JPH0676047A - Picture processor - Google Patents

Abstract

(57) [Summary] [Purpose] An image processing apparatus capable of accurately identifying noise in an image and a target object by setting a threshold value with higher reliability and automatically updating according to environmental changes. To provide. [Structure] Image input means for capturing and inputting an optical image, statistics of a plurality of images obtained in time series by this image input means are calculated, and an average luminance value and a threshold value as a background image are calculated. Threshold calculation means for obtaining, change area detection means for obtaining a difference between the newly imaged input image and the average brightness value, and a change area for a portion where the difference is larger than the threshold, and the newly picked up image An updating means for updating the average brightness value and the threshold value according to the input image is provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus used in an image monitoring apparatus for automatically detecting an intruder from a captured image.

[0002]

2. Description of the Related Art In recent years, image monitoring devices used for the purpose of automatically detecting intruders in banks and offices have been actively developed. In such an image monitoring device,
Image processing technology is required to extract only intruders as target objects from the captured images. Here, as a method of detecting the target object, (1) a method of examining a difference between temporally continuous images and setting a region having a difference in the image as the target object

(2) The average of the brightness values of a plurality of images taken without including the target object is set as the brightness value of each pixel corresponding to the reference image (background image), and then the target object is selected. A typical method is to capture an image including the image and set a portion having a difference from the background image as a target object.

However, due to changes in lighting and flickering of fluorescent lamps,
The brightness of the image changes from moment to moment due to changes in the brightness of daylight. In the above two methods, since the difference between the images is used to detect the target, this change causes noise. Therefore, a means for removing these noises and detecting the target object is required.

In (1), when there is a change in the brightness value between images due to flicker of a fluorescent lamp or the like, when the difference in brightness value is taken, the brightness which cannot be erased due to the difference remains at the changed position as noise. . In order to eliminate this, a means has been developed in which a threshold value is set in advance and a difference between images is set to be a target object. However, the problem of setting the optimum threshold has not been solved, and empirical values or experimental values under a specific environment are used. Therefore, if the environment to be photographed changes, the threshold value must be manually set again, and if the change in lighting is larger than the threshold value, noise remains, which causes misidentification. .

In (2), a means is developed in which fixed weighting is performed between corresponding pixels of a newly photographed image and the previously obtained background image to obtain the average, and the luminance value of the background image is obtained again. is doing. By updating the reference image, it is possible to respond to changes in illumination within the image. However, although it is possible to deal with the gradual change in the brightness of the illumination by updating, it is not possible to deal with the variation in the brightness due to the flicker of the fluorescent lamp and the noise cannot be sufficiently removed. Therefore, although the same means as the threshold processing of (1) is used, there is a problem of threshold setting.

[0007]

As described above, the conventional method has a drawback that the noise in the image and the target object cannot be accurately discriminated because the preset threshold value cannot correspond to the change of the environment. It was

The present invention eliminates the above-mentioned conventional drawbacks, sets the threshold value more reliably, and automatically updates the threshold value in response to environmental changes, so that the noise in the image and the target object can be accurately measured. An object is to provide an image processing device that can be identified.

[0009]

According to the present invention, an average value and a standard deviation are obtained based on the brightness value of an image which does not include a target imaged by a preset number of images. Next, the background image is estimated using the average value, and the difference in luminance value between the newly input image and the background image is calculated. Further, the threshold value is obtained based on the target deviation, and the probability of the difference due to the variation in the luminance value of the background image is obtained by comparing the difference between the luminance values and the threshold value. This is,
If it is smaller than the required detection error rate, it is determined as a target. The image capturing time for estimating the background image is determined from the average value or the number of images determined from the reliability required for the standard deviation and the image capturing interval. If it is determined that the target object is present, the area of the target object is obtained, and the brightness value and threshold value of the next background image are estimated from the brightness values of the other areas and the brightness value of the background image. When it is determined that the target object is not the target object, the next background image and the threshold value are estimated based on the brightness value of the image and the brightness value of the background image.

[0010]

In the present invention, by using the standard deviation, the threshold value is set on the basis of the degree of variation of the background image, so that noise caused by a change in the brightness value due to, for example, flicker of illumination is eliminated. It is possible to detect the change in the brightness value due to the target object.

Further, by updating the background image or the threshold value in accordance with a new image, it is possible to use an appropriate threshold value corresponding to a slow change in the brightness value due to the shading of the sun or the like.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows the overall structure. The image processing apparatus is an image pickup apparatus 1 for photographing an image, an image input circuit 2 for digitizing the photographed image and recording it in an image memory 3, an initial background image and an initial background image for calculating an initial standard deviation from the image.
A deviation calculation circuit 4, a change area detection circuit 5 for detecting a change area from an image, a background image and a standard deviation of brightness values, an average of brightness values from the image and the background image, and a brightness value from the image and the standard deviation. A background image / deviation updating circuit 6 for updating the standard deviation, and an image output device 7 for outputting the changed area. Each constituent device will be described below.

FIG. 2 shows the configuration of the initial background image / deviation calculation circuit 4. Further, FIG. 3 is a graph showing a frequency distribution of luminance values of continuous images. When the frequency distribution of the brightness value is taken for each pixel of the images taken continuously in terms of time, it has an average value and a width as shown schematically in FIG. 3 due to the flicker of illumination. To be distributed. That is, the brightness value has this frequency distribution as a probability distribution. When a brightness value that does not fit into this distribution is obtained, it is considered that an object that is not in the environment was photographed. Therefore, in order to obtain this distribution of brightness values, the standard deviation of the brightness value for each pixel and the background image having the average brightness value of the recorded image as the brightness value is estimated. , For example, using a mean test as described below. In the following estimation, the average value and standard deviation of the brightness value for each pixel are obtained according to the following formula.

[0015]

[Equation 1]

[0016]

[Equation 2] p [i, t] = time order i-th image of the t-th image taken
Elementary luminance value (background image luminance value) <p> [i, N] ＝ t = 1th to t = Nth image, i-th image
Average luminance value σ [i, t] = t = 0 to t-th image, i-th pixel
Standard deviation of the luminance value of

Let T be the number of images required to estimate the background image. Grouping circuit 8 with T being an even number of 2 m
Then, the image is divided into two groups a and b. The even-numbered images are grouped a and the odd-numbered images are grouped b according to the number taken. In each group, the average <p> [i, T] a, <p> [i,
T] and variances S [i, T] a and S [i, T] b are obtained. Considering the probability distributions of each of the two groups, it is considered that the population mean and population variance are equal between the groups. Therefore, the hypothesis that the population means are equal is established, the mean and variance are calculated while adding new images to the group, the hypothesis test circuit 10 tests the hypothesis, and the images until the rejection is not performed are the background images. Used to estimate. That is, T is set to be twice the m at that time. The details of this method are shown below. For the mean and variance of the two groups

[0018]

[Equation 3] However, according to the t distribution with 2m-2 degrees of freedom. Therefore, if the risk rate β is determined in advance and τ does not fall within the rejection area, 2
Estimate the background image and standard deviation from the group mean and variance. The estimated background image and the standard deviation are recorded in the background image memory 11 and the deviation memory 12, respectively.

[0019]

[Equation 4] <p> [i, T] = (<p> [i, t] a + <p> [it] b) / 2

[0020]

[Equation 5]

The method of estimating the background image and the standard deviation is not limited to the above. For example, the difference between the average luminance value <p> [i, t] of t images captured in time series and the average value <p> [i, t + 1] obtained by adding another image Calculate Δp.
Take the ratio Δp / <p> between this and <p> [i, t] or <p> [i, t + 1], and calculate t or t + 1 when this is smaller than the predetermined rate of change γ. As T, the background image and the standard deviation may be estimated from the used image. Further, in this method, the standard deviation may be used instead of the average value. Alternatively, the product or the sum of the ratio in the average value and the ratio in the standard deviation may be obtained and compared with the rate of change. In addition, the background image, standard deviation, and estimation can be performed by various methods without departing from the scope of the present invention.

FIG. 4 shows the configuration of the change area detection circuit 5.
A newly photographed image is obtained from the memory 3, and the luminance calculation circuit 13 takes the luminance difference from the background image obtained from the background image memory 11 for each pixel. The calculation formula is as follows.

[0023]

## EQU6 ## δ [i, t '] = p [i, t']-<p> [i, T] δ [i, t '] = t'th image up to the i'th pixel and T'th Difference from the luminance value of the i-th pixel of the background image obtained from the image

As schematically shown in FIG. 5A, the average value and standard deviation of the probability distribution of the brightness value p [i, t '] of the new image are
It is approximated by the average <p> [i, T] obtained immediately before and the standard deviation σ [i, T]. That is, the probability distribution of δ [i, t '] is approximated with 0 being the average and σ [i, T] being the standard deviation, as shown in FIG. 5B.
Therefore, the threshold value calculation circuit 14 causes the deviation memory 12
The threshold value is determined by using σ [i, T] obtained from the above, and the threshold value processing circuit 15 extracts an area in which the absolute value of the brightness difference is larger than the standard deviation, and sets this as a change area and the output device 7 Output by. For the t'th image taken after the T'th image, this is performed by the following algorithm, for example. IF {δ [i, t ']> δ (MAX) [i, t'] OR δ [i, t '] <-δ (MAX) [i, t']} THEN flag [i] = change ELSE flag [i] = unchange

Flag [i] indicates whether the i-th pixel is a change area (change) or not (unchange). In this algorithm, the change region determination result flag [i] is rewritten every time a new image is obtained, but the change region for each image may be recorded as flag [i, t '].

Δ (MAX) [i, t '] is a threshold value for the i-th pixel, and δ [i, t in the probability distribution of FIG. 5 (b).
t '] is greater than or equal to δ (MAX) [i, t'], or -δ (MAX) [i, t ']
Set the probability that is less than or equal to the required value P1. For example, when P1 is 0.01, δ (MAX) [i, t '] is 2.576.
Set as σ [i, T]. Further, the number of pixels exceeding the threshold value is output to the weight calculation circuit 18 for use in updating the background image and standard deviation described below.

FIG. 6 shows the configuration of the background image / deviation updating circuit 6. FIG. 7 is a graph showing changes over time in the average luminance value and standard deviation. As schematically shown in FIG. 7, for example, the average <p> of the luminance values of the pixels and the standard deviation σ change with time due to the shadow of the day. Therefore, from the background image and the pixel values other than the change area of the image, the weight w is used to calculate the average value by the brightness value average calculation circuit 16 and the brightness value deviation calculation circuit 17
The standard deviation is obtained by the following, and is recorded in the background image memory 11 and the deviation memory 12 as a new background image and standard deviation corresponding to the time change.

The weight w is determined by the weight calculation circuit 18 as follows, for example. FIG. 8 is a graph showing the probability distribution of brightness values. As shown in FIG. 8, the luminance value p is <p> [i, T] -δp and <p> in the probability distribution of mean <p> [i, T] and standard deviation σ [i, T].
δp is determined so that the probability within the range bounded by [i, T] + δp is a preset value P2. The percentage of pixels whose luminance value is larger than the upper limit <p> [i, T] + δp of the pixels in the new image is w (+) and lower than the lower limit <p> [i, T]-δp. Let w (-) be the proportion of pixels with small values

[0029]

7] It is assumed that w = w (+)-w (-). The average and standard deviation of brightness values are calculated by the following formula.

[0030]

[Equation 8] <p> [i, t '] = w-p [i, t'] + (1-w)-<p> [i, T]

[0031]

[Formula 9] σ [i, t '] ² = wp [i, t '] ² -<p> [i, t '] ² -(1-w)
(σ [i, T] ² -<p> [i, T]) ²

In the present embodiment, the threshold is updated by updating the standard deviation calculated based on the image and the difference between the newly photographed image and the background image. The threshold updating is not limited to this. The threshold value determined empirically may be regarded as the standard deviation, and the updating by the weighting may be performed. In addition, in the environment where the lighting has flicker, but the brightness and the width of the flicker are stable, the background image, or
It is not necessary to update the threshold value. Other modifications can be made without departing from the scope of the present invention.

[0033]

According to the present invention, since the threshold value is determined from the standard deviation, it is possible to set a natural threshold value in consideration of the variation of the natural phenomenon such as the change of the brightness value due to the change of the illumination. Therefore, it is possible to reduce the false recognition of the change of the brightness value due to the change of the illumination and the change due to the target object, improve the reliability of the discrimination, and improve the reliability of the image monitoring device and the image identifying device.

Further, according to the new input image, the background image,
Alternatively, by updating the threshold value, it is possible to reduce the misidentification of the slow change of the brightness value due to the shading of the day and the change due to the target object and improve the reliability of the distinction, thereby improving the image monitoring device and the image identification. The reliability of the device can be improved.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of an embodiment of the present invention.

FIG. 2 is a configuration diagram of an initial background image / deviation calculation circuit according to an embodiment of the present invention.

FIG. 3 is a diagram showing a frequency distribution of luminance values of continuous images.

FIG. 4 is a configuration diagram of a change area detection circuit according to an embodiment of the present invention.

FIG. 5 is a diagram showing an average luminance value and a standard deviation of a background image and a diagram showing a probability distribution of a luminance value difference between a new image and a background image.

FIG. 6 is a configuration diagram of a background image / deviation updating circuit according to an embodiment of the present invention.

FIG. 7 is a diagram showing changes over time in average luminance value and standard deviation.

FIG. 8 is a diagram showing a probability distribution of luminance values.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Imaging device 2 ... Image input circuit 3 ... Image memory 4 ... Initial background image / deviation calculation circuit 5 ... Change area detection circuit 6 ... Background image / deviation update circuit 7 ... Image output device

Claims (3)

[Claims] 1. An image inputting means for picking up and inputting an optical image, and a statistic amount of a plurality of images time-sequentially obtained by the image inputting means is calculated to calculate an average luminance value and a threshold value as a background image. Threshold value calculating means for obtaining a value, change area detecting means for obtaining a difference between the newly captured image and the average luminance value, and a change area having a portion where the difference is larger than the threshold value as a change area; An image processing apparatus, comprising: an updating unit that updates the average luminance value and the threshold value according to the image captured and input. 2. The threshold value calculating means calculates a standard deviation of corresponding pixels of a plurality of images as a statistic, and obtains a threshold value based on the standard deviation. Image processing device. 3. The image processing apparatus according to claim 2, wherein the updating means updates the average luminance value and the threshold value based on the difference and the standard deviation.