JP2002034027A - Method for recognizing picture and device for the same and monitoring system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent any erroneous warning due to any erroneous recognition under bad conditions such as rainy weather or luminance shortage. SOLUTION: Even when an object recognizing part 7 recognizes once any object other than the target of warning as an intruder being the target of warning, a warning delaying part 8 delays the warning only in a prescribed period under a condition that any object other than the the target of warning is preliminarily detected, or the notification of a warning delay signal is received from the other picture recognizing device or the like, and when it is re-recognized that the object in a picture is any object other than the target of warning during that time, a warning canceling part 9 cancels the warning. When any object other than the target of warning is detected, a delay signal outputting part 10 outputs the warning delay signal to the other picture recognizing device so that it is possible to prevent any erroneous warning due to the other picture recognizing device.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image recognition method and apparatus for processing an image to accurately detect an intruder or an intruder in the image, and a monitoring system using the apparatus.

[0002]

2. Description of the Related Art A conventional image recognition method and apparatus (for example, Japanese Patent Application Laid-Open No. H11-203567) process an image, calculate the area, speed, vertical and horizontal dimensions, etc., of an intruding object as feature quantities. Are averaged over a certain period of time, and when a determination condition based on a preset parameter is satisfied, the object is determined to be an alarm target and an alarm is immediately issued by an alarm means such as a buzzer.

[0003]

However, in the above-described conventional image recognition method and apparatus, the image quality of the image output from the image pickup device is degraded, for example, in bad weather or low illuminance. In the case where the feature amount cannot be obtained with sufficient accuracy, even if the object is not originally a target of the alert, the alert may be issued when the feature amount satisfies the determination condition based on the predetermined parameter. After such an erroneous alert, if the object approaches the imaging device, the feature amount can be obtained with relatively high accuracy, and even if it is found that the feature is not the object to be originally alerted, Cannot be cancelled, resulting in a false alarm.

The present invention has been made to solve such a problem, and an object of the present invention is to provide an image recognition method and apparatus which are less likely to give a false alarm, and a monitoring system using this apparatus. I do.

[0005]

According to the image recognition method of the present invention, when an object which is not an alarm target is detected from an image obtained by capturing an image, an alarm is issued to the alarm target detected from the image. Is postponed for a predetermined period, and when an object that is not a target of alarm is detected from the image during the period of postponing the alarm, the alarm for the postponed alarm is canceled. Things. According to this method, if an object that is not a target for alarm is detected from an image, even if an object that is a target for alarm is subsequently detected, there is a possibility of erroneous recognition. If an object that is not the alarm target is detected in advance, it is determined that the previous alarm target is erroneously recognized, and by canceling the postponed alarm itself, erroneous alarm can be prevented.

Further, the image recognition method of the present invention, when a notification postponement signal indicating that a non-reporting object has been detected from another system, is notified, the Postponing the alarm for a predetermined period, and when detecting an object that is not the alarm target from the image during the period in which the alarm is postponed, canceling the alarm for the alarm target that is postponed. It is assumed that. According to this method, if an alarm postponement signal indicating that an object that is not an alarm target has been detected is notified from another system, the possibility of erroneous recognition may be increased even if an object that is an alarm target is subsequently detected. There is a postponement of the report for a predetermined period, and if an object that is not a report target is detected during that time, it is assumed that the detection of the previous report target is erroneous recognition, and by canceling the postponed report itself ,
False alarms can be prevented.

Further, the image recognition method according to the present invention is characterized in that, when an object which is not an alarm target is detected from the image, an alarm postponement signal is notified to an image recognition apparatus of another system. . According to this method, when an object that is not a notification target such as a patrol vehicle is detected, a notification postponement signal is notified to an image recognition device that monitors an adjacent section, so that the other image recognition devices may detect an error. It is possible to prevent the alarm from being issued.

Further, the image recognition apparatus of the present invention switches from the non-delivery postponement mode to the postponement notification mode when detecting an object that is not to be reported from the image obtained by capturing the image. When an alarm target is detected from the image during the mode, an alarm postponing unit that postpones the alarm for the detected alarm target for a predetermined period, and during the period in which the alarm is postponed, When a non-reporting object is detected from the image, a report canceling unit for canceling the report about the postponed report object is provided. With this configuration, if an object that is not a target for alarm is detected from an image, even if an object that is a target for alarm is subsequently detected, the alarm is postponed for a predetermined period because there is a possibility of erroneous recognition. If an object that is not the alarm target is detected in advance, it is determined that the previous alarm target is erroneously recognized, and by canceling the postponed alarm itself, erroneous alarm can be prevented.

Further, the image recognition apparatus of the present invention switches from the non-delivery postponement mode to the report postponement mode when a notification postponement signal indicating that a non-reporting object has been detected is notified from another system. In addition, when an alarm target is detected from the image during the alarm postponement mode, an alarm postponing unit that postpones the alarm for the detected alarm target for a predetermined period, and postpones the alarm. And a notification canceling unit for canceling a report on the postponed report target when an object that is not a report target is detected from the image during the period. . With this configuration, when an alarm postponement signal indicating that an object that is not an alarm target has been detected is notified from another system, the possibility of erroneous recognition is increased even if an object that is an alarm target is subsequently detected. There is a postponement of the report for a predetermined period, and if an object that is not a report target is detected during that time, it is assumed that the detection of the previous report target is erroneous recognition, and by canceling the postponed report itself Erroneous notification can be prevented.

Further, the image recognition apparatus of the present invention further comprises a postponement signal output unit for notifying the postponement notification signal to an image recognition apparatus of another system when an object not to be notified is detected from the image. It is characterized by the following. With this configuration, when an object that is not to be alerted, such as a patrol vehicle, is detected, the alerting postponement signal is notified to the image recognition device that monitors the adjacent section, so that the other image recognition devices are erroneously notified. It is possible to prevent the alarm from being issued.

Further, the image recognition apparatus of the present invention displays the position of the object to be alerted superimposed on the image as a figure, and changes the display color of the figure depending on whether or not the alert is postponed. A display unit is provided. With this configuration, when a figure such as a rectangle is superimposed and displayed in an image for the purpose of indicating the position of an object, the display color of the figure is switched between during the postponement of the alarm and after the postponement, that is, after the alarm is issued. It will be possible to explicitly indicate that the report has been postponed.

Further, the image recognition device of the present invention predicts the moving direction of the object when detecting an object that is not the target of the alarm from the image, and selects a notification destination of the alarm postponement signal accordingly. A notification destination selecting unit is provided. With this configuration, by notifying the alarm postponement signal only to the image recognition apparatuses in the monitoring area in the moving direction of the passing object, it is possible to efficiently prevent the other image recognition apparatuses from erroneously issuing an alarm.

Further, the monitoring system of the present invention comprises a plurality of image pickup devices, and the plurality of image pickup devices for processing an image of the image pickup device and issuing an alert when an object to be alerted is present in the image. A recognition device, a plurality of frame buffers for delaying an image, transmission means for transmitting and receiving information between the devices, and a plurality of image display devices for displaying an image of the imaging device and a processing result of the image recognition device. It is characterized by having. With this configuration, when a certain image recognition device detects a passing object that is not a target to be processed by processing an image, a notification postponement signal is notified to another image recognition device through a transmission unit, thereby generating a false alarm. A monitoring system capable of effectively preventing the above can be realized.

[0014] The monitoring system of the present invention processes an image of the imaging device, a passing object detector that identifies a passing object that is not an alarm target, and notifies the alarm postponement signal, and an image of the imaging device. 9. The image recognition device according to claim 4, which issues a notification when an intruding object to be reported exists in the image, a frame buffer for delaying the image, and each device. And a transmission means for transmitting and receiving information between the imaging device and an image display device for displaying an image of the imaging device and a processing result of the image recognition device. This configuration realizes a monitoring system that can effectively prevent false alarms by notifying the image recognition device of an alarm postponement signal when the passing object detector detects an object that is not an alarm target. You can do it.

The monitoring system according to the present invention is characterized in that the passing object detector that identifies a passing object that is not a target of notification and notifies the postponement signal of the notification identifies the passing object by sound. It is. With this configuration, it is possible to realize a monitoring system that can effectively prevent erroneous alarms by detecting an object that is not an alarm target by a passing sound of the object and notifying the image recognition device of an alarm postponement signal. Becomes

The monitoring system according to the present invention is characterized in that a passing object detector that identifies a passing object that is not a target of notification and notifies the postponement signal of the notification identifies the passing object by magnetic fluctuation. Things. With this configuration, a surveillance system that can effectively prevent false alarms by detecting objects that are not the object of alarm due to magnetic fluctuations caused by passing and notifying the image recognition device of the alarm postponement signal You can do it.

[0017]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below with reference to FIGS. (Embodiment 1) FIG. 1 shows the configuration of an image recognition apparatus according to a first embodiment of the present invention. Imaging device 1
Captures an image of the monitored area and outputs it to the image recognition device 12. The image recognition device 12 processes an image obtained by the imaging device 1 and issues an alert when an intruding object to be alerted is detected. Although it is rare for the image recognition device 12 to misrecognize an object, it is rare for an object not to be reported due to a deterioration in an image from the imaging device due to weather conditions or illuminance conditions and a feature amount not being obtained correctly. May be erroneously recognized as an alarm target.

A video signal input from the image pickup device 1 is A / D converted in the image input unit 2 and stored in the input image memory 3. The object extraction unit 4 includes an image in which no object is stored in the background image memory 5, that is, a so-called background image,
By comparing the current input image stored in the input image memory 3 with the current input image, an area where a change has occurred is extracted as an object area. The feature amount calculation unit 6 calculates an area ratio,
Calculate the moving amount, moving speed, tracking period, and the like. The object identifying unit 7 compares the feature amount obtained by the feature amount calculating unit 6 with a preset threshold to identify the object, and instructs to issue an alert if the object is an object to be alerted. . Postponement part 8
When the object identification unit 7 detects an object that is not a notification target, and subsequently detects an object that is a notification target, the notification is postponed for a predetermined period. The alarm canceling unit 9 cancels the alarm during the postponement when the object identification unit 7 detects an object that is not to be issued during the alarm postponing period. Postponement signal output unit 10
When the object identification unit 7 detects an object that is not a notification target, the notification unit notifies a notification postponement signal to another system of image recognition devices. When displaying the position of the alert target using a graphic such as a frame, the display unit 11 switches the color of the graphic depending on whether the alarm is postponed or after the alarm.

Next, the image input unit 2, the object extraction unit 4, the feature amount calculation unit 6, the object identification unit 7, the alarm postponement unit 8, the alarm cancellation unit 9, the postponement signal output unit 10, and the display unit 11 The operation will be described in further detail.

The image input unit 2 converts the video signal output from the image pickup apparatus 1 into digital image data by A / D conversion. For example, 256-gradation monochrome image data having a pixel size of 192 pixels in the horizontal direction (X) and 242 pixels in the vertical direction (Y) is acquired. When the imaging apparatus 1 outputs digital image data, A / D conversion is not performed, but resolution conversion and the like are performed in accordance with subsequent processing.

The object extracting section 4 calculates the absolute difference S (X, Y) of the luminance between the input image I (X, Y) and the background image B (X, Y) which is an image in which no intruding object exists. The change in luminance on the image caused by the moving object is detected by the following equation (1). S (X, Y) = | I (X, Y) −B (X, Y) | (1) A threshold process is performed on the absolute value of the difference between the luminance values, and a threshold value TH is calculated.
Pixels exceeding bi are set to 1 and pixels not exceeding the threshold THbi are set to 0, for example, to be binarized, and a luminance change area in the image is extracted. When the image quality of the image output from the imaging device 1 is lowered and the contrast is low, the region of the object may not be correctly extracted in some cases.

The above-described object extraction method is known as a background subtraction method. Next, a labeling process is performed on the pixels at level 1, and a unique label is given to a region where adjacent pixels are at level 1 from each other. The area and the center of gravity of each label are obtained by the labeling process. When the object is extracted by the object extraction unit 4, a method of extracting an area of the object by pattern matching or optical flow may be used in addition to the background difference method.

The feature value calculator 6 calculates a feature value for identifying an object. Here, a case where the area ratio R and the tracking period TT are used as the feature amount will be described. First, the positions of the objects extracted by the object extraction unit 4 are temporally correlated and tracked. Tracking is established when the object position one frame before and the current object position are sufficiently close to each other and the change amount of the area or shape of the object area is within a predetermined range. Using the area Sob of the object extracted by the object extraction unit 4 during the tracking of the same object and the reference area Sst, the area ratio indicating the size of the object is calculated for each processing cycle by the following equation (2). The reference area is an area value on an image when a person of average physique is extracted at the same position where the object is extracted. The reference area can be calculated from the depression angle of the imaging device, the size of the imaging device, and the focal length of the lens. If an average intruder is extracted, the area ratio will be around 100%. Area ratio R = (Sob / Sst) × 100 [%] (2) The tracking period TT is an elapsed time from the start of tracking to the present. When the image quality is low, such as when the input image has low contrast, the feature amount may not be obtained with high accuracy. In the above description, the case where the area ratio and the tracking period are calculated as the feature amounts has been described. However, the movement amount, the movement direction, the movement speed, the luminance, and the like may be calculated.

The object identifying unit 7 determines whether the object is an object to be notified using the feature amount. Here, it is assumed that the object to be alerted is an intruder into the monitored area, and the objects not to be alerted are small animals such as cats and birds and patrol vehicles. FIG. 2 shows a flow of a process for determining an intruder using the area ratio R and the tracking period TT. Maximum area ratio Rm from the start of tracking
ax is stored, and after the tracking period TT exceeds the threshold TT _TH (step S1), the maximum area ratio Rmax exceeds the lower threshold R _TH1 at least once (step S2), and the upper threshold R _TH2 is once exceeded. If the number does not exceed (Step S3), it is determined that the user is an intruder. The maximum area ratio Rmax is the lower threshold R _TH1
Objects that are less than are determined to be small animals, noise, etc.
An object larger than the upper threshold _RTH2 is determined to be an object such as a vehicle. By increasing the threshold TT _TH for the tracking period and performing recognition over a long period of time, false recognition can be reduced.However, there is a delay between the occurrence of intrusion and the generation of an alert, Since a problem such as overlooking an intruding object that is not satisfied occurs, the tracking period threshold TT _TH is appropriately about several seconds. It is appropriate that the lower limit R _TH1 of the area ratio is about the maximum area ratio that a small animal or the like can take, and the upper threshold R _TH2 is about the maximum area ratio that an intruder can take. In the above description, an example has been described in which the object is recognized using the tracking period TT and the area ratio R as the feature amount. It is also possible to do.

The notification postponing section 8 postpones the notification when the object discriminating section 7 detects a non-reporting object and thereafter detects a notification target object by a predetermined period TA. The mode is the postponement mode, and when an object to be issued is detected in the postponement mode, the postponement is postponed for the postponement period TB. After the elapse of the predetermined period TA, a non-delivery postponement mode is set, and postponement of the notification is not performed. FIG. 3 shows a state transition in each mode. In the period TA during which the alarm is postponed, the time required for the object to move to another monitoring section is appropriate, and in many cases, about several tens of seconds is appropriate. Postponement period
For example, when an object approaches from a distant place, TB is about the time required for the object to move from a distant place to the camera side, and several seconds to several tens of seconds are appropriate.

The alarm canceling unit 9 cancels any postponed alarm, if an object other than the alarm target, for example, a patrol vehicle, is detected by the object identification unit 7. The alarm canceling unit 9 does not function except during the postponement period of the alarm.

The postponement signal output unit 10 outputs a postponement notification signal to another image recognition device that monitors an adjacent area when an object that is not a notification target, for example, a patrol vehicle, is detected by the object identification unit 7. Notice.

When the position on the image of the intruding object is indicated by a graphic such as a rectangular frame, the display unit 11 displays the rectangular frame in yellow when the report is postponed, and displays the rectangular frame in red after the report. By changing the color of the figure, it is easy for the observer or the like to understand that the report is postponed. In addition, the display unit 11 displays the identification result of the object by superimposing a character or a figure on the image.

In the above description, it is assumed that the object to be alerted is an intruder into the monitoring area, and the objects not to be alerted are small animals and patrol vehicles. The object that is not the alarm target may be a worker, a construction vehicle, or the like.

As described above, according to the first embodiment, when the object discriminating unit 7 detects an object that is not a notification target, the notification postponing unit 8 switches from the non-report postponing mode to the predetermined period TA. If the mode is switched to the alert postponement mode and an object to be alerted is detected after that, there is a possibility of erroneous recognition.Therefore, the alert is postponed for the predetermined alert postponement period TB, and If an object other than the target is detected, it is determined that the detection of the previous notification target is erroneous recognition, and the notification cancellation unit 9 cancels the postponed notification itself to prevent the erroneous notification. Can be.

When the postponement signal output unit 10 detects an object that is not a target of notification, for example, a patrol vehicle, in the object identification unit 7, the postponement of the notification is performed to another image recognition device that monitors the adjacent area. By notifying the signal, it is possible to prevent another image recognition device from recognizing an object that is not an alarm target as an alarm target object and causing an erroneous alarm.

(Embodiment 2) Next, a second embodiment of the present invention will be described. In the configuration of the second embodiment, the alarm deferral signal is input to the alarm deferral unit 8 of the image recognition device described in the first embodiment from an image recognition device of another system that monitors an adjacent area. Only different
The image input unit 2, input image memory 3, object extraction unit 4, background image memory 5, feature amount calculation unit 6, object identification unit 7, alarm cancellation unit 9, postponement signal output unit 10, and display unit 12 are implemented. As described in the first embodiment, the description is omitted.

In the alarm postponement unit 8 according to the second embodiment, the mode switching of whether or not to postpone the alarm is performed according to the presence / absence of the notification postponement of the alarm postponement signal from the image recognition apparatus of another system which monitors the adjacent area. I do. When the notification of the alarm postponement signal is received from the outside, the predetermined period TA is in the alarm postponement mode, and in the alarm postponement mode, the alarm is postponed by the alarm postponement period TB. After a lapse of a predetermined period TA from the last received notification postponement signal, the non-report postponement mode is set, and the postponement of the notification is not performed. FIG. 5 shows the state transition in each mode. Here, the notification postponement signal is a signal that notifies the image recognition device 12 when a patrol vehicle or the like that is not a target of notification is detected by an image recognition device of another system that monitors an adjacent section or various sensors. And Period during which the alarm is postponed
As for the TA, the time required for the object to move to another monitoring section is appropriate, and about tens of seconds is often appropriate.
The period TB is, for example, when an object approaches from a distance, is a time period in which the object moves from a distance to the camera.
Several seconds to several tens of seconds are appropriate.

As described above, according to the second embodiment, even if an object that is not an alarm target is once recognized as an intruder, the object that is not an alarm target is detected in advance. The notification is postponed if a notification is received from an image recognition device or the like, and if the object in the image is re-recognized as not being a target for the notification during that time, the notification is canceled. In addition, false alarms can be prevented.

(Embodiment 3) Next, a third embodiment of the present invention will be described. The third embodiment is a case where a part of the second embodiment is realized by software, and will be described with reference to the processing flow of FIG.

In the object extracting step F1, an object area is extracted by the background subtraction method. The background subtraction method is a method in which an input image is compared with a past image that is separately stored, that is, a background image, and an image region that has changed is extracted as an object region. In the feature amount calculating step F2, an area ratio, a moving amount, a moving speed, a tracking period, and the like are calculated as the feature amounts of the object. In the object identification step F3, the feature amount calculation step F2
Is compared with a threshold value set in advance to identify whether the object should be an alarm. Alert postponement process F4
Then, the notification is postponed for a predetermined period when the notification postponement signal is received. In the alarm canceling process F5, the alarm that is being delayed is canceled when an object that is not an alarm target is detected from the image.
In the postponement signal output step F6, when an object that is not a notification target is detected in the object identification step F3, a notification postponement signal is notified to another system of image recognition. The alarm postponement step F4, the alarm cancellation step F5, and the postponement signal output step F6 will be described in further detail below.

In the alarm postponement step F4, when an alarm postponement signal is being input from the outside or in the alarm postponement mode within a predetermined period TA after the input, an object to be output is detected in the object identification step F3. Is issued after a delay of TB for a predetermined period. In the case where the notification postponement signal has not been notified, or in the non-report postponement mode after a predetermined period TA has elapsed since the last notification, the notification is issued without postponement. The period during which the notification postponement mode is set is appropriately about the time required for the object to move to another monitoring section, and is often about several tens of seconds. Also, the period TB for delaying the alarm instruction signal is
For example, when the object is approaching from a distance, it is about the time required for the object to move from a distance to the camera, and several seconds to several tens of seconds are appropriate.

In the notification canceling step F5, when a patrol vehicle or the like which is not to be notified from the image is detected, if there is a report that is being delayed, this is canceled.

In the postponement signal output step F6, when it is determined in the object identification step F3 that there is an object not to be alerted such as a patrol vehicle in the image, the patrol vehicle or the like is sent to another adjacent image recognition device. A notification postponement signal indicating that an object not to be issued is present in the vicinity is notified.

As described above, according to the third embodiment, even if an object that is not a target of alarm is erroneously recognized as an intruder, it is later determined that the object is not a target of alarm. In such a case, the previous false alarm can be canceled.

(Embodiment 4) Next, a fourth embodiment of the present invention will be described with reference to FIG. The configuration according to the fourth embodiment is obtained by adding a notification destination selection unit 13 to the image recognition device 12 described in the first and second embodiments, and includes an image input unit 2, an input image memory 3, an object The extraction unit 4, the background image memory 5, the feature amount calculation unit 6, the object identification unit 7, the alarm postponement unit 8, the alarm cancellation unit 9, the postponement signal output unit 10, and the display unit 12 are the same as those in the first embodiment. The description is omitted because it is as described in the second embodiment.

The notification destination selection unit 13 first receives the result of the object tracking processing in the feature amount calculation unit 6 and predicts the moving direction of an object that is not to be notified. For example, as shown in FIG. 8A, on an image having the upper left as the origin (0, 0),
The position at the start of tracking is (X0, Y0), and the current position is (X0, Y0).
1, Y1). If the horizontal X direction is not considered for simplicity, if Y0 <Y1, the object moves downward in the image, that is, it is approaching the imaging device, and will continue to move downward in the image Can be expected. The prediction of the moving direction may be performed in the left-right direction of the image.

Next, the notification destination of the alarm postponement signal output from the postponement signal output unit 10 is switched according to the predicted moving direction. For example, as shown in FIG. 8B, a case where the monitoring area A0 is located farther than the monitoring area A1 of interest and the monitoring area A2 is located before the monitoring area A1, and A0 and A1 and A1 and A2 are adjacent to each other. Suppose. Assuming that there is a non-reporting object moving upward in the image capturing the monitoring area A1, that is, moving away from the imaging device 1, the object is now located in the monitoring area A0 adjacent to the back of the monitoring area A1. Is expected to move. Therefore, the notification postponement signal is notified only to the image recognition device monitoring the monitoring area A0. Conversely, when there is an object that moves downward in the image, the notification postponement signal is notified only to the image recognition device in the adjacent monitoring area A2 before the monitoring area A1. The notification postponing unit 8 of the image recognition device 12 that has received the notification of the notification postponement signal postpones the report for a predetermined period, and cancels the report if the object is re-recognized as being out of the report target during the postponement. In the part 9, the alarm is canceled.

As described above, according to the fourth embodiment, even if an object that is not a target of alarm in an image is erroneously recognized as a target of alarm once, the alarm is notified by the notification of the alarm postponement signal. Since the postponement, if the object is re-recognized as a non-reporting object within the predetermined period, it is possible to cancel the postponement of the postponement, and when notifying the postponement signal to another system, Notifying all the adjacent image recognition apparatuses, but not only the image recognition apparatuses that are expected to enter the object, it is possible to prevent the image recognition apparatus that does not enter the object from being delayed.

(Embodiment 5) Next, a fifth embodiment of the present invention will be described with reference to FIG. The fifth embodiment is a monitoring system using the above-described image recognition device 12, and the plurality of imaging devices 1a and 1b capture images of different adjacent monitoring areas A and B, respectively. The image recognition devices 12a and 12b that have received the images of the respective imaging devices 1a and 1b recognize the approaching objects in the images and issue an alert if there is an object to be alerted. The passing object detector SEN detects an object that is not the target of the alarm and outputs an alarm postponement signal. The frame buffer FB is
The moment of the alarm and the images before and after it are stored. Transmission means CAB
Transmits the images output by the imaging devices 1a and 1b, the image recognition devices 12a and 12b, the passing object detector SEN, and the alarm postponement signal. The alarm device CAU is provided with an alarm means such as a buzzer to issue an alarm. The display device MON displays images output from the image recognition devices 12a and 12b and the imaging devices 1a and 1b. As described in the first to fourth embodiments, the image recognition devices 12a and 12b process images input from the imaging devices 1a and 1b and issue an alert if an object to be alerted exists in the images. . Further, when an object not to be alerted is detected, an alert postponement signal is output to an image recognition device monitoring an adjacent section.

An example of the operation of the present system will be described. It is assumed that the object CAR not to be alerted moves from left to right in FIG. The object enters the monitoring area A within the angle of view of the imaging device 1a, and is recognized by the image recognition device 12a as a patrol vehicle that is an object that is not an alarm target. The image recognition device 12a is an image recognition device 1 in the adjacent monitoring area.
Notify 2b of the alarm postponement signal. The patrol vehicle then enters the monitoring area B, and the image recognition device 12b recognizes the patrol vehicle. Even if the image recognition device 12b erroneously recognizes as an intruder, the notification is postponed because the notification postponement signal has been notified in advance, and during the postponement, the previous object is a patrol vehicle again. If so, cancel the delayed alert. In addition, the image recognition device
In 12b, if it is not re-recognized as a patrol vehicle, that is, if the patrol vehicle does not enter the monitoring area B but there is actually an intruder, etc. become. The image at the moment of the alarm and the images before and after the alarm are recorded by the frame buffer FB, and are used by the supervisor to confirm the alarm factor.

It is conceivable that the alarm postponement signal is transmitted from the passing object detector SEN. When the passing object detector SEN detects an object that is not a reporting target such as a patrol vehicle, it notifies the nearby image recognition device of a report postponement signal. The image recognition device that has received the alarm postponement signal, even if erroneously recognized, receives the alarm postponement signal and postpones the alarm, and if it is later re-recognized as an object such as a patrol vehicle that is not an alarm target Cancel postponed alerts. As the passing object detector SEN, for example, a voice recognition device that recognizes sound when the vehicle passes, a magnetic sensor that captures a change in magnetism when the vehicle passes, and the like are considered.

As described above, according to the fifth embodiment, even if the object is once erroneously recognized as an object to be alerted, the image recognition device and the passing object detector that monitor the adjacent section are not required. If an object that is not an alarm target is detected, the alarm is delayed, and if the object in the image is recognized again and it is found that the object is not an alarm target, the alarm can be canceled. As a result, false alarms can be prevented.

[0049]

As described above, according to the present invention,
Even if an object that is not an alarm target is once recognized as an intruder, the object that is not an alarm target is detected in advance, or a notification of an alarm postponement signal is sent from another image recognition device or the like. If the notification is received, the notification is postponed for a predetermined period, and if the object in the image is re-recognized as being out of the notification target during that time, the notification is canceled so Prevention can improve the reliability of the monitoring device.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an image recognition device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing an object identification process according to the first embodiment of the present invention.

FIG. 3 is a schematic diagram showing a state transition of an alarm postponing unit according to the first embodiment of the present invention.

FIG. 4 is a block diagram illustrating a configuration of an image recognition device according to a second embodiment of the present invention.

FIG. 5 is a schematic diagram showing a state transition of an alarm deferral unit according to Embodiment 2 of the present invention.

FIG. 6 is a flowchart showing an image recognition process according to the third embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration of an image recognition device according to a fourth embodiment of the present invention.

8A is a schematic diagram when detecting a moving direction of an object according to a fourth embodiment. FIG. 8B is a schematic diagram when detecting a destination monitoring area of an object according to a fourth embodiment.

FIG. 9 is a block diagram showing a configuration of an intruder monitoring system according to a fifth embodiment of the present invention.

[Explanation of symbols]

 1 Imaging device 2 Image input unit 3 Input image memory 4 Object extraction unit 5 Background image memory 6 Feature value calculation unit 7 Object identification unit 8 Alarm postponement unit 9 Alarm cancellation unit 10 Postponement signal output unit 11 Display unit 12 Image recognition Device F1 Object extraction process F2 Feature amount calculation process F3 Object identification process F4 Notification postponement process F5 Notification cancellation process F6 Postponement signal output process 13 Notification destination selection section 1a, 1b Imaging device 12a, 12b Image recognition device CAR Object SEN Pass Object sensor FB Frame buffer CAB Transmission means CAU alarm device MON display device

 ──────────────────────────────────────────────────続 き Continuing from the front page (72) Inventor Katsumasa Onda 4-3-1 Tsunashima Higashi, Kohoku-ku, Yokohama-shi, Kanagawa Prefecture Inside Matsushita Communication Industrial Co., Ltd. 1-4-1 Tokai Passenger Railway Co., Ltd. (72) Inventor Shoji Morishita 1-4-1 Meiji Station, Nakamura-ku, Nagoya-shi, Aichi Prefecture Tokai Passenger Railway Co., Ltd. (72) Inventor Naoki Yamanashi Nakamura, Aichi Prefecture 1-4-1 Kuname Station Inside Tokai Passenger Railway Co., Ltd. (72) Inventor Yuji Suzuki 1-4-1 Meiji Station Nakamura-ku Nagoya-shi, Aichi Prefecture Tokai Passenger Railway Co., Ltd. F-term (reference) 5C054 AA05 CA04 CC03 ED07 ED17 FA00 FC04 FC05 FC12 FC15 FE16 FE28 GA04 HA18 5C087 AA02 AA16 BB74 CC46 DD03 EE08 FF01 FF04 GG40 GG66 GG70 GG83

Claims (12)

[Claims] When detecting an object that is not an alarm target from an image obtained by capturing an image, the alarm for the alarm target detected from the image is postponed for a predetermined period, and the alarm is postponed. An image recognizing method comprising: canceling a report on the postponed report target when an object that is not a report target is detected from the image during the period. 2. When an alarm postponement signal indicating that an object that is not an alarm target is detected is notified from another system, the alarm for the alarm target detected from the captured image is postponed for a predetermined period after that, An image recognizing method, wherein, when an object that is not an alarm target is detected from the image during the period in which the alarm is postponed, the alarm for the postponed alarm target is canceled. 3. The image recognition method according to claim 1, wherein, when an object that is not an alarm target is detected from the image, an alarm postponement signal is notified to an image recognition apparatus of another system. 4. When a non-reporting object is detected from an image obtained by imaging, the mode is switched from the non-reporting postponement mode to the report postponement mode, and the image is transmitted from the image during the report postponement mode. An alarm postponing unit that postpones the alarm for the detected alarm target for a predetermined period when an alarm target is detected, and an object that is not an alarm target from the image during the period in which the alarm is postponed. And an alarm canceling unit for canceling the alarm for the postponed alarm target when the alarm is detected. 5. When a notification postponement signal indicating that a non-reporting object has been detected is notified from another system, the non-report postponement mode is switched to the report postponement mode, and the report postponement mode is set. When an alarm target is detected from the image during the period, an alarm postponing unit that postpones the alarm for the detected alarm target for a predetermined period, and from the image during the period during which the alarm is postponed. An image recognizing device comprising: a report canceling unit that cancels a report on the postponed report target when an object that is not a report target is detected. 6. A postponement signal output unit for notifying an alarm postponement signal to another type of image recognition device when an object not to be notified is detected from the image. 5. The image recognition device according to 5. 7. A display unit which superimposes and displays the position of the notification target object as a graphic on an image and changes a display color of the graphic depending on whether or not the notification is postponed. The image recognition device according to any one of claims 4 to 6. 8. A notification destination selection unit for predicting a moving direction of an object when an object not to be notified is detected from the image and selecting a notification destination of the notification postponement signal in accordance with the prediction. The image recognition device according to claim 6, wherein: 9. A plurality of image pickup devices and a plurality of image pickup devices for processing an image of the image pickup device and issuing an alert when an object to be alerted is present in the image. 8. An image recognition device according to any one of the items 8, a plurality of frame buffers for delaying the image, a transmission unit for transmitting and receiving information between the devices, and an image of the imaging device and a processing result of the image recognition device. A monitoring system comprising a plurality of image display devices. 10. An image pickup apparatus, a passing object detector that identifies a passing object that is not an alarm target and notifies the alarm postponement signal, and processes an image of the image pickup apparatus to generate an intruding object that is an alarm target. 4. An alarm is issued when the image is present in an image.
6. The image recognition device according to any one of 6, 7, or 8, a frame buffer for delaying the image, a transmission unit for transmitting and receiving information between the devices, and an image of the imaging device and a processing result of the image recognition device. A monitoring system comprising an image display device for displaying. 11. The surveillance system according to claim 10, wherein a passing object detector that identifies a passing object that is not an alarm target and notifies the alarm postponement signal identifies the passing object by sound. 12. The surveillance system according to claim 10, wherein a passing object detector that identifies a passing object that is not an alarm target and notifies the alarm postponement signal identifies the passing object by magnetic fluctuation. .