JP5726595B2 - Image monitoring device - Google Patents

Description

  The present invention relates to an image monitoring apparatus that detects the number of people existing in a monitoring space such as the vicinity of important monitoring objects and monitors whether or not an emergency such as a burglary has occurred, and in particular, an image that prevents erroneous detection of the number of people. The purpose is to provide a monitoring device.

  2. Description of the Related Art Conventionally, there has been known an image monitoring apparatus that detects an emergency such as a burglary in a monitoring space from an image input from an imaging apparatus and automatically outputs a report to the outside. In general, suspicious individuals such as robbers aim to seize money by conquering the surveillance area, and often appear by multiple people to increase the probability of success. Therefore, in the image monitoring apparatus, when detecting a suspicious person such as a robber who appears in a plurality of persons, it is necessary to count the number of persons in the image.

Here, in order to count the number of persons from an image, it is effective to count the number of “faces” present in the image. In some cases, it is not possible to correctly count the persons who cannot properly capture the “face” (for example, landscape or backward). In addition, suspicious individuals such as robbers often hide their real faces with a balaclava or full-face helmet, and it may not be possible to extract facial features such as the eyes and nose and mouth. It is difficult to count the number of people.
In other words, in counting the number of people using “faces”, it is difficult to accurately measure the number of people in a surveillance camera image in which a person with various faces or a person whose face is hidden is mixed.

  Therefore, for example, Patent Document 1 is known as a technique for extracting not the “face” but the shape of the head itself on the image. Japanese Patent Application Laid-Open No. 2004-228561 discloses a technique for extracting a human head region as a region similar to an elliptical shape by regarding the human head as an ellipse in order to determine the presence of a person. In Patent Document 1, an edge image is generated from an input image, and matching processing using an elliptical template is performed on the edge image, thereby extracting a region where the edge shape is substantially elliptical as a head region of a human body.

JP 2010-186274 A

By detecting the human body using the head region as in Patent Document 1, it is possible to detect the presence of a person with various face orientations or a person whose face is hidden, and the number of people can be counted. It becomes.
However, if the head of a human body is extracted as an elliptical area and this is counted as the number of persons in the image, the following problem occurs.
That is, when trying to extract the head region with an elliptical edge, a shape similar to an ellipse caused by clothing stitches or heels may be extracted as the head region, and there is a risk of erroneous detection of the number of people.

This problem will be specifically described with reference to FIG. FIG. 6 shows an input image from the imaging device at a certain point in time. In the input image, a person with his arms folded facing diagonally forward is captured. In this case, when the head region is extracted with an elliptical edge, as shown by the dotted line in the figure, in addition to the human head, an ellipse is formed at the shoulder stitch seen from the side and the bent elbow edge. Shape edges can be detected and three head regions can be extracted.
Therefore, in this case, if the number of people is measured based on the number of head regions, the number of people is erroneously measured even though only one person is included in the image.

  The present invention aims to solve such a problem, and when detecting a suspicious person such as a robber who appears in multiple people using the number of head regions, the influence of a false detection factor such as a shoulder is detected. An object is to provide a reduced image monitoring apparatus.

  In order to achieve the above object, an image monitoring apparatus according to the present invention is an image monitoring apparatus that detects an emergency in a monitoring space, and periodically captures the monitoring space to acquire a monitoring image; A head extraction unit that extracts an input head region similar to a human head shape from the monitoring image; a number detection unit that detects the number of people in the monitoring space based on the number of input head regions; and the detection An emergency determination unit that determines an emergency if the number of persons performed is greater than or equal to a predetermined number, and the number detection unit performs another input within a predetermined distance for the input head region extracted by the head extraction unit When the head region exists, the count flag is invalidated in any region, and the number of persons in the monitoring space is detected based on the input head region in which the count flag is not invalidated.

  In such a configuration, the image monitoring apparatus extracts an input head region corresponding to the head of a person from the monitoring image, and checks whether there is another input head region located within a predetermined distance for this input head region. . If the other input head area exists, the count flag is invalidated for any one of the input head areas according to a predetermined rule, and the number of persons is detected based only on the input head area where the count flag is not invalid. In other words, the image monitoring device acts to detect the number of persons with a valid count only for any one of the input head regions for a plurality of input head regions where the distance between the input head regions is within a predetermined distance. .

  If a shape similar to an ellipse caused by clothing stitches or heels is mistakenly extracted as the head region, the incorrect head region and the genuine head region (person's head) are located at a certain close position. It will be in a relationship. Therefore, in such an image monitoring apparatus, not all of the head regions in a certain close positional relationship are counted, but the counting is enabled only for any one of the head regions, so that the same person can be counted. It is possible to prevent the number of people from being counted twice in the wrong head region and the genuine head region, thereby improving the accuracy of the number of people counting.

  In the image monitoring apparatus according to the present invention, the number detection unit may invalidate a count flag in a region located relatively below in the input head region and the other input head region. Good.

  A shape similar to an ellipse generated by clothing stitches, heels, or the like can be detected in a clothing portion below a genuine head region (person's head). Therefore, according to such a configuration, for the head regions in a close positional relationship, the counting is valid only for the head region located relatively above, thereby extracting the true head and counting the number of people. It is possible to improve accuracy and head region extraction accuracy.

  Furthermore, in the image monitoring apparatus according to the present invention, the number detection unit is within the predetermined distance from the input head region, and the other input head region whose area difference from the input head region is within a predetermined range. May be compared, and the region located relatively below may be invalidated as a count flag.

  When a monitoring image is taken with the horizontal direction as the depth direction, when attention is paid to the same person in the image, there are few area differences between the head and the shoulder or elbow of the clothes. On the other hand, when comparing this person with other persons separated in the depth direction (perspective direction) in the image, the sizes of the heads and the sizes of the shoulder elbows are different. Therefore, according to such a configuration, the head region in the close positional relationship is compared with another person's head by determining whether the count is valid or invalid by comparing an area having a certain area difference as a comparison target. Thus, it is possible to extract the true head from the region extracted for the same person. This makes it possible to improve the accuracy of counting people and the accuracy of extracting the head region.

  Furthermore, in the image monitoring apparatus of the present invention, the image monitoring apparatus further includes a variable region extracting unit that extracts the variable region by comparing the monitor image with a past image, and the number detection unit has a predetermined distance with respect to the input head region. It may be determined whether or not the other input head region exists within the same fluctuation region.

  According to such a configuration, the head area in the close positional relationship is compared with another person's head by determining whether the count is valid or invalid by comparing the extracted head area as the same fluctuation area. Thus, it is possible to extract the true head from the region extracted for the same person, and it is possible to improve the accuracy of counting people and the extraction accuracy of the head region.

  Further, in the image monitoring apparatus of the present invention, a storage unit for storing human face feature information in advance, and a face for determining whether the input head region has face feature information or not using the face feature information. A feature determination unit, and the number detection unit may not invalidate the count flag for the input head region determined to have a face feature by the face feature determination unit.

  According to such a configuration, it is possible to prevent the head area that is determined to be a face from being set to be invalid by counting the person's facial appearance with respect to the extracted head area. It is possible to improve the extraction accuracy of the head region.

According to the present invention, when counting the number of persons according to the number of head areas, not all the head areas in a certain close positional relationship are counted, but only one of the head areas is counted. Can be made effective.
As a result, even if a shape similar to an ellipse caused by clothing stitches or heels is mistakenly extracted as the head region, the same head region overlaps with the wrong head region and the true head region. It is possible to prevent the number of people from being counted.

It is the schematic which shows the whole structure of the monitoring system by the image monitoring apparatus of this invention. It is a block diagram which shows the structure of the image monitoring apparatus of this invention. It is the figure which showed typically the number detection process of the image monitoring apparatus of this invention. It is a flowchart which shows the suspicious person monitoring process of the image monitoring apparatus of this invention. It is a flowchart which shows the number detection process of the image monitoring apparatus of this invention. It is the schematic which shows the state of the head area | region extraction in background art.

Embodiments of the present invention will be specifically described below with reference to the drawings.
In the present embodiment, the case of monitoring the vicinity of important monitoring objects in an office area such as a financial institution or a store as a monitoring area (monitoring space) is illustrated, but the scope of the present invention is not limited to this, and It may be used to monitor the entire room or street as a monitoring area.

FIG. 1 is a configuration diagram showing a monitoring system 1 using the image monitoring apparatus of the present invention.
FIG. 1 schematically shows a monitoring area in which a storage 2 having a storage space inside a safe or the like and an image monitoring device 3 are installed as important monitoring objects. As shown in FIG. 1, the monitoring system 1 of the present embodiment communicates a monitoring device 3 placed on the upper part of a storage 2 fixedly installed on the floor of a monitoring area and a remote monitoring center 5. It is configured to be connected via the line network 4.

In the housing of the image monitoring apparatus 3, a monitoring camera 31 as an imaging unit is arranged in the same direction as the storage door. The monitoring camera 31 is arranged so that a person approaching the storage 2 can be imaged using the front of the storage door as an imaging field of view.
Note that the monitoring camera 31 is not limited to the example built in the housing of the image monitoring device 3, and a monitoring camera is separately installed in the upper space or the rear wall of the storage 2 so that at least a person in the monitoring area can take an image, It may be configured to be connected to the image monitoring apparatus 3. The image monitoring device 3 is not limited to the example of being placed on the upper part of the storage 2 and may be set in the upper space of the storage 2 or the rear wall surface.

  The monitoring camera 31 captures a monitoring area at every predetermined shooting interval (for example, 0.2 seconds), generates image data as a monitoring image, and outputs it. Based on the information output from the monitoring camera 31, the image monitoring apparatus 3 refers to periodically captured image data to determine whether or not an emergency situation caused by a suspicious person has occurred around the storage 2 To do.

  In the present embodiment, an emergency means that there is a risk of damaging the maintenance of a monitored area such as a burglar, and specifically, there are more than a predetermined number of people around the storage. In general, suspicious individuals such as robbers aim to seize money by conquering the surveillance area, and often appear by multiple people to increase the probability of success. Further, there are cases where a plurality of people including the user exist around the storage due to acts such as threatening a user who has the authority to unlock the storage. Therefore, in the present embodiment, when it is detected that there are more than a predetermined number of people around the storage, it is determined that there is a suspicious person and an emergency situation has occurred.

  In particular, as a characteristic matter in the present embodiment, the image monitoring apparatus 3 extracts a head region similar to the head shape of a person and detects the positions of a plurality of head regions when detecting the number of people from image data. An appropriate head area is counted after evaluating whether the count is invalid from the relationship. Then, the counted number of head regions is regarded as the number of people in the monitoring area, and it is determined whether or not there are more than the specified number. When there are more than the specified number of persons, the image monitoring apparatus 3 determines that an emergency has occurred due to a suspicious person, and detects an emergency signal and a suspicious person indicating that an emergency has occurred in the monitored area. The image data used for the transmission is transmitted to the remote monitoring center 5.

  The monitoring center 5 is a facility provided with a center device 51 operated by a security company or the like. The center device 51 is composed of one or a plurality of computers, and realizes the function of the monitoring center 5 related to the present invention. In the monitoring center 5, various devices are controlled by the center device 51, the emergency signal received from the image monitoring device 3 is recorded, and the image data transmitted from the image monitoring device 3 is displayed on the display 52, so that the monitoring staff Monitors multiple monitored areas.

<Image monitoring device>
Next, the configuration of the image monitoring apparatus 3 will be described with reference to FIG. FIG. 2 is a block diagram illustrating a configuration of the image monitoring apparatus 3.
The image monitoring device 3 is connected to the communication network 4 so as to be communicable with the monitoring center 5, and is fixedly placed on the upper part of the storage 2.

  The image monitoring apparatus 3 includes a monitoring camera 31 that captures the front of the storage, a communication unit 32 connected to the communication network 4, a storage unit 33 including an HDD and a memory, an MPU, a microcomputer, and the like. And a control unit 34 that controls each unit.

  The monitoring camera 31 generates a monitoring image of the monitoring area including the front of the storage. The monitoring camera 31 converts an image signal input from the image sensor into a digital signal at a predetermined shooting interval (for example, a cycle of 0.2 seconds, that is, 5 fps (frame / second)), performs compression encoding processing, and performs a predetermined standard. Image data compliant with (for example, JPEG standard) is generated. As a wavelength band used for image capturing, a visible light wavelength capable of capturing a color image is used. The image data generated by the monitoring camera 31 is output to the control unit 34.

  The communication unit 32 is connected to the center device 51 via the communication line network 4 and communicates with the monitoring center 5. When the control unit 34 determines that there is a suspicious person in the monitoring area, the communication unit 32 transmits an emergency signal including its own address information and the image data stored in the storage unit 33 to the monitoring center 5.

  The storage unit 33 is configured by a ROM, a RAM, or an HDD, stores address information for identifying itself and various programs, and further stores various information for operating the image monitoring apparatus 3. Specifically, the storage unit 33 determines the presence of a suspicious person, a reference image serving as background information for extracting a moving object, face feature information storing feature information of various facial images acquired in advance, and A suspicious person image for storing the image at the time and a count flag indicating whether or not to count the number of people for each head region are stored.

  The reference image is comparison reference information used to extract a moving object in the monitoring area in comparison with image data (hereinafter also referred to as an input image) input from the monitoring camera 31 in a fluctuation area extraction process described later. It is image data acquired by capturing an image of a monitoring area when there is no driver.

  The face feature information is the feature information for determining the face image of a person. For the face image obtained by capturing images of faces of various men and women from multiple directions in advance, the brightness pattern of each part such as the entire face and eyes, nose, mouth, etc. It is stored as facial feature information. Also, the brightness pattern of the face image stored in the face feature information is not only the face image of a real person, but also a person wearing sunglasses, a person wearing a mask, a person whose face is hidden by hand, and a balaclava It also includes image information of the face and head of a person who is assumed to be a suspicious person, such as a person or a person wearing a full-face helmet, and the brightness associated with the attributes of the face image as the input source for these various face images The pattern is stored. The attributes of the face image are the “original face” that does not hide the face, and “Sunglasses + mask”, “hand”, “balancing cap”, “full face”, etc. This is information indicating the state of the image.

The suspicious person image is stored in the case where it is determined by the emergency determination process by the control unit 34 that the current image data includes a suspicious person (there are more than the allowable number of persons).
The count flag is flag information that is set for each unique label of the head region and is used by the control unit 34 in the number detection process. Details of the count flag will be described later.

  The control unit 34 includes a microcomputer including a CPU, a ROM, a RAM, and the like and peripheral circuits thereof, and controls each unit described above. For this purpose, the control unit 34 includes, as a functional module realized by the microcomputer and a computer program executed on the microcomputer, a variable region extraction unit 341 that extracts a person region in the input image, and an input head from the input image. A head extraction unit 342 that extracts a human head region as a partial region; a face feature determination unit 343 that determines whether the head region in the input image is a head having facial feature information; A number detection unit 344 that detects the number of people based on the head region, and an emergency determination unit 345 that determines whether or not an emergency has occurred in the monitored area.

  The fluctuation region extraction unit 341 performs image processing on the input image to detect a person who becomes a moving object. The variation area extraction unit 341 compares the input image with the reference image stored in the storage unit 33, extracts the variation pixels whose luminance value variation is equal to or greater than a predetermined value, and sets a substantially continuous variation pixel group as one. Group as one variable area.

  The head extraction unit 342 extracts the region of the person's head part as the input head region within the variation region extracted by the variation region extraction unit. The head extraction unit 342 extracts the edge component of the fluctuation region, detects an elliptical edge distribution that approximates the contour shape of the face in the edge image data, and determines the elliptical region surrounded by the edge distribution as the head of the person. Extract as a partial area. Various methods described in Japanese Patent Application Laid-Open No. 2005-25568 and Japanese Patent Application Laid-Open No. 2010-286274 by the present applicant can be adopted for the extraction process of the head region.

  Specifically, the head extraction unit 342 applies a known edge extraction filter such as a Sobel filter to the fluctuation region of the input image to obtain an edge strength and an edge angle (an edge direction with respect to the horizontal) to obtain an edge strength image. And generate an edge angle image. Then, the head extraction unit 342 performs shift matching on the edge strength image and the edge angle image using a plurality of ellipse templates having different sizes stored in advance, and obtains the similarity between the ellipse template and the edge component, A position having a high degree of similarity is extracted as a head region of a person, and is labeled with a label unique to each head region.

Note that the head region extraction method is not limited to this, and various known methods can be used. For example, the head region may be extracted using HOG (Histograms of Oriented Gradients) feature values. In this case, a luminance gradient direction histogram is obtained for each of a plurality of cells constituting the variable region to obtain a nine-dimensional feature vector, and the cell feature vector is normalized in the block region of the plurality of cells. A head region is extracted by obtaining a multidimensional feature vector for each block region and comparing it with a feature value of a head shape learned in advance. As a reference for HOG features, N. Dalal and
B. Triggs, "Histograms of oriented gradients for
human detection ", Proc. of IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), pp.886-893, 2005, etc.

The face feature determination unit 343 covers the face of the head region extracted by the head extraction unit 342 with a face image of the face, sunglasses, a mask, a hand, or the like, or wearing a full-face helmet or the like. It is determined whether the face image is a face image in which a part or all of the part is hidden and the human phase is not known (no facial feature information).
The face feature determination unit 343 divides the extracted head region into a plurality of rectangular regions in the input image, and stores the divided head regions in the face feature information for each divided rectangular region and each block obtained by integrating adjacent divided rectangular regions. Pattern matching with the luminance pattern of the face image is performed. Then, from the attribute of the brightness pattern calculated with a high degree of similarity, an identification result as to which attribute face image is included or not included for each head region with a unique label Output.

For example, if a facial feature part (feature information) such as the eyes and nose and mouth can be extracted, it is determined that the face image has a “real face” attribute that does not conceal the face. When dark pixels are concentrated at the center in the direction and bright pixels are concentrated below the vertical direction, it is determined that the face is concealed as a face image having the “sunglasses + mask” attribute, and facial features such as the eyes, nose and mouth Cannot be extracted, and there are few edges in the head region, it is determined that the face is concealed as a face image having the “full face” attribute. Further, when a high similarity cannot be obtained with any of the patterns stored in the face feature information, it is determined that the head region is not a human head (a face image is not included).
The identification result of the face feature determination unit 343 is stored in the work area on the storage unit 33 in association with the input image and the unique label of the head area.

Note that the face image determination processing by the face feature determination unit 343 is not limited to this, and various known methods can be used. For example, an attribute identifier of the head region may be determined by an Adaboost discriminator using Haar-like features. The Haar-like feature is a luminance difference between a plurality of adjacent divided rectangular areas arbitrarily set in the head area of the input image. The Adaboost classifier includes a plurality of weak classifiers and a strong classifier that is determined by integrating the determination results of the weak classifiers. Each weak classifier calculates a different Haar-like feature from the input image region, and based on the calculated Haar-like feature, which attribute image is included in the head region or the face Outputs the identification result of whether an image is not included. Each weak classifier is generated by learning the Haar-like feature of the face image stored in the face feature information, and acquires the attribute of the face image accordingly. For details on Haar-like features and Adaboost classifiers, see, for example, Paul Viola and Michael Jones, "Rapid Object Detection using a
Boosted Cascade of Simple Features ", IEEE CVPR, vol.1, pp.511-518, 2001.

The number-of-people detection unit 344 evaluates the head region with the unique label from the positional relationship between the plurality of head regions, and counts the appropriate head region.
If a shape similar to an ellipse caused by clothing stitches or wrinkles is mistakenly extracted as the head region, these erroneous head regions are located on the clothing part below the genuine head region (the person's head). Can be detected. Therefore, the number detection unit 344 examines the distance between the head regions, and among the plurality of head regions existing in a certain distance range, only the uppermost one is effective, and the other heads located below The number of head regions is counted with the region invalid. At this time, an isolated head region in which no other head region exists in the close distance range is counted as valid.

Specifically, the number detection unit 344 pays attention to any head region in the input image (attention head region), sets a search region that is a predetermined distance in the vertical and horizontal directions from the center of gravity, and It is determined whether there is another head area extracted from the same fluctuation area as the attention area. The size of the search area is set according to the size of the target head area. For example, when the width W and the height H of the target head area are set, the search area is set as a rectangular area having a size of 3W / 3H. If there is another head region (comparison head region) including the center of gravity position in the search region, the comparison head region is compared with the target head region, and the difference in area is within a predetermined range. Whether the area of the comparative head region is within the range of 70% to 130% of the area of the head region of interest.
This is because when the monitoring image is taken with the horizontal direction as the depth direction as in this embodiment, the area difference between the head and the shoulder or elbow of clothing as a false detection factor is small if the person is the same person. Based on the knowledge that. On the other hand, if the persons are separated from each other in the depth direction (perspective direction) in the image, the sizes of the heads will be different even if they appear in the image overlapping one fluctuation region.

  Then, the number detection unit 344 compares the center-of-gravity position coordinates of the target head region and the comparative head region, and sets the count flag to “invalid” for any head region positioned downward in the vertical direction. More specifically, the number detection unit 344 refers to the identification result of the face feature determination unit 343 stored in the storage unit 33 for any head region located below and includes a face image. If it is determined that the face feature determination unit 343 determines that the face image is included, the count flag is set to “invalid”. do not do.

The count flag is flag information indicating whether or not the head region is to be counted, and is stored in the storage unit 33 in association with the unique label of the head region. The count flag is generated with an initial value of “valid” when a head region is extracted by the head extraction unit 342 and a unique label is attached. If the value of the count label is “valid”, the head region is counted, and if it is “invalid”, the head region is not counted.
The number detection unit 344 counts the number of head regions in which the count flag is set effectively, and outputs the number as the number of people existing in the monitoring area.

  The emergency determination unit 345 determines whether or not the number of people detected by the number of people detection unit 344 exceeds a preset allowable number of people. The allowable number of persons is the upper limit number of persons allowed to exist in the monitoring area (not abnormal), and is preset for each monitoring area and stored in the storage unit 33. In the present embodiment, the allowable number of people is set to two, for example. The emergency determination unit 345 compares the number of persons detected by the number of persons detection unit 344 with the allowable number of persons in the monitoring area, and determines that there is a suspicious person and an emergency has occurred if the number of detected persons is greater. .

  When the emergency determination unit 345 determines from the input image that there is an emergency situation in which a suspicious person exists, the emergency determination unit 345 generates an emergency signal including its own address information, and attaches image data that determines the presence of the suspicious person to the communication unit 32. To the monitoring center 5 from the communication unit 32. At this time, the image data determined to be present is stored in the storage unit 33 as a suspicious person image.

<Description of operation>
The operation of the monitoring system 1 configured as described above will be described with reference to the drawings. First, an outline of processing of the image monitoring apparatus 3 will be described with reference to FIG. FIG. 3 is a diagram schematically showing the number detection process by the image monitoring apparatus 3.

  In the image monitoring device 3, the monitoring camera 31 acquires image data at predetermined intervals (for example, 0.2 seconds) and inputs the image data to the control unit 34. When image data (input image) is input from the monitoring camera 31, the fluctuation region extraction unit 341 calculates the difference between the reference image of the storage unit 33 and the input image as shown in FIG. Regions 10, 11, and 12 are extracted. Next, the head extraction unit 342 detects an elliptical edge distribution within the fluctuation region, extracts an elliptical region surrounded by the edge distribution as a human head region, and is unique to each head region. Label with the label. In FIG. 3A, five regions f1 to f5 are extracted as the head region. At this time, count flags are generated for the head regions f1 to f5, and “valid” is stored as an initial value.

  Next, as shown in FIG. 3B, the face feature determination unit 343 performs pattern matching between the extracted head region and the face feature information stored in the storage unit, and the calculated similarity is high. The attribute of the luminance pattern is read, and an identification result indicating which face image of which attribute is included for each head region to which the unique label is attached or which face image is not included is output. In FIG. 3B, the head region f1 is determined to be “real face”, f2 and f3 are determined not to include a face image, f4 is determined to be “real face”, and f5 is “full face”. An example of determination is shown.

  Then, as shown in FIG. 3 (c), the number-of-people detection unit 344 sets a search region by paying attention to each head region, and another head region (comparison) extracted from the same variation region in the search region. Check if head area exists. In FIG. 3C, the search area A1 is set with the head area f1 as the target head area. In this case, since the center of gravity of the head region f2 extracted from the same fluctuation region 10 as the target head region f1 is included in the search region A1, the head region f2 becomes a comparative head region. Subsequently, the number detection unit 344 checks whether the area of the comparative head region f2 is within a predetermined range (for example, 70% to 130% of the target head region) with respect to the area of the target head region f1. If the condition is satisfied, the count flag of the comparative head region f2 positioned below in the target head region f1 and the comparative head region f2 is set to “invalid”. At this time, if the face feature determination unit 343 determines that a face image is included in the comparative head region f2, the count flag is not set to invalid.

  The number detection unit 344 sequentially executes the determination process for each head region. In the example of FIG. 3C, the count flag of the head region f3 is set to “invalid” in the process focusing on the head region f2, and the comparison head region is not found in the processing focusing on the head region f4. The count flag is not changed, and the count flag is not changed because the comparison head region is not found even in the process focusing on the head region f5. When the determination process is completed for each head region, the number-of-people detection unit 344 counts the number of head regions for which the count flag is “valid” and outputs the number as the number of people in the monitoring area. In the example of FIG. 3C, the head regions f1, f4, and f5 are effective, and the number of people is output as three. Then, the emergency determination unit 345 determines that there is a suspicious person because the number of detected persons is 3 with respect to 2 allowable persons, and notifies the monitoring center 5.

Next, the operation of the suspicious person monitoring process performed by the image monitoring apparatus 3 will be described with reference to FIGS. 4 and 5. FIG. 4 is a flowchart showing the operation of the monitoring program repeatedly executed by the control unit 34 of the image monitoring apparatus 3.
The image monitoring device 3 receives the output of the image data captured by the monitoring camera 31 every predetermined cycle (for example, 0.2 second cycle, that is, 5 fps (frame / second)) and inputs it to the control unit 34 (step ST1). The fluctuation area extraction unit 341 calculates the difference between the input image and the reference image in the storage unit 33 and extracts the fluctuation area (step ST2). If the fluctuation region cannot be extracted, the series of processes is terminated, and the process is executed from step ST1 at the next execution timing.

  In step ST3, the head extraction unit 342 extracts a person's head region from the variation region extracted by the variation region extraction unit 341. The extracted head region is labeled with a unique label. At this time, a count flag is generated corresponding to the unique label for each head region, and “valid” is stored as an initial value. If the head region cannot be extracted, the series of processing ends, and the processing is executed from step ST1 at the next execution timing. Next, the face feature determination unit 343 performs pattern matching between the extracted head region and the face feature information stored in the storage unit 33 to determine which attribute of the face image is included or whether the face image is The identification result of whether it is not included is output corresponding to the unique label of the head region (step ST4).

  In step ST5, the number-of-people detection unit 344 performs evaluation based on the positional relationship between the plurality of head regions, and counts the number of appropriate head regions as the number of people. The number detection process will be described later.

  The emergency determination unit 345 compares the number of people detected by the number of people detection unit 344 with the allowable number of people stored in advance, and when the number of detected people is larger than the allowable number of people (step ST6-Yes), there is a suspicious person in the monitoring area and the emergency It is determined that a situation has occurred (step ST7), and an emergency signal including its own address information is transmitted from the communication unit 32 together with image data for which the presence of a suspicious person has been determined (step ST12).

The basic operation of the image monitoring apparatus 3 has been described above.
Next, the number detection process in step ST5 of FIG. 4 will be described with reference to FIG. FIG. 5 is a flowchart of the number detection process by the number detection unit 344. In FIG. 5, the number-of-persons detection unit 344 focuses on any of the head regions extracted from the input image as a focused head region (step ST21), and a predetermined range (for example, focused on the center of gravity of the focused head region). A search area having a width W and a height H of the head area (in the range of 3W and 3H) is set (step ST22).

  Then, a search is made as to whether another head region exists in the search region, and if it exists, the other head region is set as a comparison head region (step ST23-Yes). Hereinafter, an example in which one comparative head region exists will be described. However, if a plurality of other corresponding head regions are detected, the plurality of head regions may be used as comparative head regions. In step ST23, if there is no other head region in the search region (step ST23-No), the process proceeds to step ST32 to determine whether or not to focus on another head region. Thereby, in the image, an elliptical area due to clothes stitches or heels existing within a certain distance from the head of the person is detected.

  At this time, if the other head region is not the same variation region as the target head region, it is excluded from the comparison head region (step ST24). In this way, by comparing the head regions extracted as the same variation region with each other, the head of one person is extracted from the same person while preventing the head from being compared with the head of another person. It is possible to extract the true head from the selected area.

  Next, in step ST25, the number-of-people detection unit 344 calculates the areas of the attention head region and the comparison head region, and the difference between the area of the attention head region and the comparison head region is within a predetermined range. Find out. If the area of the comparison head region is within a predetermined range with respect to the area of the attention head region (for example, 70% to 130% of the attention head region) (step ST26-Yes), comparison processing between the two is performed. continue. By this processing, even when persons who are separated in the depth direction (perspective direction) are extracted as the same fluctuation region, the same person is extracted while preventing heads of different persons from being compared. Extract the true head from the region.

  Subsequently, the number-of-people detection unit 344 determines which of the head region of interest and the comparative head region is positioned downward in the vertical direction (step ST27). Which is located below may be determined by comparing the positions of the center of gravity, and may be a comparison of the upper end position and the lower end position. This detects an elliptical shape caused by clothing stitches or heels that can be detected in the clothing part below the head of the person.

  When it is determined in step ST27 that the comparative head region is positioned below (step ST27-Yes), the comparison head is referred to by referring to the identification result of the face feature determination unit 343 corresponding to the unique label of the comparative head region. It is determined whether or not a face image is detected in the region (step ST28). If no face image is detected (step ST28-No), the count flag corresponding to the unique label of this comparison head region is set to “invalid” (step ST29). On the other hand, when it is determined in step ST27 that the head region of interest is located below (step ST27-No), it is determined whether or not a face image is detected in the head region of interest. In step ST30-No), the count flag corresponding to the unique label of the head region of interest is set to “invalid” (step ST31).

  Then, in step ST32, it is determined whether or not all the head regions have been noticed. If there is a head region that has not been noticed yet (step ST32-No), the processing is executed while paying attention to the head region. . When the processing focused on all the head regions is completed (step ST32—Yes), the number detection unit 344 determines the number of head regions whose count flag is not “invalid”, that is, the number of count flags “valid”. Count and output this as the number of people in the input image (step ST33).

  As described above, the number of persons detection unit 344 counts a flag for any one of the input head areas based on a predetermined rule when another input head area exists within a predetermined distance for the head area included in the input image. Is set to invalid, and the number of persons is detected based only on the head region where the count flag is not invalid.

  In particular, as a result of comparison with the head region within a predetermined distance, the count flag of the head region located relatively below is invalidated. Further, in the comparison of the head regions, those having a certain area difference are excluded from comparison, and those having a large area difference are excluded from comparison. Furthermore, in the comparison of these head regions, only the head region extracted from the same fluctuation region is targeted. In addition, the head region determined to be a face is excluded from the targets for which the count flag is invalidated.

  As a result, even when elliptical shapes caused by clothing stitches or heels that appear prominently on the shoulders and elbows are erroneously extracted as head regions, these are counted as the number of heads, that is, the number of people. This makes it possible to appropriately count the number of people in the image. Further, since the number of people is counted using the number of head regions, the number of people can be counted even when there is a user whose back head is photographed or a suspicious person who hides his / her face.

  The embodiments of the present invention have been described above by way of example, but the scope of the present invention is not limited to these embodiments, and can be changed or modified according to the purpose within the scope of the claims. is there.

DESCRIPTION OF SYMBOLS 1 Monitoring system 2 Storage 3 Image monitoring apparatus 31 Surveillance camera 32 Communication part 33 Storage part 34 Control part 341 Fluctuation area extraction part 342 Head extraction part 343 Face feature determination part 344 Number of persons detection part 345 Emergency determination part 4 Communication network 5 Monitoring center 51 Center device 52 Display 10, 11, 12 Fluctuating area f1, f2, f3, f4, f5 Head area A1 Search area

Claims (4)

An image monitoring device for detecting an emergency in a monitoring space,
An imaging unit that periodically images the monitoring space to obtain a monitoring image;
A head extraction unit that extracts an input head region similar to a human head shape from the monitoring image;
A number detection unit that detects the number of people in the monitoring space based on the number of the input head regions;
An emergency determination unit that determines an emergency if the detected number of persons is greater than or equal to a predetermined number of persons,
The number of persons detecting unit is
When another input head region exists within a predetermined distance with respect to the input head region extracted by the head extraction unit, the input head region and the other input head region are relatively downward An image monitoring apparatus characterized in that a count flag is invalidated in a region that is positioned , and the number of persons in the monitoring space is detected based on an input head region in which the count flag is not invalidated.
The number of persons detecting unit is
The other input head region that is within a predetermined distance from the input head region and whose area difference from the input head region is within a predetermined range is compared, and the region positioned relatively below is counted. The image monitoring apparatus according to claim 1 , wherein the flag is invalid.
further,
A fluctuation area extraction unit that extracts a fluctuation area by comparing the monitoring image with a past image,
The number of persons detecting unit is
3. The image monitoring apparatus according to claim 1, wherein the input head region is determined to determine whether the other input head region is within a predetermined distance and within the same fluctuation region. 4.
further,
A storage unit for storing face feature information of a person in advance;
A face feature determination unit that determines whether the input head region has facial feature information using the facial feature information;
The number of persons detecting unit is
Image monitoring apparatus according to any one of the face for the input head region of determination of the presence of facial features by the feature determining unit does not disable the counting flag claims 1 to 3.