JP3824934B2 - Image processing device for monitoring - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a monitoring image processing apparatus having an improved region integration method.
[0002]
[Prior art]
FIG. 1 is a block diagram showing a conventional monitoring image processing apparatus. In FIG. 1, 1 is an image input unit for inputting a video signal, 2 is a current image storage unit for storing a current image based on the video signal, and 3 is a video signal. A background image accumulating unit for accumulating a background image according to 4 obtains a background image by sequentially processing video signals, and a background updating unit for accumulating the updated background image in the background image accumulating unit 3. It is a background difference unit that extracts a change area of the current image from the difference by comparison with an image.
6 is a threshold value calculation unit for calculating a threshold value for binarizing the change region of the current image, 7 is a binarization unit for binarizing the change region of the current image with the threshold value, and 8 is a block for the binary data. A labeling unit that assigns a label to each label and 9 is a feature quantity computing unit that computes a feature quantity for each numbered label.
Reference numeral 10 denotes an area integration unit that integrates change areas that are considered to belong to the same change target for each numbered label and combines them into one change area. 11 is a preset feature value to be reported and 12 is stored, 12 is a recalculation of the feature value of the change area integrated by the area integration unit 10, and the recalculated feature value is calculated by the feature value calculation unit 9. A recognition processing unit that compares the feature quantity of the changed area that has not been integrated by the area integration unit 10 and the notification target feature quantity 11 to determine whether they match, and 13 indicates that the recognition processing unit 12 determines a match. This is a notification processing unit that issues a notification.
FIG. 6 is a flowchart showing the process of the conventional region integration unit, and FIG. 7 is an explanatory diagram showing the process of the conventional region integration unit.
[0003]
Next, the operation will be described.
In FIG. 1, the video signal from the surveillance camera is input to the image input unit 1 and after A / D conversion, for example, the current image storage unit 2 and the background as 8-bit image data for each of R, G, and B This is supplied to the image storage unit 3. The current image storage unit 2 stores the image data. The background image storage unit 3 supplies the image data to the background update unit 4, and the background update unit 4 calculates, for example, a background image from the average of the past 30 image data and stores it in the background image storage unit 3. The background image to be updated is sequentially updated. Here, the reason why the background image is sequentially updated is to cope with the fact that the scenery of the area captured by the surveillance camera differs between day and night or due to movement of an object. The background difference unit 5 compares the current image of the current image storage unit 2 with the background image of the background image storage unit 3 and extracts a change area such as a moving object or a change object in the current image from the difference.
The threshold value calculation unit 6 calculates a threshold value for binarizing (1 bit) the background difference data (8 bits) from the background difference unit 5. In general, an optimum threshold value is obtained based on background difference data. The binarization unit 7 binarizes the background difference data with the threshold value obtained by the threshold value calculation unit 6. The labeling unit 8 numbers the binary data output from the binarizing unit 7 for each block. Here, the block of binary data means that adjacent data “1” pixels are regarded as one block. Before the region integration shown in FIG. 7, there are three groups of binary data, but here, each is numbered. The feature amount calculation unit 9 calculates the feature amount for each label numbered by the labeling unit 8. Here, the feature amount refers to each set of binary data, that is, the area of each change region, vertical and horizontal dimensions, and the like.
[0004]
The area integration unit 10 integrates the change areas that are considered to belong to the same change target for each label numbered by the labeling unit 8 and combines them into one change area.
6 and 7 describe the processing of the region integration unit 10 in detail, and will be described below with reference to FIGS. 6 and 7.
First, the ferret diameter determination unit ST1 determines the ferret diameter obtained from the labeling unit 8. Here, the ferret diameter is a circumscribed rectangle of the change region, as shown in FIG. Since the labeling unit 8 assigns labels to each change region, the total number of ferret diameters can be obtained from the labeling unit 8. Before the region integration shown in FIG. 7, the ferret diameter is 3, and in this case, the processing of steps ST2 to ST6 is performed three times. The processes of steps ST2 to ST6 are performed for the total number of ferret diameter combinations. When the ferret diameter number is n, the processes are performed _n C ₂ times.
Thereafter, the following integration determination is performed between the ferret diameters.
(Conventional condition 1) The inclusion determination unit ST2 determines whether one ferret diameter is completely included in the other ferret diameter. If it is included, both ferret diameters are integrated.
(Conventional condition 2) The contact determination unit ST3 determines whether one ferret diameter is in contact with the other ferret diameter. When they are in contact, both ferret diameters are integrated.
(Conventional condition 3) In the midpoint coordinate calculation unit ST4, the midpoint coordinates of the ferret diameter are calculated, and in the virtual ferret diameter calculation unit ST5, the midpoint is the same, double in the vertical direction and double in the horizontal direction (multiple can be changed by setting). The virtual ferret diameter is calculated. In the virtual ferret diameter determination unit ST6, when the midpoint of the other ferret diameter is included in the virtual ferret diameter, the two ferret diameters are integrated.
FIG. 7 shows a state in which the ferret diameter is changed from before the region integration to after the region integration by the processing of the region integration unit 10.
[0005]
The recognition processing unit 12 recalculates the feature amount of the change area of the ferret diameter integrated by the region integration unit 10, calculates the recalculated feature amount, and the feature amount calculation unit 9, and integrates it by the region integration unit 10. The feature amount of the change area of the ferret diameter that has not been performed is compared with the feature amount 11 to be notified of, for example, a person or a vehicle, which is set in advance, and it is determined whether they match. The notification processing unit 13 issues a BEEP sound or the like when the recognition processing unit 12 determines a match.
[0006]
[Problems to be solved by the invention]
Since the conventional monitoring image processing apparatus is configured as described above, in general, the monitoring camera takes a form of looking down from a high position in order to widen the shooting range, so that an image with a sense of perspective is obtained. That is, an object that is close in distance is imaged at the lower part of the screen and a far object is imaged at the upper part of the screen. Therefore, the area integration unit 10 in the conventional monitoring image processing apparatus has the following problems.
Since the areas in the screen are integrated under the same conditions, over-integration is likely to occur for objects that are photographed large on the screen, and incomplete integration is likely to occur for objects that are photographed small.
In overintegration or incomplete integration, individual moving objects cannot be extracted accurately, and therefore an error occurs when compared with the previously-reported report target feature quantity 11. Thereby, even if it is a report object, the non-detection which does not determine with a report object, and the unnecessary detection which determines with the report object although it is not a report object generate | occur | produces.
[0007]
The present invention has been made to solve the above-described problems, and can provide a monitoring image processing apparatus that prevents undetection and unnecessary detection due to over-integration and incomplete integration and accurately determines individual objects. With the goal.
[0008]
[Means for Solving the Problems]
A monitoring image processing apparatus according to the present invention obtains a background image in accordance with an input video signal and extracts a change region of the current image by comparing the current image with the background image by the video signal. Extraction means, feature amount calculation means for binarizing the change area extracted by the change area extraction means, and calculating a feature quantity for each set of change areas after binarization, and feature quantity calculation means It is determined whether or not the circumscribed rectangles in the change area corresponding to the feature amount calculated by (1) can be integrated as one change target, and the integration condition that becomes difficult to integrate as the size of the circumscribed rectangle is larger than a predetermined value. while, in the integrated conditions the size of the circumscribed rectangle tends to integrate more smaller than a predetermined value, if possible integration, change area integrated hand for generating a bounding rectangle that integrates And re-calculating the feature quantity of the circumscribed rectangle integrated by the change area integrating means according to the feature quantity calculated by the feature quantity calculating means, and the feature quantity of the circumscribed rectangle and the preset target And a target area determination unit that determines whether or not they match , and the change area integration unit sets a circumscribed rectangle maximum value for each range in which the image area is divided in advance, A range determination means for determining whether or not it corresponds to a range, the size of one circumscribed square is calculated, and the magnification for determining the size of the virtual circumscribed square is larger when the size of the circumscribed square is larger than a predetermined value, While setting smaller than the standard multiple, the larger the size of the circumscribed rectangle is smaller than the predetermined value, the larger the magnification that determines the size of the virtual circumscribed rectangle is set relative to the standard multiple, It is determined whether the virtual circumscribed quadrangle whose one circumscribed square has been changed with the magnification set in (1) includes the other circumscribed quadrangle. If the size of the bounding rectangle is calculated and the size of the bounding rectangle does not exceed the maximum value of the bounding rectangle according to the range of one circumscribed rectangle determined by the range determination means, the virtual circumscribed rectangle determination to be integrated Means .
[ 0009 ]
DETAILED DESCRIPTION OF THE INVENTION
An embodiment of the present invention will be described below.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a monitoring image processing apparatus according to Embodiment 1 of the present invention. In the figure, 1 is an image input unit for inputting a video signal, and 2 is a current image storage for storing a current image by the video signal. Unit 3 is a background image storage unit that stores a background image by a video signal, 4 is a background update unit that sequentially processes the video signal to obtain a background image, and stores the updated background image in the background image storage unit 3; A background difference unit 5 extracts a change area of the current image from a difference between the current image and the background image. The change area extraction unit is configured as described above.
6 is a threshold value calculation unit for calculating a threshold value for binarizing the change region of the current image, 7 is a binarization unit for binarizing the change region of the current image with the threshold value, and 8 is a block for the binary data. A labeling unit that assigns a label to each label and 9 is a feature quantity computing unit that computes a feature quantity for each numbered label. The feature amount calculation means is configured as described above.
Reference numeral 10 denotes an area integration unit that integrates change areas that are considered to belong to the same change target for each numbered label and combines them into one change area. The change area integration unit is configured as described above.
11 is a preset feature value to be reported and 12 is stored, 12 is a recalculation of the feature value of the change area integrated by the area integration unit 10, and the recalculated feature value is calculated by the feature value calculation unit 9. A recognition processing unit that compares the feature quantity of the changed area that has not been integrated by the area integration unit 10 and the notification target feature quantity 11 to determine whether they match, and 13 indicates that the recognition processing unit 12 determines a match. This is a notification processing unit that issues a notification. The object determination unit is configured as described above.
FIG. 2 is a flowchart showing the processing of the region integration unit according to the first embodiment of the present invention, and FIG. 3 is an explanatory diagram showing the processing of the region integration unit according to the first embodiment of the present invention.
[ 0010 ]
Next, the operation will be described.
In FIG. 1, the video signal from the surveillance camera is input to the image input unit 1 and after A / D conversion, for example, the current image storage unit 2 and the background as 8-bit image data for each of R, G, and B This is supplied to the image storage unit 3. The current image storage unit 2 stores the image data. The background image storage unit 3 supplies the image data to the background update unit 4, and the background update unit 4 calculates, for example, a background image from the average of the past 30 image data and stores it in the background image storage unit 3. The background image to be updated is sequentially updated. Here, the reason why the background image is sequentially updated is to cope with the fact that the scenery of the area captured by the surveillance camera differs between day and night or due to movement of an object. The background difference unit 5 compares the current image of the current image storage unit 2 with the background image of the background image storage unit 3 and extracts a change area such as a moving object or a change object in the current image from the difference.
The threshold value calculation unit 6 calculates a threshold value for binarizing (1 bit) the background difference data (8 bits) from the background difference unit 5. In general, an optimum threshold value is obtained based on background difference data. The binarization unit 7 binarizes the background difference data with the threshold value obtained by the threshold value calculation unit 6. The labeling unit 8 numbers the binary data output from the binarizing unit 7 for each block. Here, the block of binary data means that adjacent data “1” pixels are regarded as one block. Before the area integration shown in FIG. 3, there are eight chunks of binary data, but here each is numbered. The feature amount calculation unit 9 calculates the feature amount for each label numbered by the labeling unit 8. Here, the feature amount refers to each set of binary data, that is, the area of each change region, vertical and horizontal dimensions, and the like.
[ 0011 ]
The area integration unit 10 integrates the change areas that are considered to belong to the same change target for each label numbered by the labeling unit 8 and combines them into one change area.
FIGS. 2 and 3 describe the processing of the area integration unit 10 in detail, and will be described below with reference to FIGS. 2 and 3.
First, the ferret diameter determination unit ST1 determines the ferret diameter obtained from the labeling unit 8. Here, as shown in FIG. 3, the ferret diameter is a circumscribed rectangle of the change region. Since the labeling unit 8 assigns labels to each change region, the total number of ferret diameters can be obtained from the labeling unit 8. Before the region integration shown in FIG. 3, the ferret diameter is 8, and in this case, the processing of steps ST2 to ST6 is performed 28 times. The processes of steps ST2 to ST6 are performed for the total number of ferret diameter combinations. When the ferret diameter number is n, the processes are performed _n C ₂ times.
Thereafter, the following integration determination is performed between the ferret diameters.
(Invention 1 Condition 1) The inclusion determining unit ST2 determines whether one ferret diameter is completely included in the other ferret diameter. If it is included, both ferret diameters are integrated.
(Invention 1 Condition 2) The contact determination unit ST3 determines whether one ferret diameter is in contact with the other ferret diameter. When they are in contact, both ferret diameters are integrated.
(Invention 1 Condition 3) The midpoint coordinate calculation unit ST4 calculates the midpoint coordinates of the ferret diameter, and the size determination unit ST11 determines the vertical size and horizontal size of each ferret diameter, and matches the image conditions to be determined in advance. Are classified into three in each of the vertical and horizontal directions (a total of nine combinations). The size-specific virtual ferret diameter calculator ST12 calculates a virtual ferret diameter based on the size information in step ST11.
The calculation method is for each size classified vertically and horizontally,
Large size: Standard multiple * m times (0 <m <1)
Medium size: standard multiple
Small size: standard multiple * n times (1 ≤ n)
The standard multiple is double vertical and double horizontal (multiplier can be changed by setting).
In the virtual ferret diameter determination unit ST6, when the midpoint of the other ferret diameter is included in the virtual ferret diameter, the two ferret diameters are integrated.
The midpoint coordinate calculation unit ST4, the size determination unit ST11, the size-specific virtual ferret diameter calculation unit ST12, and the virtual ferret diameter determination unit ST6 constitute a virtual circumscribed rectangle determination unit.
[ 0012 ]
The recognition processing unit 12 recalculates the feature amount of the change area of the ferret diameter integrated by the region integration unit 10, calculates the recalculated feature amount, and the feature amount calculation unit 9, and integrates it by the region integration unit 10. The feature amount of the change area of the ferret diameter that has not been performed is compared with the feature amount 11 to be notified of, for example, a person or a vehicle, which is set in advance, and it is determined whether they match. The notification processing unit 13 issues a BEEP sound or the like when the recognition processing unit 12 determines a match.
[ 0013 ]
In FIG. 3, the ferret diameter changes from before the region integration to after the region integration by the processing of the conventional region integration unit 10, and the ferret from before the region integration to after the region integration by the processing of the region integration unit 10 according to the first embodiment. It shows how the diameter changed.
In an image with perspective as shown in FIG. 3, sufficient integration is performed in (Conventional Condition 1) and (Conventional Condition 2) in the case of a large ferret diameter, and overconsolidation is performed in (Conventional Condition 3). It will cause.
On the other hand, in the case of a small ferret diameter, there are few cases where (Conventional condition 1) and (Conventional condition 2) can be integrated, and even (Conventional condition 3) cannot be integrated completely. This is because a small change in the distance often has a small image contrast, and a change region obtained by the difference becomes small.
In the first embodiment, it is difficult to integrate a large ferret diameter, that is, an object that appears in the vicinity, and it is easy to integrate with a small ferret diameter, that is, an object that appears in the distance, thereby accurately extracting the object. be able to.
[ 0014 ]
Embodiment 2. FIG.
The configuration diagram of the monitoring image processing apparatus according to the second embodiment is the same as FIG.
FIG. 4 is a flowchart showing the processing of the region integration unit according to the second embodiment of the present invention, and FIG. 5 is an explanatory diagram showing the processing of the region integration unit according to the second embodiment of the present invention.
[ 0015 ]
Next, the operation will be described.
4 and 5 describe the processing of the region integration unit 10 in detail, and will be described below with reference to FIGS. 4 and 5.
First, the ferret diameter determination unit ST1 determines the ferret diameter obtained from the labeling unit 8, and the midpoint coordinate calculation unit ST4 calculates the midpoint coordinates of each ferret diameter.
Next, in the range attribute adding unit ST21, as in the range setting example shown in FIG. 5, the ferret diameter maximum value is set for each range in which the image area is divided according to the image condition to be determined in advance. It is determined in which range the midpoint coordinates of the ferret diameter are included, and the attribute of the determined range is added to each ferret diameter. The range can be set in an arbitrary shape in an arbitrary area on the screen. Further, the range attribute adding unit ST21 constitutes a range determination unit.
Thereafter, the following integration determination is performed between the ferret diameters.
(Invention 2 Condition 1) The inclusion determination unit ST2 determines whether one ferret diameter is completely included in the other ferret diameter. If it is included, both ferret diameters are integrated.
(Invention 2 Condition 2) The contact determination unit ST3 determines whether one ferret diameter is in contact with the other ferret diameter. If they are in contact, the post-integration ferret diameter length determination unit ST22 calculates the ferret diameter length in the case of integration, and compares it with the maximum ferret diameter value in the range added by the range attribute addition unit ST21. . If the maximum ferret diameter is not exceeded, integration is performed. In addition, when the ferret diameter spans the range, it is sufficient to satisfy one of the conditions.
(Invention 2 Condition 3) In the size determination unit ST11, the vertical size and horizontal size of each ferret diameter are determined, and each of the vertical and horizontal sizes is set in accordance with image conditions to be determined in advance (total of 9 combinations). ). The size-specific virtual ferret diameter calculator ST12 calculates a virtual ferret diameter based on the size information in step ST11.
The calculation method is for each size classified vertically and horizontally,
Large size: Standard multiple * m times (0 <m <1)
Medium size: standard multiple
Small size: standard multiple * n times (1 ≤ n)
The standard multiple is double vertical and double horizontal (multiplier can be changed by setting).
In the virtual ferret diameter determination unit ST6, when the midpoint of the other ferret diameter is included in the virtual ferret diameter, the post-integration ferret diameter length determination unit ST23 calculates the ferret diameter length when integrated, Comparison with the maximum value of the ferret diameter in the range added by the attribute adding unit ST21 is performed. If the maximum ferret diameter is not exceeded, integration is performed. When the ferret diameter spans the range, it is sufficient to satisfy one of the conditions.
The virtual circumscribing is performed by the midpoint coordinate calculation unit ST4, the range attribute addition unit ST21, the size determination unit ST11, the size-specific virtual ferret diameter calculation unit ST12, the virtual ferret diameter determination unit ST6, and the post-integration ferret diameter length determination unit ST23. A quadrangle determination means is configured.
[ 0016 ]
In FIG. 5, the ferret diameter changes from before the region integration to after the region integration by the processing of the region integration unit 10 according to the first embodiment, and the region integration from before the region integration by the processing of the region integration unit 10 according to the second embodiment. It was shown that the ferret diameter changed later.
As shown after integration according to the first embodiment in FIG. 5, when there are a plurality of large ferret diameters in the vicinity, over-integration is caused even in the integration conditions according to the first embodiment.
On the other hand, as shown after integration according to Embodiment 2 in FIG. 5, by setting the maximum value of the ferret diameter when the ferret diameters belonging to the range 1 are integrated to the vertical y pixel and the horizontal x pixel, FIG. Even under such conditions, the occurrence of overintegration can be suppressed and the target can be extracted accurately.
[ 0017 ]
【The invention's effect】
As described above, according to the present invention, the integration condition between circumscribed rectangles in the change area integrating means is set to an integration condition that becomes difficult to integrate as the size of the circumscribed rectangle is larger than a predetermined value. Since the integration condition is such that the smaller the size of the smaller than the predetermined value, the easier it is to integrate, without causing over-integration in which different circumscribed rectangles are integrated when the circumscribed rectangle is large, In addition, when there is a small circumscribed rectangle, there is no incomplete integration that results in multiple circumscribed rectangles for a single change target, preventing undetected and unnecessary detection due to overintegration and incomplete integration, and There exists an effect which can be determined correctly.
Further, in the virtual bounding rectangle determination means, as when a large listening magnitude than a predetermined value at one of the circumscribed rectangle, a ratio which determines the size of the virtual circumscribing rectangle is set low relative standard multiples, the circumscribed As the size of the rectangle is smaller than the predetermined value, the virtual circumscribed rectangle in which one circumscribed rectangle is changed with a magnification that determines the size of the virtual circumscribed rectangle larger than the standard multiple is the other circumscribed rectangle. If a rectangle is included, it is determined whether the size of the circumscribed rectangle when the circumscribed rectangles are integrated does not exceed the maximum value of the circumscribed rectangle according to the range of one circumscribed rectangle determined by the range determining means. If it does not exceed, it is configured to integrate, so the setting of the circumscribed rectangle maximum value for each range in the image area by the range determination means, for example, the far image area As a result, it is easy to integrate, making it difficult to integrate near image areas, thereby further preventing non-detection and unnecessary detection due to over-integration and incomplete integration, and having the effect of accurately determining individual objects. .
[Brief description of the drawings]
FIG. 1 is a configuration diagram illustrating a monitoring image processing apparatus according to the related art and Embodiment 1. FIG.
FIG. 2 is a flowchart showing processing of a region integration unit according to Embodiment 1 of the present invention.
FIG. 3 is an explanatory diagram showing processing of a region integration unit according to Embodiment 1 of the present invention.
FIG. 4 is a flowchart showing processing of a region integration unit according to Embodiment 2 of the present invention.
FIG. 5 is an explanatory diagram showing processing of a region integration unit according to Embodiment 2 of the present invention.
FIG. 6 is a flowchart showing processing of a conventional region integration unit.
FIG. 7 is an explanatory diagram showing processing of a conventional region integration unit.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Image input part, 2 Current image storage part, 3 Background image storage part, 4 Background update part, 5 Background difference part, 6 Threshold calculation part, 7 Binary conversion part, 8 Labeling part, 9 Feature-value calculation part, 10 area | region Integration unit, 11 Notification target feature, 12 Recognition processing unit, 13 Notification processing unit, ST1 Ferre diameter number determination unit, ST2 Inclusion determination unit, ST3 Contact determination unit, ST4 Midpoint coordinate calculation unit, ST6 Virtual ferret diameter determination Part, ST11 size judgment part, ST12 virtual ferret diameter calculation part by size, ST21 range attribute addition part, ST22, ST23 after ferret diameter length judgment part.

Claims (1)

A change area extracting means for obtaining a background image according to an input video signal and extracting a change area of the current image by comparing the current image by the video signal and the background image;
A feature amount calculation means for binarizing the change region extracted by the change region extraction means and calculating a feature amount for each set of change regions after the binarization;
It is determined whether or not the circumscribed rectangles in the change area corresponding to the feature amount calculated by the feature amount calculating means can be integrated as one change target, and integration is performed when the size of the circumscribed rectangle is larger than a predetermined value. while the harder becomes integrated condition, the size of the circumscribed rectangle in the easily become integrated condition by integrating more smaller than a predetermined value, if possible integration, a change region integrating means for generating an enclosing rectangle that integrates ,
The feature amount of the circumscribed rectangle integrated by the change area integration unit is recalculated according to the feature amount calculated by the feature amount calculation unit, and the re-calculated feature amount of the circumscribed rectangle and the preset target feature An object determination means for comparing the amounts and determining whether they match ,
The change area integration means is:
Range determination means for setting a circumscribed rectangle maximum value for each range in which the image area is divided in advance and determining which range the circumscribed rectangle corresponds to;
When the size of one circumscribed rectangle is calculated and the size of the circumscribed rectangle is larger than a predetermined value, the magnification for determining the size of the virtual circumscribed rectangle is set smaller than the standard multiple and As the size of the rectangle is smaller than the predetermined value, the virtual circumscribed rectangle in which the magnification that determines the size of the virtual circumscribed rectangle is set larger than the standard multiple, and one of the circumscribed rectangles is changed by the set magnification. Is included, and if so, the size of the circumscribed rectangle when the one and the other circumscribed rectangles are integrated is calculated, and the size of the integrated circumscribed rectangle is A virtual circumscribed quadrangle determining unit that integrates when the maximum value of the circumscribed quadrangle corresponding to the range of one circumscribed quadrangle determined by the range determining unit is not exceeded. Monitoring the image processing apparatus according to symptoms.

Images