ₀ (x, y).

This G ₀ (x, y) is And G ₀ (x, y) is expressed by the following equation.

G ₀ (x, y) = S (x, y) + C (x, y) + T (x, y)
(4) However, S (x, y): Information obtained by integrating the measured object information n times and averaged C (x, y): Information obtained by integrating the unique information of the imaging means 1 n times and averaging T (x, y): Information obtained by integrating and averaging information due to thermal noise n times.

However, since the thermal noise information T (x, y) is so small that it can be ignored by n times of integration, T (x, y) 0 (5) can be obtained. Therefore, the standard pattern G ₀
(X, y) becomes G ₀ (x, y) S (x, y) + C (x, y) (6) from the equations (4) and (5).

On the other hand, G (x, y) (that is, G ₀ (x, y)) under normal conditions is
It is stored in the frame memory 3 as one-time observation information, and this observation information is output to the subtractor 5.

The subtracter 5 calculates the difference Gd ₀ (x, y) between the input normal image information G ₀ (z, y) and the standard pattern G ₀ (x, y). This Gd ₀ (x, y) is calculated from the equations (2) and (6) as follows: Gd ₀ (x, y) = G ₀ (x, y) −G ₀ (x, y) = S (x, y) -S (x, y) + C (x, y) -C
(X, y) + T (x, y) (7)

Here, since S (x, y) S (x, y) (8) can be used, Gd ₀ (x, y) is (7), (8)
From the formula, Gd ₀ (x, y) C (x, y) −C (x, y) + T (x, y) ...
It can be expressed as (9).

C (x, y) -C (x, y) in the above equation is a signal due to a drift component specific to the image pickup means 1 and is information of a DC signal component.

This Gd ₀ (x, y) is output to the frequency distribution calculation unit 6, and the frequency distribution calculation unit 6 calculates the grayscale frequency distribution f ₀ under normal conditions. Graphs of this frequency distribution f ₀ are as shown in FIGS. FIG. 2 (b) is symmetrical with respect to the density value 0, and has a normal distribution. 2 (a) and 2 (c) are shifted by the DC component C (x, y) -C (x, y), and when normalized, they are the same as those in FIG. 2 (b). The figure (a) is C (x, y)
<C (x, y), C (x, y)> C in FIG.
In the case of (x, y), the figure (c) shows C (x, y) C (x,
y) in each case.

Further, the frequency distribution f ₀ is output to the deviation value calculation unit 7, and the deviation value calculation unit 7 calculates the deviation (average value deviation, standard deviation or maximum value deviation) of f ₀ and outputs this to the determination unit 8. To do. Then, the determination unit 8 stores the deviation of f ₀ as a reference deviation that serves as a reference for determination.

By the way, assuming that the image information G (x, y) observed when detecting the presence or absence of abnormality is the detection target image information _Gv (x, y), _Gv (x, y) is expressed by the following equation. It

G _v (x, y) = S (x, y) + C (x, y) + T (x, y) + V (x,
y) (10) However, V (x, y) is the information of the abnormal portion (for example, due to the uneven color portion) as described above. This G _v (x, y)
Is also input to the frame memory 3 and output to the subtractor 5 as in the case of G ₀ (x, y). The subtractor 5 outputs G _v (x, y) and the standard pattern G ₀ (x, y). Calculate the difference between and. This difference is G
_{If dv} (x, y), G _dv (x, y) is (6), (7),
From equation (10), G _dv (x, y) C (x, y) -C (x, y) + T (x, y) + V
(X, y)… (11) This G _dv (x, y) is output to the frequency distribution calculation unit 6, and the grayscale frequency distribution f is calculated.

For the sake of simplicity, C (x, y) C (x, y)
T (x, y) V ( x, y) the density frequency distribution of f _T, When _{f V f, f T, f} V is as shown in Figure 3. Here, f _T has the same distribution as the grayscale frequency distribution f _{0 under} normal conditions.

From FIG. 3, it can be seen that the grayscale frequency distribution differs depending on the presence or absence of an abnormal portion.

The deviation value calculation unit 7 calculates various deviation values from the grayscale frequency distribution f and outputs them to the determination unit 8. The determination unit 8 compares the stored standard deviation value in the normal state with the deviation value, and when the difference between the deviation values becomes a certain value or more, the object 10 to be measured is abnormal. Judges and outputs an abnormality detection signal.

The operation of the deviation value calculator 7 will be described in detail below. That is, the deviation value calculation unit 7 calculates, for example, the average value deviation M by the following formula based on the grayscale frequency distribution.

However, Bi: density value f (Bi): frequency of density value Bi: average brightness A: number of target pixels, and the average brightness is It is represented by.

This average value deviation M is output to the determination unit 8, but the determination unit 8
Stores the reference mean value deviation Ms that serves as the reference for orientation,
The average value deviation M is compared with the reference average value deviation Ms, and when the following expression is satisfied, it is judged that the object 10 to be measured is abnormal.

Ms <M or Ms + δ <M (13) (where δ is a marginal value) Here, the reference mean value deviation Ms is the frequency distribution f measured by observing the normal object 10 to be measured and performing the same method as the above method. It is obtained by calculating.

Although the above embodiment is a method of making a determination based on the average value deviation, it can also be made by obtaining the maximum value deviation if there is a standard deviation from the grayscale frequency distribution. In this case, the standard deviation σ and the maximum value deviation R can be calculated by the following equations.

R = (Bi) max− where (Bi) max is the maximum density value.

Assuming that the standard standard deviation and the standard maximum value deviation, which serve as the standard of orientation, are σs and R _S >, respectively, it can be determined that the abnormality is satisfied when the following equation is satisfied.

σs <σ or σs + δ <σ (15) R _S <R or R _S + σ <R (16) (where δ is a margin) (Effect of the invention) As described above, the present invention Since it was decided to compare the average value deviation of thermal noise information in the normal condition of the image information with the average value deviation of the sum of the thermal noise information and the information of the abnormal part in the measurement, the conventional method It is possible to detect with high accuracy a minute change in image information that is buried and cannot be detected.

Further, as the deviation, any one of the average value deviation, the standard deviation or the maximum value deviation, or a combination of these deviations is used as a judgment parameter to detect an appropriate image abnormality according to the property of the object to be measured. Is possible.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a graph showing a grayscale frequency distribution due to thermal noise for explaining the operation of the present invention, and FIG. To explain the operation of the invention,
It is a graph which shows a density distribution when there is an abnormality. 1 ... Image pickup means, 2 ... A / D converter, 3 ... Frame memory, 4 ... Standard pattern generating means, 5 ... Subtractor, 6 ... Frequency distribution calculation section, 7 ... Deviation value calculation section, 8 ... Judgment unit 10 ... Object to be measured, 10 ... Deviation value calculation means

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Nobuyuki Kanehira No. 1 Shinden Tanabe, Kawasaki City, Kanagawa Prefecture Fuji Electric Co., Ltd. (72) Kenmitsu Kodama No. 1 Shinden Tanabe, Kawasaki City, Kanagawa Prefecture Within Fuji Electric Co., Ltd.

Claims (1)

[Claims] 1. An image pickup means for picking up an object to be measured, an A / D converter for converting the analog image information showing the light and shade picked up by the image pickup means into digital image information and outputting the digital image information, and the digital image information A frame memory that stores and outputs one-time observation information for each pixel, and sequentially inputs and stores the digital image information, and the digital image information is integrated a plurality of times for each pixel and averaged. Standard pattern generating means for outputting as a standard pattern, and subtraction for inputting each output of the frame memory and the standard pattern generating means and calculating and outputting a difference between the standard pattern and the one-time observation information for each pixel And the subtractor output for each pixel are used as density values, and the number of pixels with the same density value is used as the frequency to calculate the grayscale frequency distribution for all pixels. , A deviation value calculating means for calculating and outputting an average value deviation, a standard deviation or a maximum value deviation of this frequency distribution, and a difference between the deviation in the normal time and the deviation in the measuring time outputted from the deviation value calculating means. An abnormality detection device comprising: a determination unit that outputs an abnormality detection signal when exceeds a predetermined value.