JP2011123726A - Image processing device for monitoring - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing device for monitoring, which precisely detects an event even from a captured image with low brightness and low contrast. <P>SOLUTION: A background difference calculation unit 6 extracts a difference between a background image and a current image. A binarization unit 8 binarizes a difference value between the background image and the current image. A recognition processing unit 11 detects whether an event has occurred or not by comparing a characteristic value which is binarized at the binarization unit 8 and calculated at a characteristic value calculation unit 9 with a preset characteristic value parameter value 10. A frequency expansion unit 14 calculates a spatial frequency of the current image and a frequency component extraction unit 15 extracts predetermined frequency components. A correction performing unit 16 corrects operating parameters in the background difference calculation unit 6 based on the frequency components extracted at the frequency component extraction unit 15. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a monitoring image processing apparatus that extracts a change area from an image obtained by sequentially photographing a monitoring target area and detects the occurrence of an event based on the change area.

Conventionally, there has been a monitoring device that extracts a change area from a time-series image obtained by sequentially capturing a monitoring target area, determines whether or not an event has occurred based on the change area, and issues a notification when an event occurs. (For example, refer to Patent Document 1).
In such an image processing apparatus used for a monitoring apparatus, a sequentially acquired image is accumulated as a current image, and a new background image is created by performing arithmetic processing with a past background image, and this is accumulated as a background image. To do. Then, difference data between the current image and the background image is created, binarized to calculate the feature value, and when the feature value satisfies a predetermined set value, the monitoring device generates an event occurrence. Was configured to report.

Japanese Patent No. 4065137

  In the conventional image processing apparatus as described above, the feature amount is calculated based on the change area. The change area is created by the multi-value difference data and the binarization threshold value, and the basis of the calculation is multi-value difference data. The multi-value difference data is data that is obtained when there is a difference between two images (the current image and the background image). However, in the following situation, it is difficult for the two images to have a difference. This is called a “low contrast” situation.

Situation 1: Contrast is lowered due to darkness due to insufficient lighting → Both the current image and the background image are dark. For example, even in a scene where a person walks on the ground, the difference data of the person is not sufficiently calculated, and the change area is It does not become a silhouette of a beautiful person, but a silhouette that collapses small.
Situation 2: To compensate for the lack of lighting, the camera uses an electronic sensitization function to lower the contrast. → The outline of both the current image and the background image is blurred against moving objects, and people are walking even when people are walking on the ground. The difference data is not sufficiently calculated, and the change area does not become a beautiful silhouette of a person, but becomes a silhouette collapsed small.

As described above, when the contrast is low, the difference becomes small. When the difference is small, the change area cannot be accurately cut out even in the subsequent binarization. Furthermore, if the change region cannot be obtained with high accuracy, the accuracy in the subsequent feature amount calculation also deteriorates. In that case, as a result, the feature amount calculated therefrom is also affected, and the area and vertical and horizontal dimensions tend to be smaller than those of an actual pedestrian.
As described above, in the conventional image processing apparatus, the above-mentioned tendency cannot be eliminated, which is a major cause of deterioration in accuracy of the recognition process. As a result, the monitoring device also has a drawback that the report is inaccurate.

  The present invention has been made to solve the above-described problems, and provides a monitoring image processing apparatus capable of accurately detecting an event even if a captured image is a low-brightness and low-contrast image. The purpose is to obtain.

  The monitoring image processing apparatus according to the present invention extracts a change area in the monitoring target area from a plurality of images in video data obtained by sequentially capturing the monitoring target area based on a predetermined operation parameter, and based on the change area. Image recognition means for detecting the presence or absence of an event, a frequency expansion unit for calculating a spatial frequency of an image obtained by imaging a monitoring target region, and a frequency element for extracting a predetermined frequency element from the spatial frequency calculated by the frequency expansion unit Based on the frequency element extracted by the frequency element extraction unit and the extraction unit, the situation of the monitoring target region is inferred, and the correction execution unit that corrects the operation parameter in accordance with the analogy result is provided. is there.

  The monitoring image processing apparatus according to the present invention calculates the spatial frequency of the image of the monitoring target region, analogizes the situation of the monitoring target region with frequency elements based on the spatial frequency, and corresponds to the analogy result of the image recognition means. Since the operation parameters are corrected, even when the captured image is a low-brightness and low-contrast image, event detection can be performed with high accuracy.

1 is a configuration diagram showing a monitoring image processing apparatus according to Embodiment 1 of the present invention; FIG. It is explanatory drawing which shows an example of the present image of the monitoring image processing apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows an example of the present image which accumulated the change area | region of the monitoring image processing apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows the example of calculation of the spatial frequency of the monitoring image processing apparatus by Embodiment 1 of this invention. It is a flowchart which shows operation | movement of the recognition process part of the monitoring image processing apparatus by Embodiment 1 of this invention. It is a block diagram which shows the monitoring image processing apparatus by Embodiment 2 of this invention. It is a block diagram which shows the monitoring image processing apparatus by Embodiment 3 of this invention.

Embodiment 1 FIG.
FIG. 1 is a block diagram of a monitoring apparatus to which a monitoring image processing apparatus according to Embodiment 1 of the present invention is applied.
1 includes a video input unit 1, an A / D converter 2, a current image storage unit 3, a background image storage unit 4, a background update unit 5, a background difference calculation unit 6, a threshold value calculation unit 7, and binarization. 8, a feature amount calculation unit 9, a feature amount parameter setting value 10, a recognition processing unit 11, a notification processing unit 13, a frequency expansion unit 14, a frequency element extraction unit 15, and a correction execution unit 16.

  The video input unit 1 is an input interface for inputting video data captured by an imaging device (not shown). The A / D converter 2 is a conversion unit that quantizes video data that is an analog signal input from the video input unit 1 and converts the image data into a digital signal. The current image storage unit 3 is a data storage unit that inputs and stores current video data (hereinafter referred to as current image data) converted into a digital signal by the A / D converter 2. The background image storage unit 4 is a data storage unit that performs a predetermined calculation on the current image data input from the A / D converter 2 to generate and store background image data for comparison. The background update unit 5 is a data update unit that performs a predetermined operation on the background image data stored in the background image storage unit 4 to update the data content (background). For example, the average of the pixel values of the current image data corresponding to the past five scenes is used as the updated background image data.

  The background difference calculation unit 6 is a calculation unit that compares the image data stored in the current image storage unit 3 and the background image storage unit 4 and calculates the difference image data. The threshold value calculation unit 7 is a calculation unit that calculates a threshold value for binarizing the difference image data calculated by the background difference calculation unit 6. In general, an optimum threshold value is obtained by a simple rule based on difference image data, and for example, an average of pixel values of difference image data is set as the threshold value. The binarization unit 8 is a calculation unit that binarizes the difference image data with the threshold obtained by the threshold calculation unit 7. The feature amount calculation unit 9 is a calculation unit that receives the binary data calculated by the binarization unit 8 and calculates a feature amount. The parameters that define the feature amount include, for example, the area of the change area indicating the pixel area that has changed over time in the image, the vertical and horizontal dimensions, and the like.

  The feature amount parameter setting value 10 is a feature amount parameter setting value (predetermined setting value), which is a parameter value of each feature amount that is set in advance under a predetermined condition that defines a notification target. The feature parameter setting value 10 is stored in a storage device that can be appropriately read out by an arithmetic processing unit of a computer device functioning as the recognition processing unit 11. The recognition processing unit 11 is a determination unit that determines whether an event has occurred by comparing the feature amount data calculated by the feature amount calculation unit 9 with the feature amount parameter setting value 10 to be notified. The recognition processing unit 11 is configured to output an event detection signal 12 when an event occurrence is detected.

  In addition, the background difference calculation unit 6 to the recognition processing unit 11 extract a change region in the monitoring target region from a plurality of images in video data obtained by sequentially capturing the monitoring target region based on predetermined operation parameters, Image recognition means for detecting the presence or absence of the occurrence of an event based on the change area is configured.

  When the event detection signal 12 is output from the recognition processing unit 11, the notification processing unit 13 is a processing unit that issues a notification to the operator in response to the output. As a reporting method, for example, the reporting processing unit 13 may be configured together with an appropriate sound source, and a beep sound may be generated for reporting.

  The frequency expansion unit 14 is a calculation unit that calculates the spatial frequency of the image quantized by the A / D converter 2. The frequency element extraction unit 15 is a calculation unit that obtains a spatial frequency element such as a spatial frequency, a spatial frequency fluctuation period, and a spatial frequency fluctuation width. The correction execution unit 16 corrects the operation parameter in the background difference calculation unit 6 based on the frequency element extracted by the frequency element extraction unit 15. The correction value table 17 is a table showing the correction values of the operation parameters.

  The monitoring image processing apparatus of the present invention is realized by using a computer, and the background update unit 5 to the feature amount calculation unit 9 and the recognition processing unit 11 to the correction execution unit 16 are configured with software corresponding to each function. Alternatively, it may be configured by hardware such as a CPU or memory for executing these software. Alternatively, either configuration may be configured with dedicated hardware.

Next, the operation of the monitoring image processing apparatus according to the first embodiment will be described.
The camera image is captured by the video input unit 1 and converted into digital data by the A / D converter 2, and the current image storage unit 3 stores it as “current image”. At the same time, the background update unit 5 performs an arithmetic process on the current image with the past background image (for example, averages the background for the past n seconds), creates a new background image, and sends the background image storage unit 4 a new “ Accumulated as “background image”. The background difference calculation unit 6 performs a difference calculation between the “current image” and the “background image” to create difference data. This is not a simple difference, but it is general to add so-called noise removal, in which a small difference is judged as noise and removed.
The difference data is multi-value data at this point, but is binarized by the binarization unit 8 to become binary data. The threshold value for binarization is generally calculated using the “difference data” in the threshold value calculation unit 7 (example: a peak is obtained from a histogram of difference data, and m value is subtracted from there. In some cases, the value is a threshold value).
The binarized difference data is indicated by 1, 0. When the pixels “1” are aligned and overlapped on the screen, a silhouette of a “person” is obtained as shown in FIG. 3 in a scene where a person is walking on a flat background as shown in FIG. This silhouette image is generally called a change area.

  On the other hand, the frequency developing unit 14 divides the image into small blocks (for example, 16 × 16 pixels), and calculates the spatial frequency of the image therein. For example, as shown in FIG. 4, in one block (eg, 16 × 16 pixels), a certain line (16 pixels) is taken out, and the luminance value is plotted on a graph. This is a graph in which a single line is drawn in the coordinate system with the X axis of 1 to 16 and the Y axis of 0 to 255. If this graph is expanded into a Fourier series and a series whose shape matches 90% or more is obtained, it indicates the spatial frequency of the corresponding line (16 pixels). If this operation is performed on all the lines in the block and the most series is selected, it indicates the spatial frequency of the block (16 × 16 pixels), etc. (the above is an example of how to determine the spatial frequency) Is).

The frequency element extraction unit 15 stores the spatial frequency obtained in each block for a long time (eg, 48 hours), observes the variation of the spatial frequency during that time, calculates the frequency element of each block, and stores it. Examples of frequency elements include the following.
(1) Spatial frequency (the above-mentioned spatial frequency itself)
(2) Spatial frequency fluctuation period (if there is a cycle in fluctuation during the above storage period, the cycle period)
(3) Spatial frequency fluctuation width (variation width when viewing the spatial frequency of all blocks in the screen simultaneously)
(4) Spatial frequency peak value (maximum value when viewing the spatial frequency of all blocks in the screen at the same time)

First, the correction execution unit 16 looks at the frequency elements and infers the situation of each block of the video.
For example, (1) the spatial frequency has a correlation that decreases when the contrast in the block is low and increases when the contrast is high. If the spatial frequency is low, the block is “dark due to insufficient illuminance”, “it is dark due to insufficient illuminance, so the electronic sensitization is turned on and it looks bright but the motion is blurred” It can be inferred that it is either “one”.
Further, (4) the spatial frequency peak value indicates the maximum value of the spatial frequency in the block in the screen, and is the value of the block that is viewing the most complicated landscape. If this value is high, even if there are many blocks with a low spatial frequency, if we discuss it with the entire field of view of the camera, it is actually dark because of insufficient illuminance and dark due to insufficient illuminance. It can be inferred that the movement is not blurred.
In addition, (2) the spatial frequency fluctuation period can be predicted that if the period is similar to 24 hours, the camera observes the outdoors, and if it is irregular, the camera observes the indoors. This is not observed in each block, and a plurality of representative blocks may be selected and measured from the blocks on the entire screen.
(3) The spatial frequency fluctuation width is large if there is a spatial frequency difference between the blocks in the screen, and is small if there is no difference. If the spatial frequency fluctuation range is low, the area viewed by the camera can be predicted to be uniform in complexity (eg, only the floor surface is viewed), and conversely if the spatial frequency fluctuation range is high, the camera can see it. Area can vary in complexity (eg, roads, trees and houses are visible).

Specific examples will be shown below.
<Example 1>
Observed for 24 hours,
(A) The spatial frequency is always low (b) The spatial frequency fluctuation period is unclear (c) The spatial frequency fluctuation width is always small (d) The spatial frequency peak value is always low In such a case, (a) (d ), The corresponding block is “dark due to insufficient illumination”, “because it is dark due to insufficient illumination, the electronic sensitization is turned on, it looks bright but the motion is blurred”, “the scenery of the visible block happens to be single”, etc. It can be inferred that this is the situation.
From (b), it can be inferred that the camera field of view is indoors because the illumination fluctuation is small.
Furthermore, it can be inferred from (c) that the camera field of view is capturing a flat landscape. (Example: indoor passage etc.)
From the above, it can be inferred that this camera is imaging a flat landscape such as an indoor passage.

<Example 2>
Observed for 24 hours,
(A) Spatial frequency fluctuates periodically (b) Spatial frequency fluctuation period tends to be high around 24 hours and near noon (c) Spatial frequency fluctuation range is large during daytime and small during nighttime ( d) The peak value of the spatial frequency is high during the daytime period and small during the nighttime period. In this case, it can be inferred from this (b) that the camera is capturing the outdoors.
In addition, from (c), the camera captures a flat place and a complicated place. (Example: Outdoor floor and branches and leaves of trees)
In addition, from (a) and (d), the illuminance of the field of view during the day is sufficient, but the illuminance at night is insufficient, and it is "dark due to insufficient illuminance". It can be inferred that the situation is “I can see it but the movement is blurry”.
→ From the above, this camera can be analogized as "capturing a complex landscape including outdoor floors and trees, and no lighting at night."

<Example 3>
Observed for 24 hours,
(A) Spatial frequency fluctuates periodically (b) Spatial frequency fluctuation period tends to be high around 24 hours and near noon and at night (c) Spatial frequency fluctuation range is always small (d) Spatial frequency peak value is day Medium time zone is small, night time zone is large. In such a case, it can be inferred from (b) that this camera is shooting outdoors.
Also, from (c), the camera is photographing a flat place. (Example: Outdoor floor, etc.)
Furthermore, from (a) and (d), it can be inferred that the illuminance of the visual field during the day is insufficient, but the illuminance at night is sufficient. Therefore, it can be inferred that there is no situation where “dark due to insufficient illumination” or “dark due to insufficient illumination, so that electronic sensitization is turned on and it appears bright but the motion is blurred”.
From the above, it can be inferred that this camera “captures an image of a flat landscape such as an outdoor floor and is well lit at night”.

As described above, in this embodiment, it is possible to analogize the installation environment of the camera and the scenery to be imaged by observing the frequency element for a long time.
The installation environment of the camera and the scenery to be imaged directly affect the images that can be acquired from the camera. If it is outdoors, there will be a drastic change in the image of day and night, and if it is insufficient, the contrast at night will be reduced.
However, in the conventional apparatus, there is no such information on “camera installation environment and scenery to be imaged”, and therefore it is only possible to perform common image processing for any video.
Since this apparatus has means for analogizing “the camera installation environment and the scene to be imaged”, it is possible to predict these effects and correct the image processing.

When the correction execution unit 16 grasps the situation of “the camera installation environment and the scene to be imaged” as described above, the correction execution unit 16 refers to the correction value table 17 and performs correction according to the situation.
In this case, the correction may be changed for each block, or the same correction may be performed on the entire screen. In the correction value table 17, a plurality of correction patterns are registered in advance according to combinations of frequency elements. The correction value corrects the operation parameter in the background difference calculation unit 6. An example is shown below.

[Example 1]: It is predicted that an image of a flat place such as an indoor passage is taken. → Since the fluctuation in illuminance is small, there is a high possibility that even a small difference value that is usually determined as noise is not noise. Therefore, the background difference calculation unit 6 sets operation parameters so as not to remove noise.
[Example 2]: An image of a complex landscape including an outdoor floor and trees is taken, and it is expected that there will be no illumination at night. → Since it is outdoors, there is no problem because the illuminance is sufficient during the day.
→ Because there is no night lighting, it is difficult to extract differences at night. At night when the spatial frequency is lowered, the background difference calculation unit 6 sets an operation parameter so as to perform contour enhancement processing on the current image and the background image, and then calculates the difference to compensate for the lack of contrast at night.
[Example 3]: An image of a flat landscape such as an outdoor floor surface is taken, and it is expected that the illumination is sufficient at night. → Although it is outdoors, the illumination at night is solid and the illuminance is sufficient. No problem.
→ It is a shadow that the sunshine does not reach, and it is difficult to extract the difference during the day depending on the weather. In the daytime when the spatial frequency is lowered, the background difference calculation unit 6 performs contour enhancement processing on the current image and the background image, and then calculates the difference to compensate for the lack of contrast.

  As described above, the frequency expansion unit 14 to the correction execution unit 16 appropriately compensate for the lack of contrast, and as a result, the changed region is restored to a shape close to the original silhouette. This is a result of alleviating the problem that until then the change area could not be cut out with low contrast and accuracy. As a result, the various feature values calculated by the feature value calculation unit 9 take an approximate value with the situation in which there is sufficient contrast and the change area is cut out with high accuracy. Since the area of the change area, vertical and horizontal dimensions, movement vector, and the like are similar to the original values, deterioration of recognition accuracy can be prevented.

These feature quantities are compared with the feature quantity parameter setting value 10 in the recognition processing unit 11, and if the feature quantity corresponds to the set value, an event detection signal 12 indicating the occurrence of an event is output.
FIG. 5 is a flowchart showing the operation of the recognition processing unit 11.
First, when the recognition processing unit 11 acquires feature amount data from the feature amount calculation unit 9 (step ST1), the recognition processing unit 11 determines whether or not the “continuity” matches the feature amount parameter setting value 10 (step ST2). ). For the determination of “continuity”, for example, a continuous time indicating how many frames the change area in the difference image data to be processed survives in the video is used. Specifically, when the range of the feature parameter setting value 10 is “minimum 4 frames, maximum 10 frames”, it is determined that the change areas match if the continuous time in which the change area survives is 4 to 10 frames. Otherwise, it is determined that it does not match the range of the feature parameter setting value 10. Here, if it is determined that the value matches the range of the feature parameter setting value 10, the process proceeds to step ST3, and if not, the process ends.

  In step ST <b> 3, the recognition processing unit 11 determines whether or not the “area” of the feature amount data obtained by the feature amount calculation unit 9 matches the feature amount parameter setting value 10. For the determination of the “area”, for example, the number of pixels in the binary change area of the difference image data is used. Specifically, when the range of the feature parameter setting value 10 is “minimum 500 pixels, maximum 1000 pixels”, it is determined that the binary change area matches if the area is 500 to 1000 pixels. Otherwise, it is determined that it does not match the range of the feature parameter setting value 10. Here, if it is determined that it matches the range of the feature parameter setting value 10, the process proceeds to step ST4, and if not, the process ends.

  Next, the recognition processing unit 11 determines whether or not the “vertical and horizontal dimensions” of the feature amount data obtained by the feature amount calculating unit 9 matches the feature amount parameter setting value 10 (step ST4). For the determination of the “vertical and horizontal dimensions”, for example, the vertical length or horizontal length of a circumscribed rectangle that defines a binary change region is used. Specifically, when the range of the feature parameter setting value 10 is “minimum vertical 50 pixels, maximum 100 pixels” and “horizontal minimum 50 pixels, maximum 100 pixels”, both the vertical and horizontal dimensions of the binary change area are 50 to 50. If it is 100 pixels, it is determined that they match. Otherwise, it is determined that it does not match the range of the feature parameter setting value 10. Here, if it is determined that it matches the range of the feature parameter setting value 10, the process proceeds to step ST5, and if not, the process ends.

  Subsequently, the recognition processing unit 11 determines whether or not the “speed” of the feature amount data obtained by the feature amount calculating unit 9 matches the feature amount parameter setting value 10 (step ST5). For the determination of the “speed”, for example, the moving speed in the video of the change area is used. Specifically, when the range of the feature parameter setting value 10 is “minimum 50 pixels / second, maximum 100 pixels / second”, the moving speed of the binary change area in the video may be 50 to 100 pixels / second. Are determined to match. Otherwise, it is determined that it does not match the range of the feature parameter setting value 10. Here, if it is determined that the value matches the range of the feature parameter setting value 10, the process proceeds to step ST6, and if not, the process ends.

  The above-mentioned “area”, “vertical / horizontal dimensions”, and “moving speed” are obtained by calculating the average value of the feature values of each image during the lifetime of the binary change area used for the determination in step ST2. It is common to compare with the quantity parameter set value of 10.

In step ST <b> 6, the recognition processing unit 11 determines that it matches all the feature parameter setting values 10 that specify the regular reporting target, and outputs an event detection signal 12 to the reporting processing unit 13. In addition, the process from step ST1 to step ST6 is realizable by executing the software which has the function of the recognition process part 11 with a computer apparatus.
Upon receiving the event detection signal 12 from the recognition processing unit 11, the reporting processing unit 13 performs a reporting process.

  As described above, according to the monitoring image processing apparatus of the first embodiment, the change area in the monitoring target area is determined based on the predetermined operation parameter from the plurality of images in the video data obtained by sequentially capturing the monitoring target area. An image recognition means for extracting and detecting the occurrence of an event based on the change area, a frequency expansion section 14 for calculating a spatial frequency of an image obtained by photographing the monitoring target area, and a spatial frequency calculated by the frequency expansion section 14 Based on the frequency element extracted from the frequency element extraction unit 15 and the frequency element extracted by the frequency element extraction unit 15, the situation of the monitoring target area is inferred, and the operation parameter And a correction execution unit 16 that performs correction, so that even when the captured image is a low-brightness and low-contrast image, event detection can be performed with high accuracy. .

  Further, according to the monitoring image processing apparatus of the first embodiment, the image recognition unit includes a background difference calculation unit 6 that calculates a difference between the background image of the monitoring target area and the current image based on a predetermined operation parameter; A binarization unit 8 that binarizes the difference value calculated by the background difference calculation unit 6 and extracts a change region, a feature amount of a monitoring target region based on the change region, and a preset value of a feature amount And the recognition processing unit 11 that determines whether or not an event has occurred in the monitoring target region, and the correction execution unit 16 corrects the operation parameter, so that the captured image is a low-brightness, low-contrast image. Even if it exists, the difference of a background image and the present image can be calculated | required correctly, As a result, an event detection can be performed with sufficient precision.

Embodiment 2. FIG.
FIG. 6 is a configuration diagram of the monitoring image processing apparatus according to the second embodiment.
In the second embodiment, the correction execution unit 16a corrects the threshold value in the threshold value calculation unit 7 as the correction of the operation parameter. That is, in the correction value table 17a held by the correction execution unit 16a, the threshold correction value in the threshold calculation unit 7 is registered, and the correction execution unit 16a corrects the threshold calculation unit 7 based on the correction value. Do. Since the other configuration is the same as that of the first embodiment shown in FIG.

Next, the operation of the second embodiment will be described.
Here, the basic operation from the video input unit 1 to the notification processing unit 13 and the operation from the frequency expansion unit 14 to the frequency element extraction unit 15 are the same as those in the first embodiment, and thus description thereof is omitted here. To do.

When the correction execution unit 16a grasps the status of “camera installation environment and imaging target landscape” based on the frequency elements extracted by the frequency element extraction unit 15, the correction execution unit 16a refers to the correction value table 17a and corresponds to the situation. Make corrections. Here, regarding the correction method, the correction may be changed for each block, or the same correction may be performed on the entire screen.
In the correction value table 17a, a plurality of correction patterns are registered in advance according to combinations of frequency elements, and the correction execution unit 16a corrects the threshold value of the threshold value calculation unit 7 using the correction pattern. Examples are shown below.

[Example 1]: It is predicted that an image of a flat place such as an indoor passage is taken. → Since the fluctuation in illuminance is small, there is a high possibility that even a small difference value that is usually determined as noise is not noise. Therefore, the threshold value calculation unit 7 obtains the threshold value by a certain amount smaller than usual so that the change area can be cut out larger.
[Example 2]: An image of a complex landscape including an outdoor floor and trees is taken, and it is expected that there will be no illumination at night. → Since it is outdoors, there is no problem because the illuminance is sufficient during the day.
→ Because there is no night lighting, it is difficult to extract differences at night. At night when the spatial frequency is lowered, the threshold value calculation unit 7 calculates the threshold value by a certain amount smaller than usual so that the change area can be cut out larger, and compensates for the lack of contrast.
[Example 3]: An image of a flat landscape such as an outdoor floor surface is taken, and it is expected that the illumination is sufficient at night. → Although it is outdoors, the illumination at night is solid and the illuminance is sufficient. No problem.
→ It is a shadow that the sunshine does not reach, and it is difficult to extract the difference during the day depending on the weather. During the day when the spatial frequency is lowered, the threshold value calculation unit 7 obtains the threshold value by a certain amount smaller than usual so that the change area can be cut out larger, thereby compensating for the lack of contrast.

  As described above, as a result of appropriately compensating for the lack of contrast by the frequency expansion unit 14 to the correction execution unit 16a, the changed region is restored to a shape close to the original silhouette. This is a result of alleviating the problem that until then the change area could not be cut out with low contrast and accuracy. As a result, the various feature values calculated by the feature value calculation unit 9 take an approximate value with the situation in which there is sufficient contrast and the change area is cut out with high accuracy. Since the area of the change area, vertical and horizontal dimensions, movement vector, and the like are similar to the original values, deterioration of recognition accuracy can be prevented.

  In addition, the various feature amounts are compared with the feature amount parameter setting values stored in advance in the feature amount parameter setting value 10 in the recognition processing unit 11, and if they correspond to the setting value, the event detection signal 12 indicating event detection. Is output. The processing in the recognition processing unit 11 is as shown in FIG. When the event detection signal 12 is received from the recognition processing unit 11, the notification processing unit 13 performs a process reporting process.

  As described above, according to the monitoring image processing apparatus of the second embodiment, the image recognition unit includes the background difference calculation unit 6 that calculates the difference between the background image of the monitoring target area and the current image, and the background difference calculation unit. Based on the difference value calculated in 6, the threshold value calculation unit 7 that calculates a threshold value for binarization as a predetermined operation parameter and the difference value calculated in the background difference calculation unit 6 are binarized with the threshold value. A binarization unit 8 for extracting a change area, and a recognition for determining whether or not an event has occurred in the monitoring target area by comparing a feature amount of the monitoring target area based on the changing area with a preset value of the feature quantity Since the correction execution unit 16a corrects the threshold value calculated by the threshold value calculation unit 7, even if the captured image is a low-brightness and low-contrast image, the threshold value corresponding to this image is provided. And the result can be Accuracy can be an event detection.

Embodiment 3 FIG.
FIG. 7 is a configuration diagram of the monitoring image processing apparatus according to the third embodiment.
In the third embodiment, the correction execution unit 16b corrects the feature parameter setting value 10 as the operation parameter correction. That is, the correction value table 17b held by the correction execution unit 16b stores the correction value of the feature parameter setting value 10, and the correction execution unit 16b corrects the feature parameter setting value 10 based on the correction value. I do. Since other configurations are the same as those of the first embodiment shown in FIG. 1, the corresponding parts are denoted by the same reference numerals and description thereof is omitted.

Next, the operation of the third embodiment will be described.
Here, the basic operation from the video input unit 1 to the notification processing unit 13 and the operation from the frequency expansion unit 14 to the frequency element extraction unit 15 are the same as those in the first embodiment, and thus description thereof is omitted here. To do.
When the correction execution unit 16b grasps the situation of the “camera installation environment and the scene to be imaged” based on the frequency element extracted by the frequency element extraction unit 15, the correction execution unit 16b refers to the correction value table 17b and performs correction according to the situation. Do. In this correction, the correction may be changed for each block, or the same correction may be performed on the entire screen.
In the correction value table 17b, a plurality of correction patterns are registered in advance according to combinations of frequency elements, and the correction execution unit 16b corrects the feature parameter setting value 10 using the correction patterns. Examples are shown below.

[Example 1]: It is predicted that an image of a flat place such as an indoor passage is taken. → Since the fluctuation in illuminance is small, there is a high possibility that even a small difference value that is usually determined as noise is not noise. Therefore, the feature value parameter setting value 10 is switched so that the setting values for the area determination and the vertical / horizontal determination are acceptable even in a smaller change area.
[Example 2]: An image of a complex landscape including an outdoor floor and trees is taken, and it is expected that there will be no illumination at night. → Since it is outdoors, there is no problem because the illuminance is sufficient during the day.
→ Because there is no night lighting, it is difficult to extract differences at night. At night when the spatial frequency is lowered, the feature parameter setting value is switched so that the setting value of the area determination and the vertical / horizontal determination is acceptable even in a smaller change area, and the lack of contrast is compensated.
Example 3: An image of a flat landscape such as an outdoor floor surface is captured, and sufficient illumination is expected at night. → Although it is outdoors, the illumination at night is solid and the illumination is sufficient. No problem.
→ It is a shadow that the sunshine does not reach, and it is difficult to extract the difference during the day depending on the weather. In the daytime when the spatial frequency is lowered, the feature value parameter setting value 10 is switched so that the setting value of the area determination and the vertical / horizontal determination is acceptable even in a smaller change area, and the lack of contrast is compensated.

  As described above, as a result of appropriately compensating for the lack of contrast by the frequency expansion unit 14 to the correction execution unit 16b, the changed region is restored to a shape close to the original silhouette. This is a result of alleviating the problem that until then the change area could not be cut out with low contrast and accuracy. As a result, the various feature values calculated by the feature value calculation unit 9 take an approximate value with the situation in which there is sufficient contrast and the change area is cut out with high accuracy. Since the area of the change area, vertical and horizontal dimensions, movement vector, and the like are similar to the original values, deterioration of recognition accuracy can be prevented.

  These feature amounts are compared with the corrected feature amount parameter setting value 10 in the recognition processing unit 11 and, if the feature amount falls within the set value, an event detection signal 12 indicating event detection is output. The recognition processing unit 11 is as shown in FIG. When the event detection signal 12 is received from the recognition processing unit 11, the notification processing unit 13 performs a process reporting process.

  In the third embodiment, the setting values for area determination and vertical / horizontal determination are corrected as the correction of the feature parameter setting value 10 in the correction execution unit 16b. However, other operation parameter corrections such as continuity and speed are also corrected. May be performed.

  As described above, according to the monitoring image processing apparatus of the third embodiment, the image recognition unit includes the background difference calculation unit 6 that calculates the difference between the background image of the monitoring target area and the current image, and the background difference calculation unit. The binarization unit that binarizes the difference value calculated in 6, the feature amount of the monitoring target region based on the binarized data, and a predetermined set value of the feature amount are compared, and the monitoring target region And a recognition processing unit 11 that determines whether or not an event has occurred in the image processing apparatus, and the correction processing unit corrects the setting value of the feature amount. Even if the captured image is a low-contrast image with low brightness, It is possible to set the feature parameter setting value according to the image, and as a result, it is possible to detect the event with high accuracy.

  In the first to third embodiments, the background difference calculation unit 6, the threshold value calculation unit 7, and the feature parameter setting value 10 are corrected. However, the correction may be performed in combination. .

  DESCRIPTION OF SYMBOLS 1 Video input part, 2 A / D converter, 3 Current image storage part, 4 Background image storage part, 5 Background update part, 6 Background difference calculating part, 7 Threshold value calculating part, 8 Binarization part, 9 Feature-value calculating part 10 Feature parameter setting value, 11 Recognition processing unit, 12 Event detection signal, 13 Notification processing unit, 14 Frequency expansion unit, 15 Frequency element extraction unit, 16, 16a, 16b Correction execution unit, 17, 17a, 17b Correction Value table.

Claims (4)

An image for extracting a change area in the monitoring target area from a plurality of images in video data obtained by sequentially capturing the monitoring target area based on a predetermined operation parameter, and detecting the occurrence of an event based on the change area Recognition means;
A frequency expansion unit for calculating a spatial frequency of an image obtained by photographing the monitoring target region;
A frequency element extraction unit that extracts a predetermined frequency element from the spatial frequency calculated by the frequency expansion unit;
Based on the frequency element extracted by the frequency element extraction unit, the situation of the monitoring target area is inferred, and a correction execution unit that corrects the operation parameter according to the analogy result is provided. Processing equipment. The image recognition means binarizes the difference value calculated by the background difference calculation unit and a background difference calculation unit that calculates a difference between the background image of the monitoring target area and the current image based on a predetermined operation parameter. A binarization unit for extracting a region, and a feature amount of the monitoring target region based on the change region is compared with a set value of a predetermined feature amount to determine whether or not an event has occurred in the monitoring target region A recognition processing unit,
The monitoring image processing apparatus according to claim 1, wherein the correction execution unit corrects the operation parameter. The image recognition means has a background difference calculation unit for calculating a difference between the background image of the monitoring target area and the current image, and a threshold value for binarization is determined based on the difference value calculated by the background difference calculation unit. A threshold value calculation unit that calculates as an operation parameter, a binarization unit that binarizes the difference value calculated by the background difference calculation unit with the threshold value and extracts a change region, and the monitoring target region based on the change region A recognition processing unit that determines whether or not an event has occurred in the monitoring target region by comparing the feature amount and a set value of a predetermined feature amount,
The monitoring image processing apparatus according to claim 1, wherein the correction execution unit corrects a threshold value calculated by the threshold value calculation unit. The image recognition means includes a background difference calculation unit that calculates a difference between the background image of the monitoring target area and the current image, a binarization unit that binarizes the difference value calculated by the background difference calculation unit, A recognition processing unit that determines whether or not an event has occurred in the monitoring target region by comparing a feature amount of the monitoring target region based on the converted data and a set value of a predetermined feature amount;
The monitoring image processing apparatus according to claim 1, wherein the correction processing unit corrects a set value of the feature amount.

Images