JP4828265B2 - Image sensor - Google Patents

Description

  The present invention relates to an image sensor that can detect a moving background object such as a plant by performing image processing on a plurality of monitoring images obtained by sequentially imaging a monitoring space at predetermined time intervals.

  In recent years, many image sensors that detect intruders or the like by processing a monitoring image captured by an imaging device such as a camera in a security system have been provided. In such an image sensor, a background image as a reference is stored, and a difference area is extracted between the monitoring image captured every moment and the background image, and an intruder is extracted based on the difference area. Is detected.

  In such an image sensor, a difference area is extracted in the same manner as when an intruder appears even when a background object such as a plant or a stream is shaken by the wind. Therefore, the image sensor is required to reliably detect only the intruder without erroneously detecting the movement of the background object.

Japanese Patent Application Laid-Open No. H10-228688 describes setting a mask region in which an intruding object is not detected based on the maximum luminance value and the minimum luminance value in each pixel constituting the monitoring image.
Patent Document 2 describes that the similarity between the edge of the background image and the edge of the monitoring image in the difference area is calculated, and an intruder is detected based on the calculated similarity. In the invention described in Patent Document 2, a high similarity is calculated in a difference area caused by a disturbance, and a low similarity is calculated in a difference area caused by an intruder.
JP2002-279429 JP2001-243475

  However, in the invention described in Patent Document 1, since the position of the pixel related to the moving background object is set in the mask area, the intruder can be detected when the moving object overlaps the intruder on the monitoring image. There is a risk of this. Therefore, it is necessary to select an installation position of the image sensor so that no detection leak occurs.

  In addition, since the edge position of a moving background object varies, the similarity of the invention described in Patent Document 2 is calculated to be low, and the moving background object is erroneously detected as an intruder.

  The present invention has been made in view of the above problems, and an image sensor capable of accurately detecting a moving background object existing in the monitoring space by performing image processing on the monitoring image obtained by imaging the monitoring space, and this detection The purpose is to realize an image sensor that can detect an intruder with high accuracy by eliminating the influence of moving background objects.

The image sensor according to claim 1 comprises:
Sequentially acquires the monitor image of the captured monitoring space, the pre-Symbol dynamic background object present in the monitoring space, an image sensor for detecting the phenomenon that the disappearance in a particular pixel on the monitoring image and the appearance of the edges is repeated ,
Edge detection means for detecting the presence or absence of an edge at each position of the monitoring image and storing the edge position;
Edge fluctuation position detection means for detecting an edge fluctuation position that has changed three or more times , the presence or absence of the edge detected by the edge detection means within the first predetermined time in the past being a first predetermined number of times ;
A difference area extracting means for extracting a difference area between the monitoring image and the background image stored in advance;
An edge matching degree calculating means for calculating a matching degree between the edge position and the edge fluctuation position in the difference region;
A moving background object detecting means for detecting the difference region having the matching degree equal to or greater than a predetermined value as a moving background object;
It is characterized by comprising.

The image sensor according to claim 2 is the image sensor according to claim 1,
Positions where the number of times the edge detection means did not detect an edge within the second predetermined time in the past having a length equal to or longer than the first predetermined time are excluded from the edge fluctuation position. Edge fluctuation position correcting means for performing

The image sensor according to claim 3 is the image sensor according to claim 1 or 2,
A position adjacent to the edge standing position by detecting an edge standing position where the number of times the edge detecting means detects an edge within a third predetermined time including the first predetermined time is equal to or more than a third predetermined number of times. Is provided with an edge fluctuation position correcting means for excluding from the edge fluctuation position.

The image sensor according to claim 4 is the image sensor according to any one of claims 1 to 3,
Intruder detection means for extracting an intruder feature amount from the monitoring image in the difference area and detecting the intruder when the feature amount exceeds a predetermined reference,
The intruder detection unit changes the predetermined criterion to be high when the moving background object detection unit detects the moving background object.

The image sensor according to claim 5 is the image sensor according to any one of claims 1 to 3,
Intruder detection means for extracting an intruder feature amount from the monitoring image in the difference area where the moving background object detection means did not detect a moving background object, and detecting the intruder when the feature amount exceeds a predetermined reference It has.

  According to the image sensor described in claim 1, when the degree of coincidence between the past edge fluctuation position and the current edge position is high, the difference area is defined as a banner or flag in which characters, symbols, and the like are described. Since it is determined that it is caused by a moving background such as a sign pole of a barber shop or a light-emitting bulletin board of an electric bulletin board, it is distinguished from a difference area caused by an intruder, so that the moving background can be accurately detected.

  According to the image sensor described in claim 2, in addition to the effect of the image sensor according to claim 1, it is possible to exclude a position that has been edgeless for a long time in the monitoring image from the edge variation position. Therefore, it is possible to accurately correct an error in which the edge fluctuation position is detected based on edge fluctuation that may be temporarily caused by a moving object such as an intruder.

  According to the image sensor described in claim 3, in addition to the effect of the image sensor according to claim 1 or 2, the position adjacent to the stationary edge position where the edge is constantly present is detected from the edge fluctuation position. Since it can be excluded, even if a part of the edge of the stationary object fluctuates due to fluctuations in environmental conditions such as sunlight and the edge fluctuation position is erroneously detected, this error can be corrected accurately.

  According to the image sensor described in claim 4, in addition to the effects of the image sensor according to claims 1 to 3, the predetermined value which is a comparison target of the feature amount of the intruder extracted from the monitoring image in the difference region is further provided. Since the reference is changed to be higher when a moving background object is detected, errors in detecting the moving background object as an intruder are reduced, and the intruder can be detected with high accuracy. That is, when there is a moving background object in the difference area, for example, by using a predetermined reference according to the degree of coincidence between the detected edge position and the edge fluctuation position (the higher the degree of coincidence, the higher the predetermined reference), etc. This effect can be achieved.

  According to the image sensor described in claim 5, in addition to the effect of the image sensor according to claims 1 to 3, in addition, when a moving background object is detected in the difference area, no intruder is detected, When a moving background object is not detected in the difference area, the feature amount of the intruder is extracted from the monitoring image in the difference area, and the intruder is detected when the feature amount exceeds a predetermined reference. An intruder is not detected for the detected difference area, and the inconvenience of erroneously detecting a moving background object as an intruder can be avoided by relatively simple data processing.

1. First, an overall configuration of an image sensor according to an embodiment of the present invention will be briefly described.
The image sensor of the present example sequentially acquires a plurality of monitoring images by imaging the monitoring space at an appropriate time interval by an imaging means such as a camera, and these monitoring images are obtained by a method described below. This is an image sensor having a function of detecting a moving background object existing in a monitoring space by distinguishing it from an intruder or the like by performing image processing.

  Here, the moving background object to be detected by the image sensor is the movement of the vegetation that appears in the surveillance image, the moving banner or flag on which characters or symbols are written, the rotating sign pole of the barber shop, the electric bulletin board This means an object that is present in the background image and has a motion, and a relatively large number of edges (contours, boundaries, etc.) are detected, such as a moving light-emitting display. An intruder means an object (including a person, an animal, an object, etc.) that does not exist in the background image but sometimes enters the monitoring image.

  The image sensor of this example sequentially detects and stores the presence / absence of an edge at each position (each pixel) of the monitoring image, and when the presence / absence of an edge detected within a certain past time has repeatedly fluctuated, That position is detected as an edge fluctuation position. Then, a difference area between the monitoring image and the background image stored in advance is extracted by the difference area extraction unit, and in this difference area, the degree of coincidence between the position of the newly detected edge and the edge fluctuation position is calculated, When this is equal to or greater than a predetermined value, that is, when it is determined from past data that many of the detected edges appear due to repeated edge fluctuations, a moving background object at the position in the difference area It is determined that there is an intruder and is distinguished

2. Configuration of Embodiment Hereinafter, a configuration of an embodiment (this example) of an image sensor of this example will be described in detail with reference to the drawings.
FIG. 1 is a functional block diagram showing the configuration of the image sensor of this example, FIG. 2 is a flowchart showing the entire basic operation of the image sensor of this example, and FIG. 3 is a flowchart showing edge variation position detection processing in the basic operation. FIG. 4 is a flowchart showing edge variation position correction processing in the basic operation, FIG. 5 is a flowchart showing edge matching degree calculation processing in the basic operation, and FIGS. 6 to 8 are explanatory diagrams of examples of the edge variation position detection processing. 8 to 11 are explanatory diagrams of an example of the edge variation position correction process, and FIGS. 12 and 13 are explanatory diagrams of another example of the edge variation position correction process.

(1) Configuration (Fig. 1)
The configuration and function of the image sensor of this example will be described with reference to FIG.
[Imaging unit 1]
The imaging unit 1 of the present example includes a camera device having an imaging element such as a CCD and a C-MOS and an illumination device such as an LED, and is connected to an image acquisition unit 2 to be described later and images at least a monitoring space. Image data that is a data series of pixel values including luminance values and holds the positions of the pixels on the image plane as the arrangement order of the data is generated, and the image data is output to the image acquisition unit 2 as an electrical signal.

  The imaging unit 1 needs to image the same monitoring space without moving at least during a series of operation periods in which an intruder is detected. However, the imaging unit 1 should not always be fixed in a fixed arrangement and orientation. You may comprise so that a different monitoring space can be imaged as needed.

  In the present specification, the image data output by the imaging unit 1 is referred to as a monitoring image. In this specification, the position is expressed as coordinates (x, y) as necessary, and the pixel value is expressed by a function of x and y, such as a luminance value I (x, y).

  Imaging and output of the monitoring image by the imaging unit 1 are performed at a preset constant time interval (indicating a unit of a fixed length of a predetermined time, for example, 0.2 seconds in this example). That is, the imaging unit 1 captures the monitoring area with a momentary operation at each time interval at a constant time interval of 0.2 seconds. In the present specification, each moment that delimits this time interval is represented by the notation of time t, and in order to indicate a specific time interval existing on the time axis, one time of the time that delimits the time interval is used. It may also indicate a time interval.

[Image acquisition unit 2]
The image acquisition unit 2 is an interface circuit that inputs and outputs a monitoring image to the imaging unit 1 and a control unit 3 described later. That is, the image acquisition unit 2 is connected to the imaging unit 1 and the control unit 3, and sequentially inputs the monitoring images output by the imaging unit 1 and outputs them to the control unit 3.

[Storage unit 4]
The storage unit 4 is connected to the control unit 3 and can be configured by a semiconductor memory such as ROM and RAM, a magnetic disk such as HDD, an optical disk drive such as CD-ROM and DVD-RAM, and any other storage medium. . The storage unit 4 stores various data such as a program for detecting an intruder by performing image processing on a monitoring image, a program for controlling each unit of the apparatus, preset data, a background image generated when the program is executed, and an edge image. These programs and data are stored and read or written by the control unit 3 as necessary.

[Control unit 3]
The control unit 3 can be constituted by a calculation device such as a CPU such as a DSP, a microcomputer, an LSI, or a numerical calculation processor, and is connected to the image acquisition unit 2, the output unit 5, and the storage unit 4. The control unit 3 receives monitoring image data output from the image acquisition unit 2 and reads out a program from the storage unit 4 and performs image processing on the monitoring image according to these programs to detect an intruder existing in the monitoring space. The detection result is output to the output unit 5. Further, when performing image processing, data is read from the storage unit 4 or written to the storage unit 4 as necessary.

[Output unit 5]
The output unit 5 is an interface circuit that outputs the detection result of the intruder to the outside. The input side is connected to the control unit 3 and an electric signal representing the detection result is input. The output side is a controller of a security system or the like. It is connected to an external device and outputs an electrical signal to the external device. For example, if the detection result indicates that an intruder has been detected, an electrical signal indicating that fact is transmitted to the external device. The output unit 5 may be provided with notification means such as a buzzer or LED. If the detection result indicates that an intruder has been detected, the notification means is energized to sound the buzzer or turn on the LED. To alert the people around you.

  Difference area extraction means 6, edge detection means 7, edge fluctuation position detection means 8, edge fluctuation position correction means 9, edge matching degree calculation means 10, moving background object detection means 11 and intruder detection means 12 described below are: As shown in FIG. 1, each block is represented as a functional block constituting the control unit 3 and can be constituted by hardware, but in this example, it is stored in a CPU such as an LSI constituting the control unit 3 and functions. It is configured as an image processing program.

[Difference area extraction means 6]
The difference area extraction unit 6 stores the monitoring image acquired by the image acquisition unit 2 in the storage unit 4 as a background image immediately after activation of the image sensor. Since this background image is a reference image for performing a difference process for extracting a difference area that changes from moment to moment from a monitoring image acquired thereafter, storage of the background image performed immediately after the activation or update described later Is called initialization of the background image in the storage unit 4. Note that this background image does not always use what was acquired immediately after startup, and in this example, the background image stored in the storage unit 4 is appropriately updated using the acquired monitoring image. And

  After the background image is initialized, the difference area extraction unit 6 performs a difference process between the monitoring image newly acquired by the image acquisition unit 2 and the background image stored in the storage unit 4 to extract the difference area. The extracted difference area includes a difference area due to a moving object (moving background object) existing in a background image such as a plant, in addition to a difference area due to a moving object that does not exist in the background image such as an intruder.

  Information regarding the extracted difference area is output from the difference area extraction means 6 to the edge coincidence calculation means 10 and the intruder detection means 12.

[Edge detection means 7]
The edge detection means 7 detects the presence or absence of an edge at each pixel position of the monitoring image. The detection result is held by the edge detection means 7 as an edge image in which the pixel value at the position where the edge is detected is 1 and the pixel value at the position where the edge is not detected is 0.

  An edge is an outline of an individual object in a monitoring image, a difference in material or color, a boundary that appears due to a shadow, or the like, and is detected in a substantially linear shape at a position where a luminance value greatly changes in the image. For example, edges are detected in the contours of leaves and branches in the image of a tree, the contours in the image of an intruder, the boundary of parts such as hair and face, the boundary of clothing such as clothing, and the pattern of the clothing. Is done.

  The edge image is output from the edge detection means 7 to the edge coincidence calculation means 10 and is written in the storage unit 4 as a past edge image.

[Edge fluctuation position detecting means 8]
The edge variation position detection means 8 reads out the edge images generated by the edge detection means 7 within the first predetermined time in the past from the storage unit 4 and compares these edge images in the order of the generation times for each pixel position. The number of times that the presence or absence of the edge fluctuates is counted, and a pixel position where the counted number is equal to or more than the first predetermined number is detected as an edge fluctuation position. The detection result is held by the edge fluctuation position detection unit 8 as a moving background object image in which the pixel value at the edge fluctuation position is 1 and the pixel value at other positions is 0. It is output to the position correction means 9.

  In a swaying tree, multiple leaves and branches repeat reciprocating motion, so if you observe a fixed point on the image (that is, if you detect an edge for a specific pixel on the image), the edges of the leaves and branches appear and disappear Will be repeated. The edge fluctuation position detection means 8 detects such a physical phenomenon and has a side surface for predicting the position where the edge of the moving background object appears at the current time. In this example, the total of the number of times an edge has appeared (no → present) and the number of disappearance (present → no) is counted. Since the appearance and disappearance of edges occur in pairs, only one of them may be counted.

  Since the number of edge fluctuations due to the movement of an intruder is typically two, the first predetermined number should be set to 3 or more so that this can be eliminated and repeated edge fluctuations due to moving objects can be detected. Is preferred. In this embodiment, the first predetermined number of times is set to three in order to detect the edge fluctuation due to the moving background the earliest.

  Further, in this embodiment, the first predetermined time is set to a time from 8 hours before to 1 hour, that is, from 1.8 seconds to 0.2 seconds before, and eight edge images generated within this time. When a change in presence or absence of an edge is recognized three times in a certain pixel, the position of the pixel is recognized as an edge change position and detected.

  In this example, the minimum unit of the time interval is set to 0.2 seconds. However, if the actual setting is different from this, the edge variation position required three times or more from the eight images described above is used. The number of images necessary for recognition and the number of times recognized as edge fluctuations may be different. That is, how many edge fluctuations are recognized from how many edge images acquired at what time interval, and whether or not the pixel should be recognized as an edge fluctuation position is, for example, the size of an intruder image that can be monitored in the monitoring image In accordance with the minimum value, the size of the pixels constituting the monitoring image, the expected movement speed of the intruder, and the like, it should be set so that optimum detection is performed based on empirical rules.

  Accordingly, the setting up to 0.2 seconds before in this example is an example, and there may be a delay of about several seconds, and the setting of 1.6 seconds is just an example. However, since it is necessary to be able to observe the situation where the edge of the moving background object repeatedly appears and disappears, the lower limit needs to be a time corresponding to three time intervals. In addition, when the first predetermined time is too long and the presence of edge fluctuation is counted from a larger number of edge images, the pixel is positioned at the edge fluctuation position as long as the minimum number of times that edge fluctuation should be recognized does not change. Therefore, since all adjacent pixels are recognized as edge fluctuation positions, the obtained moving background object image is in a state where the range in which the edge has moved is filled in a solid shape. The property that the edge positions are discretely distributed on the image plane is lost. For this reason, the upper limit is preferably about several seconds.

[Edge fluctuation position correcting means 9]
<Correction based on edgeless information>
The edge fluctuation position correction means 9 receives the moving background object image from the edge fluctuation position detection means 8 and reads out the edge image generated by the edge detection means 7 within the second predetermined time from the storage unit 4 and outputs these edges. Count the number of times that an edge was not detected at each pixel position of the image, and assume that there was no edge fluctuation due to the moving background object at the pixel position where this number is the second predetermined number or more. The pixel value at the pixel position is set to 0.

  This process corrects an error detected by the edge fluctuation position detecting means 8 for edge fluctuation temporarily caused by a moving object such as an intruder based on the fact that the edge has not been detected for a long time. is there.

  Therefore, it is preferable that the second predetermined time includes a past time than the first predetermined time, and is set as a time equal to or longer than the first predetermined time. More specifically, the start time of the first predetermined time is set as the end time, the second predetermined time having a length equal to or longer than the first predetermined time may be used, or the end time of the first predetermined time may be set as the end time. The end time may be a second predetermined time that is a past time zone including the first predetermined time and is longer than the first predetermined time.

  In this embodiment, the second predetermined time is a time including at least 16 hours before to 1 hour before, that is, a time including 3.2 seconds before and 0.2 seconds before, and at least 16 sheets generated within this time. An edge image is a processing target.

  The second predetermined number of times is determined based on the fact that it can be determined that the state without an edge is dominant, and that an edge may appear suddenly due to environmental changes such as lighting or the occurrence of noise. Set slightly less than the number of images. In the present embodiment, it is preferable that the second predetermined number of times is at least 14.

<Correction based on standing edge information>
Edge image generated and stored by the edge detection means 7 within the third predetermined time is read out from the storage unit 4, edge standing positions where edges are regularly present in these edge images are detected, and edge fluctuation position detection is performed. When the edge fluctuation position detected by the means 8 is adjacent to the edge stationary position, the edge fluctuation position is assumed to have no edge fluctuation due to the moving background object, and the pixel value of the pixel position in the moving background object image is set to 0. To do.

  Many of the edges of a stationary object continue to be detected at the same position, but some of them may or may not be detected due to changes in environmental conditions such as sunshine. Edge fluctuations are detected by the edge fluctuation position detection means 8. It may be done. Therefore, by the above processing, an edge fluctuation position where there is an edge that is stably detected in the vicinity thereof is corrected as being erroneously detected by the edge fluctuation position detection means 8.

  The detection of the edge standing position is preferably performed under the same environmental conditions as the detection of the edge fluctuation position, and it is preferable that the third predetermined time includes the first predetermined time. In this embodiment, the third predetermined time is set to be the same as the first predetermined time.

  The moving background object image corrected by the edge variation position correcting unit 9 by the above two methods is output to the edge coincidence degree calculating unit 10.

[Edge coincidence calculation means 10]
The edge coincidence degree calculation means 10 receives the difference area information from the difference area extraction means 6, receives the edge image at the current time from the edge detection means 7, and is corrected by the edge variation position correction means 9. Is entered.

  The degree of coincidence between the edge position at the current time held in the edge image and the edge variation position held in the moving background image is calculated for each pixel in the difference area, and the degree of coincidence for each pixel in each difference area Is output to the moving background object detection means 11. The degree of coincidence indicates a ratio at which the position of the edge at the current time coincides with the edge fluctuation position, and is an index for determining whether or not the difference area is due to a moving background object.

  The edge of a moving object such as an intruder may appear at the edge fluctuation position extracted for the leaves and branches of a swaying tree, but the intruder's outline etc. is different from the outline of the leaves and branches. The possibility is low compared to the edges of moving objects. When the tree is swaying, the leaves and branches repeat a reciprocating motion substantially, so the edge at the current time is likely to appear at the edge fluctuation position. In the degree of coincidence calculated for a plurality of edge fluctuation positions in the difference area, the difference in possibility is a significant difference.

[Moving Background Object Detection Unit 11]
The moving background object detection means 11 inputs the edge matching degree in the difference area from the edge matching degree calculation means 10, detects the presence or absence of a moving background object in each difference area, and detects the detection result and the matching degree as an intruder. Output to means 12.

  Compare the degree of coincidence of the edges of the difference area with a predetermined value, and if the degree of coincidence is greater than or equal to a predetermined value, there is a moving background in the difference area, otherwise there is no moving background in the difference area Is determined.

[Intruder detection means 12]
The intruder detection means 12 inputs the acquired monitoring image, inputs information on the difference area from the difference area extraction means 6, and matches the presence or absence of a moving background object in each difference area from the moving background object detection means 11. Enter the degree information.

  The feature image of the intruder described later is extracted by performing image processing on the monitoring image in the difference region, and the feature amount of the intruder is compared with a predetermined reference. When the feature amount exceeds the predetermined reference, the intruder enters the difference region. Yes, if not, it is determined that there is no intruder in the difference area.

  At this time, with reference to the information from the moving background object detection means 11, when there is a moving background object in the difference area, a predetermined criterion according to the degree of coincidence is used. More specifically, the higher the matching degree, the higher the predetermined reference. On the other hand, when there is no moving background object in the difference area, a predetermined reference fixed lower than when there is a moving background object is used. By doing in this way, it becomes difficult to detect an intruder in the difference area where the moving background object is detected, and it becomes difficult to make an error in detecting the moving background object as an intruder by mistake.

  The intruder may be detected by extracting the feature amount of the intruder and comparing it with a predetermined standard only for the difference area where the moving background object is not detected.

  The intruder detection means 12 outputs a signal indicating that an intruder has been detected to the output unit 5 when there is even one difference area where the intruder has been detected.

(2) Image sensor operation (Figures 2 to 5)
When the image sensor is turned on, the imaging unit 1 outputs a monitoring image at a predetermined time interval, and the control unit 3 reads a program from the storage unit 4 to acquire a new monitoring image. Each time the processes shown in the flowcharts of FIGS. 2 to 5 are repeatedly executed according to the program, an intruder is detected. Each processing step in the flowcharts of FIGS. 2 to 5 is given a processing step number represented by S + number for convenience in the following description.
Hereinafter, each processing step will be described.

[S21: Monitor image acquisition]
The control unit 3 acquires a monitoring image output from the imaging unit 1 via the image acquisition unit 2. The acquired monitoring image may be temporarily stored in the storage unit 4.

[S22 and S23; background image initialization]
The control unit 3 executes the following processing using the difference area extraction unit 6.
Since the background image is not initialized immediately after the image sensor is activated (NO in S22), the following initialization processing is performed in step S23.

  At the first time immediately after activation, the acquired monitoring image is newly stored in the storage unit 4 as a background image. Thereafter, for several hours (for example, 5 hours), the newly acquired monitoring image and the background image stored in the storage unit 4 are sequentially averaged, and the averaged background image is overwritten in the storage unit 4 and initialized. Become. When initialization has been completed (YES in S22), the process proceeds to step S24.

[S24: differential area extraction]
The control unit 3 executes the following processing using the difference area extraction unit 6.
A background image is read from the storage unit 4, and an absolute value of a difference in luminance value is calculated for pixels at the same coordinates of the monitoring image and the background image. It is determined that pixels whose absolute value of the calculated difference is greater than a predetermined threshold value have changed, and other pixels have no change. A difference image is generated by setting the pixel value of a pixel having changed to 1 and setting the pixel value of a pixel having no change to 0.

  If the pixel values of neighboring pixels are 1 in the difference image, these neighboring pixels are collected as one difference area. The area (number of pixels) of each difference area is calculated, and a difference area whose calculated area is equal to or larger than a preset area threshold is assigned as a target for subsequent processing and managed. In this example, the management number is represented by a variable m. A difference area whose area is less than a preset area threshold is determined as a small difference area caused by noise and is not subject to subsequent processing.

[S25: Edge detection]
The control unit 3 executes the following processing by the edge detection means 7.
The edge intensity value E (x, y) of each pixel of the monitoring image is calculated by the following equation (1).

E (x, y) = (1/2) | I (x + 1, y) -I (x-1, y) | +
(1/2) | I (x, y + 1) -I (x, y-1) | (1)

  The edge intensity value is compared with a preset threshold value, and a pixel for which the edge intensity value is determined to be greater than the threshold value is determined to have an edge, and other pixels are determined to have no edge. An edge image in which the pixel value of a pixel with an edge is 1 and the pixel value of a pixel without an edge is 0 is generated.

  The edge strength value can be calculated using a known operator such as the Sobel operator, the Prewitt operator, and the Roberts operator in addition to the above formula.

  The generated edge image is stored in the storage unit 4. The edge image of 9 hours before is deleted and the edge image of the current time is added so that the edge images from the current time to 8 hours before are always stored in the storage unit 4.

[S26 and FIG. 3; edge fluctuation position detection]
The control unit 3 executes the “edge variation position detection” subroutine process, which is shown in detail in the flowchart of FIG. 3, by the edge variation position detection means 8 in S26 of FIG. 2 as described below.

  In step S <b> 301, an edge image generated from 8 hours before 1 hour, which is the first predetermined time, is read from the storage unit 4.

  In step S302, the coordinates (x, y) of the edge image and moving background object image to be processed in the following steps S303 to S306 are sequentially set.

  In step S303, in the eight edge images read out in step S301, pixel values of the same coordinates (x, y) are compared in the order of generation, and temporally from 0 to 1 and from 1 to 0. The number of times of change is counted as the number of edge fluctuations Nf (x, y) at the coordinates.

  The variation number Nf (x, y) is compared with a preset threshold value Tf1 (first predetermined number; three times), and if the variation number Nf (x, y) is equal to or greater than the threshold value Tf1 ( In S304, the coordinate (x, y) is detected as the edge fluctuation position, and 1 is set to the pixel value L (x, y) of the moving background object image (S305).

  If variation count Nf (x, y) is less than threshold value Tf1 (NO in S304), pixel value L (x, y) is set to 0 (S306).

  When processing has been performed for all pixels (YES in S307), the edge fluctuation position detection process is completed, and the moving background object image that holds the edge fluctuation position generated in this process is used as the edge fluctuation position correction means 9. The process proceeds to step S27 in FIG. If not (NO in S307), the process returns to step S302 to set the coordinates of the pixel to be processed next and repeat the same process.

[S27 and FIG. 4; edge fluctuation position correction]
The control unit 3 executes the “edge variation position correction” subroutine process, which is shown in detail in the flowchart of FIG. 4, in S27 of FIG. 2 by the edge variation position correction means 9 as described below.

  In step S <b> 401, eight edge images generated from 8 hours before 1 hour, which is the third predetermined time, are read from the storage unit 4.

  First, correction based on the standing edge information is performed. In step S402, the coordinates (x, y) of the edge image and the standing edge image to be processed in the following steps S403 to S406 are sequentially set.

  In step S403, the number of times the pixel value of each pixel at the same coordinate (x, y) is 1 in the eight edge images read out in step S401 is counted as the number of edged Na (x, y) in the coordinate. To do.

  By comparing the number of edged Na (x, y) with a preset threshold Ta (third predetermined number), the coordinates (x, y) is detected as a standing edge position (YES in S404), and 1 is set to the pixel value A (x, y) of the standing edge image (S405).

  If edge count Nf (x, y) is less than threshold value Ta (NO in S404), pixel value A (x, y) is set to 0 (S406).

  In this embodiment, the threshold value Ta is 8.

  If all pixels have been processed (YES in S407), the process proceeds to the next step S408. If all pixels have not been processed (NO in S407), the process returns to step S402, and the next processing target The same process is repeated by setting the coordinates of the pixel.

  In step S408, a coordinate (edge variation position) having a pixel value of 1 in the moving background object image is sequentially searched, and the coordinate (x, y) is set as a processing target in the following steps S409 to S411.

  In step S409, pixels in the vicinity of the coordinate (x, y) in the standing edge image in eight directions (upper, lower, left, right, upper right diagonal, upper left diagonal, lower right diagonal, and lower left diagonal adjacent to the pixel) The number of pixels having a pixel value of 1 is calculated as the number of neighboring standing edges Ns (x, y).

The number of neighboring standing edges Ns (x, y) is compared with a preset threshold value Ts, and if the number of neighboring standing edges Ns (x, y) is equal to or greater than the threshold value Ts (YES in S410). The pixel value L (x, y) of the moving background object image at the coordinate (x, y) is corrected to 0, assuming that the standing edge has changed due to illumination fluctuation or the like (S411). That is, it is possible to correct the determination that the edge detected as originally depending on the moving background object is the standing edge.
In this embodiment, the threshold value Ts, which is a criterion for determination, is set to 3.

  When processing has been performed for all edge fluctuation positions (YES in S412), the process proceeds to step S413 to start the next correction processing. If not (NO in S412), the next edge change position to be processed is set and the same processing is repeated.

  Next, correction based on edgeless information is performed. In step S413, the edge image, moving background object image, and counter coordinates (x, y) to be processed in the following steps S414 to S423 are sequentially set.

  In step S414, the number of times that the pixel value of each pixel at the same coordinate (x, y) in the eight edge images read out in step S401 is 0 is counted as the number of edgeless Nn (x, y) in the coordinate. To do.

  If the number of edgeless Nn (x, y) is compared with a preset threshold value Tn and the number of edgeless Nn (x, y) is equal to or greater than threshold value Tn (YES in S415), counter Cn ( The value of x, y) is increased by 1 (S416). In this example, the threshold value Tn is set to 7. This counter is for accumulating edgeless information. The larger the counter value, the higher the need for correction based on edgeless information. Note that the counter Cn (x, y) is initialized to 0 immediately after the image sensor is activated.

  In step S417, in the eight edge images read out in step S401, the pixel values of the same coordinates (x, y) are compared in the order of generation, and temporally from 0 to 1 and from 1 to 0. The number of times of change is counted as the number of edge fluctuations Nf (x, y) at the coordinates.

Comparing the number of fluctuations Nf (x, y) with a preset threshold value Tf2, if the number of fluctuations Nf (x, y) is equal to or greater than threshold value Tf2 (YES in S418), counter Cn (x, y The value of y) is corrected to one half (S419), and the process proceeds to step S422. Note that the counter value is an integer with the fractional part truncated.
This prevents the edge fluctuation position from being erroneously corrected when, for example, the vegetation that has not been shaken starts to shake violently.

  On the other hand, if the number of fluctuations Nf (x, y) is smaller than threshold value Tf2 (NO in S418), the number of fluctuations Nf (x, y) is then compared with a preset threshold value Tf3. If Nf (x, y) is equal to or greater than threshold value Tf3 (YES in S420), the value of counter Cn (x, y) is decreased by 2 (S421). When the value of the counter Cn (x, y) becomes negative, the value is corrected to 0. As a result, the edge fluctuation position is prevented from being erroneously corrected when, for example, a plant that has not been shaken starts to shake.

  Since the threshold value Tf2 is a value for detecting more severe shaking, it is preferable to set Tf1 ≦ Tf3 <Tf2. In this example, the threshold value Tf3 is set to 4 and the threshold value Tf2 is set to 5. . Further, the division and subtraction of the counter value in steps S419 and S421 is an example, and each calculation can be appropriately set based on the idea that the calculation in step S419 greatly reduces the counter value than the calculation in step S421.

The counter Cn (x, y) is compared with a preset threshold value Tc, and if the counter Cn (x, y) is greater than or equal to the predetermined threshold value Tc (YES in S422), the coordinates (x , Y), assuming that the edge of the moving background object does not appear, 0 is written to the pixel value L (x, y) of the moving background object image (S423). In this example, the threshold value Tc is set to 9.
As described above, at each pixel position, the counter is incremented by 1 if the number of no-edges in the past eight times is 7 or more, and the pixel position where the counter value is 9 or more is a position where no edge appears regularly. And determine the subject of amendment. When the correction target is determined in the shortest time, the second predetermined time is 16 hours and the second predetermined number of times is 14. In this way, the correction target is determined sequentially, and the second predetermined time is dynamically set to a time of 16 hours or more in the process, and the second predetermined number of times is set to 14 or more times. Set dynamically.

  If all pixels have been processed (YES in S424), the edge fluctuation position correction process is terminated, and the moving background object image corrected in this process and holding the edge fluctuation position is sent to the edge coincidence calculation means 10. The process proceeds to step S28 in FIG. Otherwise (NO in S424), the coordinates (x, y) of the pixel to be processed next are set and the same processing is repeated.

  With the above processing, a temporary edge appears and disappears due to a moving object such as an intruder, and this can be corrected when the edge fluctuation position detecting means 8 erroneously detects the edge fluctuation position. .

[S28 and FIG. 5; edge coincidence calculation]
The control unit 3 executes the “edge variation position correction” subroutine process, which is described in detail in the flowchart of FIG. 5, in S28 of FIG.

  In step S501, difference areas m extracted in step S24 and subjected to processing in steps S502 to S507 below are sequentially set.

  In step S502, the number of edges e (m) at the current time existing inside the difference area m is counted using the edge image and the difference image. For this purpose, the sum of logical products of the edge image and the difference image may be calculated.

  If the number of edges e (m) is smaller than a preset threshold value Te (NO in S503), it is almost certainly not a moving background object, so 0 is set to a matching degree r (m) described later (S504). .

  If the number of edges e (m) is equal to or greater than a preset threshold value Te (YES in S503), there is a possibility of a moving background object, so the edge area, moving background object image, and difference image are used, and the difference area m The number d (m) of coincidence between the edge position at the current time and the edge fluctuation position existing inside is counted (S505). For this purpose, the sum of logical products of the edge image, the moving background object image, and the difference image may be calculated.

In step S506, the degree of coincidence r (m) is calculated by normalizing the coincidence number d (m) with the number of edges e (m) at the current time when the coincidence number d (m) has appeared.
r (m) = d (m) / e (m) (2)

  Instead of e (m), the sum of the edge position at the current time existing inside the difference area m and the edge fluctuation position existing inside the difference area m is used to normalize the number of matches d (m). Also good. Alternatively, the matching number d (m) may be normalized using the area (number of pixels) of the difference region m instead of e (m).

  Furthermore, the matching degree r (m) may be converted using a predetermined function. The predetermined function experimentally obtains the probability that the difference area m is due to the moving background object when r (m) is actually obtained, and converts r (m) into the probability (range 0 to 1). It is preferable to set the function to be

  If processing has been performed for all the difference areas (YES in S508), the edge coincidence degree calculation process is terminated, and the coincidence degree in each difference area is output to the moving background object detection means 11, and step S29 in FIG. Migrate to Otherwise (NO in S508), the process returns to step S501, and the same processing is repeated after setting the difference area m to be processed next.

[S29: Moving object detection]
The control unit 3 executes the following processing by the moving background object detection unit 11.
For each difference area m, the edge matching degree r (m) is compared with a preset threshold value Tr, and if the edge matching degree r (m) is equal to or greater than the threshold value Tr, the difference area m is a moving background object. If the edge matching degree r (m) is less than the threshold value Tr, it is determined that the difference area m is not due to a moving background object. The threshold value Tr for detecting a moving background object is preferably set to 0.5 or more, and is set to 0.5 in this example.

[S30: Intruder feature value calculation]
The control unit 3 executes the following processing using the intruder detection means 12.
For each difference region, various image feature amounts are extracted, each image feature amount is input to a predetermined function set in advance, converted into a value, and the converted values are weighted and added, and the humanity of the difference region is calculated. Is calculated (value range 0 to 1).

The image feature quantity to be extracted and the characteristics of the predetermined function are exemplified below.
1) Area of difference area (number of pixels)
The predetermined function outputs a high value if the area is within a range that can be regarded as a person, and outputs a low value otherwise.
2) Shape of the difference area The predetermined function outputs a high value if it is long in the vertical direction or long in the horizontal direction, and outputs a low value otherwise.
3) Similarity between the texture of the background image and the texture of the monitoring image The texture of the image means a luminance pattern or pattern. When the texture similarity is high, there is a high possibility of transmissive disturbance (light or shadow). Therefore, the predetermined function outputs a low value when the similarity is high, and outputs a high value when the similarity is low.
4) Edge image increase rate relative to background image The predetermined function outputs a high value when the increase rate is high, and outputs a low value when the increase rate is low.
5) Reduction rate of edge of monitoring image with respect to background image The predetermined function outputs a high value when the reduction rate is high, and outputs a low value when the reduction rate is low.

  Extraction of the image feature amount may be omitted for the difference area where the moving background object is not detected by the moving background object detection unit 11. In this case, the index related to the difference area is 0.

[S31; Intruder detection]
The control unit 3 performs the following processing for each difference area m by the intruder detection means 12.
1) A threshold value Th (m) for determining whether or not the difference area is due to an intruder is set.
In the case of the difference area m in which the moving background object is detected by the moving background object detection unit 11, it is considered that the higher the edge matching degree r (m), the higher the probability of being a moving background object. In order to distinguish intruders, the threshold value Th (m) is set high.
Th (m) = (α · r (m) +1) · Th0 (α> 0) (3)
Even in the case of the difference area m in which no moving background object is detected by the moving background object detection means 11, it cannot be said that it is immediately caused by an intruder, but may be detected due to other causes such as light. , It is set lower than Th (m) when a moving background object is detected.
Th (m) = Th0 (4)

  2) An index representing humanity is compared with a threshold value Th (m), and if the index is equal to or greater than the threshold value Th (m), it is determined that the difference area m is due to an intruder, and the index is a threshold value. If it is less than the value Th (m), it is determined that the difference area m is not due to an intruder.

  It is assumed that an intruder is detected when there is at least one difference area by the intruder.

  Although an example in which the threshold value Th (m) is changed according to the edge matching degree r (m) has been shown, an index representing humanity may be corrected according to the edge matching degree r (m). In this case, it is considered that the higher the edge matching degree r (m), the higher the probability that the object is a moving background object. Therefore, the threshold Th (m) is set to a constant value as a correction for reducing an index representing humanity. To do.

[S32 to S34; Processing According to Intruder Detection Result]
When an intruder is detected (YES in S32), control unit 3 outputs a detection signal to an external device via output unit 5 (S33). When no intruder has been detected (NO in S32), control unit 3 updates the background image as follows by difference area extraction means 6 in order to prevent a decrease in intruder detection accuracy due to illumination fluctuations (S34). .

1) A background image obtained by averaging the monitoring image at the current time and the background image stored in the storage unit 4 may be overwritten in the storage unit 4.
2) The stored background image may be cleared when an abrupt illumination change is detected, for example, by extracting an extremely large difference area. In this case, the background image initialization step is executed at the next time.
In place of the process in step S34, the background image may be replaced with the monitoring image at the current time at regular intervals regardless of whether or not an intruder has been detected.

(3) Examples of image data processing (FIGS. 6 to 13)
An example of image processing in the image sensor of this example described above will be specifically described.
In each figure to be referred to, when there are two sets of horizontal arrangements of images, enlarged images, etc., these are referred to as stages, called from the top to the top and bottom, and each figure arranged horizontally in each stage , Left figure, center figure, right figure.

[Example of Edge Fluctuation Position Detection Processing] (FIGS. 6 to 8)
Let t be the current time, (t-1) one hour before, (t-2) ... two hours before.
The upper part of FIG. 6 is an edge image generated by the edge detection means 7. Black with an edge and white with no edge.

The lower part of FIG. 6 is an enlarged view of a part of each edge image in the upper part.
Each block represents a pixel, a black block has an edge (pixel value 1), and a white block has no edge (pixel value 0).

  The upper left figure of FIG. 7 shows the difference image at time t created by the difference area extraction means 6. As shown in the figure, the difference area includes the difference area of the moving object (intruder) and the difference area of the moving background object.

  The middle diagram in the upper part of FIG. 7 is a moving background object image at time t generated by the edge fluctuation position detection means 8. Edge variation positions are shown in black and none are shown in white.

  The upper right diagram in FIG. 7 shows an edge image at time t. Black with edges and white with no edges.

  The lower part of FIG. 7 is an enlarged view of a part of the moving background object image at the time t (the upper middle part of FIG. 7). Each block represents a pixel, a black block indicates edge variation (pixel value 1), and a white block indicates no edge variation (pixel value 0).

Here, specific consideration will be given to several pixels constituting each image of FIGS.
At coordinates (x1, y1), the appearance and disappearance of the edge were repeated due to the shaking of the leaf, and the pixel value of the edge image was 10100101. The number of fluctuations Nf (x1, y1) is counted as 6, and since it is equal to or greater than the threshold value Tf1, the coordinates (x1, y1) are detected as the edge fluctuation position, and the pixel value L (x1, y1) of the moving background object image Became 1.
In general, the edge variation includes edge variation due to a plurality of leaves.

  At coordinates (x2, y2), an edge appeared at time (t-8) due to the passage of the intruder, and the pixel value of the edge image was 01000000. The number of fluctuations Nf (x2, y2) is counted twice and is less than the threshold value Tf1, so the coordinates (x2, y2) are not detected as edge fluctuation positions, and the pixel value L (x2, y2) of the moving background object image is detected. y2) was zero.

  At the coordinates (x3, y3), there is always an edge of the stationary trunk portion, and the pixel value of the edge image is 11111111. The number of fluctuations Nf (x3, y3) is counted as 0, and is less than the threshold value Tf1, so the coordinates (x3, y3) are not detected as the edge fluctuation position, and the pixel value L (x3, x3) of the moving background object image is detected. y3) was zero.

  The left figure of FIG. 8 shows the edge fluctuation position in the difference area. That is, the outer edge of the difference area is indicated by a dotted line, the edge fluctuation position is indicated by black, and the other edge is indicated by white.

  The center diagram of FIG. 8 shows an edge at time t in the difference area. That is, the outer edge of the difference area is indicated by a dotted line, the edge position is indicated by black, and the other edge is indicated by white.

  The right diagram in FIG. 8 shows pixels that are edge fluctuation positions and edges in the difference area at time t. The outer edge of the difference area is indicated by a dotted line, and the pixels that are edge fluctuation positions and edges are black. Those that are not are shown in white. This figure is an AND of the two images of the left figure and the central figure of FIG. 8, and the number of coincident parts of both is the coincidence number d.

  In the edge coincidence calculation means 10, first, the number of edges e (m) in the extraction region is checked from the edge image. Similarly, the number d (m) of pixels that are the edge fluctuation position and the edge position at the current time in the extraction region is examined from the moving background object image and the edge image. Then, r (m) = d (m) / e (m) is calculated. The moving background object detection means 11 detects a moving background object if r (m)> 0.5. In the case of a moving background object, since there are many pixels that are edges and are edge fluctuation positions, d (m) is large and r (m) is also large. On the other hand, since an intruder has few pixels that are edges and edge fluctuation positions, d (m) is small and r (m) is also small.

  Since the moving background object image is generated based on the information of the edge detected in a substantially linear shape, the edge fluctuation positions are in a state of being distributed discretely on the image plane. With such a distribution, moving background objects can be distinguished from intruders and detected by confirming the coincidence between the edge fluctuation position and the edge position. If the edge variation position is detected (continuously) in the entire area where the leaf has moved (when the area is filled with black), all the positions where the edge appears at the current time in this area. Coincides with the edge fluctuation position and the degree of coincidence is always 1, so that the moving background object and the intruder cannot be distinguished.

[Example of Edge Fluctuation Position Correction Processing Based on No-Edge Information] (FIGS. 9 to 11)
Let t be the current time, (t-1) one hour before, (t-2) ... two hours ago.
The upper part of FIG. 9 and the upper part of FIG. 10 are edge images generated by the edge detection means 7, and the presence of an edge is shown in black and the absence of an edge is shown in white. The lower part of FIG. 9 and the lower part of FIG. 10 are obtained by enlarging a part of each edge image shown in the upper part of FIG. 9 and the upper part of FIG. 10, black blocks are pixels with edges (pixel value 1), and white blocks are no edges. A pixel of (pixel value 0) is shown. The upper left diagram in FIG. 11 is a part of the moving background object image generated by the edge fluctuation position detecting means 8. Edge variation positions are shown in black, and other positions are shown in white. The upper right diagram in FIG. 11 is an image of the counter Cn generated at the time t by the edge fluctuation position correcting means 9. For easy understanding, coordinates whose value is greater than or equal to the threshold value Tc are shown in black, and coordinates whose value is less than the threshold value Tc are shown in white. The lower part of FIG. 11 shows the moving background object image corrected by the edge fluctuation position correcting means 9. The edge fluctuation position is shown in black, and the other position is shown in white.

Here, a number of pixels constituting each of the images in FIGS. 9 to 11 will be specifically examined.
First, the processing by the edge fluctuation position detection means 8 will be examined.
At coordinates (x1, y1), the appearance and disappearance of an edge were temporarily repeated by an intruder, and the pixel value of the edge image from time (t-9) to (t-1) was 10000010. Therefore, the number of fluctuations Nf (x1, y1) is counted as 3 at time t and becomes equal to or greater than the threshold value Tf1, and the coordinate (x1, y1) is erroneously detected as an edge fluctuation position despite the fluctuation caused by the intruder. The pixel value L (x1, y1) of the moving background object image is 1. Two other false detections occurred.

Next, processing by the edge fluctuation position correcting means 9 will be considered.
At coordinates (x1, y1), no edge is detected from time (t-25) to (t-18), and the number of edgeless times Nn (x1, y1) is counted as 8 at time (t-17). Thus, the threshold value Tn is exceeded and the counter Cn (x1, y1) is 1. Thereafter, even from time (t-16) to (t-9), the number of edgeless times is counted as 8, and the counter is increased. At time (t-9), the counter Cn (x1, y1) is equal to or greater than the threshold value Tc. 9 and the coordinates (x1, y1) were to be corrected. Further, even when the time (t-9) reaches t, the counter does not decrease and the coordinates (x1, y1) continue to be corrected.

  Since the coordinates (x1, y1) are correction targets at time t, the pixel value L (x1, y1) of the moving background object image is corrected from 1 to 0. The pixel values of the moving background object image were corrected to 0 in the same manner for the other two erroneously detected locations.

[Example of edge variation position correction processing based on standing edge information] (FIGS. 12 and 13)
Let t be the current time, (t-1) one hour before, (t-2) ... two hours before.
The upper part of FIG. 12 represents the edge image generated by the edge detection means 7. Black with an edge and white with no edge. The lower part of FIG. 12 is an enlarged view of a part of each edge image shown in the upper part of FIG. A black block indicates a pixel with an edge (pixel value 1), and a white block indicates a pixel without an edge (pixel value 0). The upper left figure in FIG. 13 shows a moving background object image generated by the edge fluctuation position detecting means 8. Black blocks indicate edge fluctuations, and white blocks indicate no edge fluctuations. The upper right diagram in FIG. 13 is a part of the standing edge image generated by the edge fluctuation position correcting means 9. The black block indicates the edge standing position, and the white block indicates the position where it is not. The lower part of FIG. 13 is a moving background object image corrected by the edge fluctuation position correcting means 9. Black blocks indicate edge fluctuations, and white blocks indicate no edge fluctuations.

  At coordinates (x1, y1), (x1, y2), and (x1, y3), the edge of the house, which is a stationary object, always exists, and the pixel value in the edge image is 11111111. In the edge fluctuation position detecting means 8, the number of fluctuations in these coordinates is counted as 0, and the pixel value of the moving background object image becomes 0. On the other hand, in the edge fluctuation position correcting means 9, the number of edges with these coordinates is counted as 8, and the pixel value of the standing edge image is 1.

  The edge that appears at the coordinates (x2, y2) is also the edge of the house, but the appearance and disappearance were repeated due to illumination fluctuations, and the pixel value in the edge image was 10111110. Therefore, the edge fluctuation position detection means 8 counts the number of fluctuations Nf (x2, y2) to be equal to or greater than the threshold value Tf1, and the coordinate (x2, y2) is erroneously detected as the edge fluctuation position, and the moving background object image The pixel value L (x2, y2) of was 1. Three other misdetections occurred.

  However, since the pixel values of (x1, y1), (x1, y2), (x1, y3) near the coordinates (x2, y2) in the standing edge image are 1, the edge fluctuation position correction means 9 Thus, the number of neighboring standing edges Ns (x2, y2) is counted as 3 and becomes equal to or greater than the threshold value Ts, so that the pixel value L (x2, y2) of the moving background object image is corrected from 1 to 0. . The pixel values of the moving background object image were corrected to 0 in the same manner at the other three locations that were erroneously detected.

FIG. 1 is a functional block diagram showing the configuration of the image sensor of this example. FIG. 2 is a flowchart showing the entire basic operation of the image sensor of this example. FIG. 3 is a flowchart showing the edge variation position detection process in the basic operation. FIG. 4 is a flowchart showing the edge variation position correction process in the basic operation. FIG. 5 is a flowchart showing an edge matching degree calculation process in the basic operation. FIG. 6 is an explanatory diagram of an example of the edge variation position detection process. FIG. 7 is an explanatory diagram of an example of the edge variation position detection process. FIG. 8 is an explanatory diagram of an example of the edge variation position detection process. FIG. 9 is an explanatory diagram of an example of the edge variation position correction process. FIG. 10 is an explanatory diagram of an example of the edge variation position correction process. FIG. 11 is an explanatory diagram of an example of the edge variation position correction process. FIG. 12 is an explanatory diagram of an example of the edge variation position correction process. FIG. 13 is an explanatory diagram of an example of the edge variation position correction process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Imaging part 2 ... Image acquisition part 3 ... Control part 4 ... Memory | storage part 5 ... Output part 6 ... Difference area extraction means 7 ... Edge detection means 8 ... Edge fluctuation position detection means 9 ... Edge fluctuation position correction means 10 ... Edge matching Degree calculation means 11 ... Moving background object detection means 12 ... Intruder detection means

Claims (5)

Sequentially acquires the monitor image of the captured monitoring space, the pre-Symbol dynamic background object present in the monitoring space, an image sensor for detecting the phenomenon that the disappearance in a particular pixel on the monitoring image and the appearance of the edges is repeated ,
Edge detection means for detecting the presence or absence of an edge at each position of the monitoring image and storing the edge position;
Edge fluctuation position detection means for detecting an edge fluctuation position that has changed three or more times , the presence or absence of the edge detected by the edge detection means within the first predetermined time in the past being a first predetermined number of times ;
A difference area extracting means for extracting a difference area between the monitoring image and the background image stored in advance;
An edge matching degree calculating means for calculating a matching degree between the edge position and the edge fluctuation position in the difference region;
A moving background object detecting means for detecting the difference region having the matching degree equal to or greater than a predetermined value as a moving background object;
An image sensor comprising: Positions where the number of times the edge detection means did not detect an edge within the second predetermined time in the past having a length equal to or longer than the first predetermined time are excluded from the edge fluctuation position. The image sensor according to claim 1, further comprising an edge variation position correcting unit that performs the operation. A position adjacent to the edge standing position by detecting an edge standing position where the number of times the edge detecting means detects an edge within a third predetermined time including the first predetermined time is equal to or more than a third predetermined number of times. The image sensor according to claim 1, further comprising an edge fluctuation position correcting unit that excludes the edge fluctuation position from the edge fluctuation position. Furthermore, it comprises an intruder detection means for extracting an intruder feature amount from the monitoring image in the difference area, and detecting the intruder when the feature amount exceeds a predetermined reference,
The image sensor according to any one of claims 1 to 3 , wherein the intruder detection unit changes the predetermined reference to a high level when the moving background object detection unit detects the moving background object. Intruder detection means for extracting an intruder feature amount from the monitoring image in the difference area where the moving background object detection means did not detect a moving background object, and detecting the intruder when the feature amount exceeds a predetermined reference image sensor according to any one of claims 1 to 3 comprising a.

Images