JP2022037331A - Lost child detectors, lost child detection methods, and programs - Google Patents

Abstract

The present invention suppresses the processing load when detecting a lost child from an input image consisting of a plurality of frame images. A lost child detection device 12 includes a first motion vector detection unit 50 that detects a motion vector in an input image to detect a human area, and a feature point extraction unit that extracts feature points in the human area. 52, a size detection section 54 that detects the size of a person appearing in an input image from the distribution of feature points, a second motion vector detection section 56 that detects a motion vector at a feature point, and a feature point a feature point tracking section 58 that obtains the locus of the feature points from the motion vector at , a behavior detection section 60 that detects the characteristic behavior of the lost child based on the trajectory of the feature points, a size detection section 54, and a behavior detection section A lost child determining section 62 is provided, which determines whether a lost child is found based on the detection result of the section 60. The first motion vector detection section 50 detects a motion vector for a certain frame image, and the second motion vector detection section 56 detects motion vectors for a plurality of subsequent frame images. [Selection diagram] Figure 2

Description

The present invention relates to a lost child detection device, a lost child detection method, and a program for detecting a lost child based on an image captured by the image pickup device.

Conventionally, in facilities such as shopping malls and home centers, it has been studied to detect lost children by analyzing images captured by surveillance cameras.

Patent Document 1 is not a device specialized for detecting lost children, but it acquires movement trajectory information of a person based on a time-series image of an image acquisition device (surveillance camera, etc.), and features and behaviors of the person (characteristics and behaviors of the person (such as a surveillance camera). A device that acquires behavioral information that identifies (lost child, shoplifting, etc.) and predicts the future movement route of a person based on the information is disclosed. As a method for acquiring the movement trajectory information of a person, the use of optical flow is mentioned.

JP-A-2018-160219

By the way, by performing an optical flow on a moving image consisting of a plurality of frame images acquired by an image pickup device (detecting a motion vector of the moving image), the movement of the person in the moving image is captured, and the movement of the person is captured. A method of detecting a lost child based on the movement can be considered. However, if the optical flow is applied to all the frame images of the moving image and the entire area of each frame image, the processing load becomes large.

An object of the present invention is to detect a motion vector for a moving image (hereinafter, also referred to as an input image) so that a processing load can be suppressed when detecting a lost child in the input image.

The lost child detection device of the present invention has a first motion vector detection unit that detects a motion vector for an input image composed of a plurality of frame images to detect a human region, which is a region in which a person is reflected, and the human region. Size detection that detects the size of the person reflected in the input image from the feature point extraction unit that extracts the portion where the motion vector is relatively large near the end of the feature point and the distribution of the feature points. The unit, the second motion vector detection unit that detects the motion vector at the feature point, the feature point tracking unit that obtains the locus of the feature point from the motion vector at the feature point, and the locus of the feature point. Based on this, a person is determined to be a child from the size of the person detected by the motion detection unit that detects the characteristic behavior of the lost child and the size detection unit, and the motion detection unit determines that the child is the lost child. The first motion vector detection unit detects a motion vector for a certain frame image, and includes a lost child determination unit that determines the child as a lost child when the characteristic behavior of the above is detected. The second motion vector detection unit is characterized in that it detects a motion vector over a plurality of subsequent frame images.

According to this configuration, the motion vector is detected for a certain frame image and the feature points are extracted, and in a plurality of frame images following the certain frame image, the motion vector is detected only for the feature points. The processing load can be reduced as compared with the case of detecting the motion vector in the entire region.

In the lost child detection device of the present invention, the first motion vector detection unit may detect a motion vector in the entire region of the input image.

In the lost child detection device of the present invention, the first motion vector detection unit may detect a motion vector in a predetermined region where a person starts to appear in the input image.

According to this configuration, since the first motion vector detection unit detects the motion vector in the predetermined region of the frame image, the first motion vector detection unit detects the entire region of the frame image. Compared with the case of detecting a motion vector, the processing load can be reduced.

In the lost child detection device of the present invention, the predetermined region may be the end region of the input image.

In the lost child detection device of the present invention, the first motion vector detection unit may also detect a motion vector in the predetermined region of a plurality of frame images following the certain frame image.

According to this configuration, when a person starts to appear in the input image, the human area can be quickly detected and the feature points can be extracted.

In the lost child detection device of the present invention, the characteristic behaviors of the lost child are that the moving speed is faster than that of the surrounding people, that the lost child is moving in a direction different from that of the surrounding people, and that the head swings around. It may include at least one of a reciprocating action in a narrow area and a crying screaming.

The lost child detection method of the present invention includes a first detection step of detecting a motion vector for an input image composed of a plurality of frame images to detect a human region, which is a region in which a person is reflected, and the vicinity of the edge of the human region. In the extraction step of extracting a portion having a relatively large motion vector as a feature point, and a size detection step of detecting the size of a person reflected in the input image from the distribution of the feature points. A second detection step for detecting a motion vector at a feature point, a feature point tracking step for obtaining a locus of the feature point from a motion vector at the feature point, and a characteristic of a lost child based on the locus of the feature point. A person is determined to be a child from the size of the person detected by the action detection step and the size detection step, and the characteristic behavior of the lost child is detected for the child by the action detection step. A lost child determination step for determining the child as a lost child is provided. In the first detection step, a motion vector is detected for a certain frame image, and in the second detection step, a plurality of subsequent steps are detected. It is characterized by detecting a motion vector over a frame image.

The program of the present invention is a first motion vector detecting means for detecting a motion vector in an input image composed of a plurality of frame images to detect a human region, which is a region in which a person is reflected, near the end of the human region. A feature point extracting means for extracting a portion having a relatively large motion vector as a feature point, a size detecting means for detecting the size of a person reflected in the input image from the distribution of the feature points, and the feature. A second motion vector detecting means for detecting a motion vector at a point, a feature point tracking means for obtaining a locus of the feature point from a motion vector at the feature point, and a characteristic of a lost child based on the locus of the feature point. A person is determined to be a child based on the size of the person detected by the motion detecting means and the size detecting means, and the motion detecting means causes the child to have a characteristic behavior of the lost child. A program that causes a computer to function as a lost child determining means for determining a child as a lost child when detected, the first motion vector detecting means detects a motion vector for a certain frame image, and the motion vector is detected. The second motion vector detecting means is characterized in that the motion vector is detected over a plurality of subsequent frame images.

According to the present invention, it is possible to suppress the processing load when detecting a lost child from an input image.

It is a figure which shows typically the whole structure of the lost child detection system. It is a functional block diagram of a lost child detection device. It is explanatory drawing of the detection timing of the 1st and 2nd motion vectors. It is a figure which shows an example of the human area detected in the input image. It is a figure which shows an example of a feature point in an input image and the locus thereof. It is a figure which shows the example of the locus of the characteristic point of the state which a child is running. It is a figure which shows the example of the locus of the characteristic point of the state which a child is swinging around. It is a figure which shows the example of the locus of the characteristic point of the state which a child is moving in a direction different from the surrounding person. It is a figure which shows the example of the locus of the feature point of the state which a child goes back and forth in a narrow area. It is a figure which shows the example of the locus of the characteristic point of the state which a child is crying and raising. It is a figure which shows an example of the detection area of the 1st motion vector detection part in another embodiment. It is a figure which shows another example of the detection area of the 1st motion vector detection part in another embodiment. It is explanatory drawing of the detection timing of the 1st and 2nd motion vectors in another embodiment.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. The configuration described below is an example for explanation, and can be appropriately changed according to the specifications of the system and the device. Further, when a plurality of embodiments, modifications, and the like are included in the following, it is assumed from the beginning that those characteristic portions are appropriately combined and used. The same elements are designated by the same reference numerals in all drawings, and duplicate description is omitted.

FIG. 1 is a diagram schematically showing the overall configuration of the lost child detection system 10 according to the present embodiment. The lost child detection system 10 is a system for detecting a lost child 20 in a commercial facility such as a shopping mall or a home center. The lost child detection system 10 includes an image pickup device 14 installed in a commercial facility and a lost child detection device 12 that detects a lost child 20 based on an image captured by the image pickup device 14. The image pickup device 14 and the lost child detection device 12 are communicably connected to each other via a network 16 such as a LAN (Local Area Network). The image pickup device 14 and the lost child detection device 12 may be connected by a dedicated line or may be wirelessly connected.

The image pickup device 14 is a video camera such as a surveillance camera having a shooting function and a communication function, and is installed on a ceiling, a wall, or the like of a commercial facility to take a picture of a predetermined area. The image pickup device 14 takes an image at a predetermined frame rate, and a moving image composed of a plurality of frame images generated by the image pickup is transmitted to the lost child detection device 12 via the network 16. A plurality of image pickup devices 14 can be provided in a commercial facility.

The lost child detection device 12 includes a computer and detects the lost child 20 by receiving and analyzing a moving image from each image pickup device 14. When the lost child 20 is detected, the display of the lost child detecting device 12 or the display of another device indicates that the lost child has been detected, and the position where the moving image of the lost child is captured (position of the image pickup device 14). And the characteristics of the detected lost child (color and type of clothes, height, etc.) are displayed. At this time, a lamp (not shown) may be turned on, or an alarm may be issued from a speaker (not shown). The guardian of the lost child 20 can quickly reunite with the lost child 20 by contacting the information directly or through a guard of a commercial facility, an usher, or the like.

Inside the alternate long and short dash line in FIG. 1, an example of the hardware configuration of the lost child detection device 12 is shown. The lost child detection device 12 includes, for example, a processor 30, a ROM 36, a RAM 38, a hard disk drive (HDD) 40, a communication unit 42 such as a network card, an input unit 32 such as a mouse or keyboard, and a display unit 34 such as a display. Is connected to the bus 46 and is configured. The processor 30 may be composed of one or a plurality of CPUs, and the plurality of CPUs may be physically separated from each other.

In the lost child detection device 12, the processor 30 executes a program stored in the HDD 40 or the like, so that the first motion vector detection unit 50, the feature point extraction unit 52, the size detection unit 54, as shown in FIG. It functions as a second motion vector detection unit 56, a feature point tracking unit 58, a behavior detection unit 60, and a lost child determination unit 62. The program executed by the processor 30 can be provided not only by communication means but also by storing it in a computer-readable recording medium such as an optical disk or a USB memory.

FIG. 2 is a functional block diagram of the lost child detection device 12, and the figure also shows the flow of processing. Hereinafter, the lost child detection performed by the lost child detecting device 12 will be described in detail with reference to FIG. 2. Hereinafter, the moving image received from the image pickup apparatus 14 is also referred to as an input image.

As shown in FIG. 2, an input image composed of a plurality of frame images is input to the first motion vector detection unit 50 and the second motion vector detection unit 56. As will be described below, the first motion vector detection unit 50 detects the motion vector for the entire area of the frame image, whereas the second motion vector detection unit 56 detects the feature points in the frame image. Only detect motion vectors.

In FIG. 3, frame images arranged in chronological order are shown with numbers, and the first motion vector detection unit 50 sets motion vectors for the second, sixth, tenth, and so on frame images. It is shown that the second motion vector detection unit 56 detects the motion vector for the 3rd to 5th, 7th to 9th, and so on frame images. Note that detecting the motion vector for the nth frame image (n is an arbitrary integer of 2 or more) means detecting the motion vector of the nth frame image for the n-1st frame image. .. Hereinafter, the frame image is also referred to as a frame.

FIG. 3 shows an example in which the first motion vector detection unit 50 detects a motion vector every four frames. However, depending on the frame rate of the input image, for example, every several frames to several tens of frames. The motion vector detection unit 50 of 1 detects the motion vector of the frame, and the second motion vector detection unit 56 detects the motion vector for each frame between the frames in which the first motion vector detection unit 50 detects the motion vector. Should be detected.

First, the first motion vector detection unit 50 detects a motion vector for the entire region of the input image, and detects a human region, which is a region in which a person is reflected. The motion vector detected by the first motion vector detection unit 50 is also referred to as a first motion vector. In the example of FIG. 3, first, the motion vector of the second frame image is detected for all the pixels of the first frame image. FIG. 4 is a diagram showing an example of a human region 72 detected from the input image 70. The first motion vector detection unit 50 detects a region where the size of the motion vector is 0 as the background region 74, and detects a region where the magnitude of the motion vector is not 0 as the human region 72.

Next, the feature point extraction unit 52 in FIG. 2 extracts a portion having a relatively large motion vector near the end of the human region 72 as a feature point. FIG. 5 is a diagram showing an example of the feature point 80 in the input image 70, and the x mark represents the feature point 80 in FIG. 5 and each of the figures described below. The number of feature points 80 can be changed as appropriate.

The feature point extraction unit 52 acquires, for example, the average value of the size of the motion vector of each part near the end of the human region 72, and extracts all or some of the parts having a motion vector larger than the average value as feature points. do. Further, the feature point extraction unit 52 may, for example, extract a portion of some upper motion vectors having a large size from the motion vectors near the end of the human region 72 as feature points. However, as described below, since it is necessary to detect the size of the person reflected in the input image 70 from the distribution of feature points, the feature point extraction unit 52 is at least the upper end and the lower end portion of the human area 72. Is extracted as a feature point.

Next, the size detection unit 54 of FIG. 2 detects the size of the person reflected in the input image 70 from the distribution of the feature points of the human area 72. This is to detect whether the person in the human domain 72 is an adult or a child.

In the size detection unit 54, for example, when the vertical width of the human area 72 is smaller than a predetermined threshold value representing the maximum value of the height of the child, the person in the human area 72 is the child. Is determined to be. The size of the human area 72 changes depending on the magnification of the lens of the image pickup apparatus 14 and whether a person appears near the edge or the center of the image. It is good to change the threshold value.

If the size detection unit 54 determines that the person reflected in the input image 70 is only an adult, the lost child detection device 12 may not perform the process described below. As a result, the processing load of the device can be reduced.

Next, the second motion vector detection unit 56 shown in FIG. 2 detects the motion vector at the feature point. The motion vector detected by the second motion vector detection unit 56 is also referred to as a second motion vector. In the example of FIG. 3, first, the motion vector of the third frame image with respect to the feature point of the second frame image is detected, and then the motion vector of the fourth frame image with respect to the feature point of the third frame image is obtained. To detect. The second motion vector detection unit 56 repeats the same processing thereafter.

Next, the feature point tracking unit 58 of FIG. 2 generates a locus 82 of the feature points by concatenating the motion vectors for the same feature points among the motion vectors obtained by the second motion vector detection unit 56. do. 5 and 6A to 6E show an example of the locus 82 of the feature point.

Next, the behavior detection unit 60 of FIG. 2 detects the characteristic behavior of the lost child 20 based on the locus 82 of the feature point. 6A to 6E are diagrams showing an example of the characteristic behavior of the lost child 20 and the locus 82 of the characteristic point at that time. FIG. 6A is a diagram showing an example of the locus 82 of the feature point in the state where the child is running, and FIG. 6B shows an example of the locus 82 of the feature point in the state where the child is swinging while walking. FIG. 6C is a diagram showing an example of a locus 82 of a feature point in a state where a child is moving in a direction different from that of a person around him, and FIG. 6D is a diagram showing an example of a locus 82 of a feature point in which the child moves back and forth in a narrow area. It is a figure which shows the example of the locus 82 of the feature point of the state which is crying, and FIG. 6E is the figure which shows the example of the locus 82 of the feature point of the state which a child is crying.

The behavior detection unit 60 stores, for example, a lost child locus model having a feature point locus 82 as shown in FIGS. 6A to 6E in the lost child detection device 12 in advance, and the lost child locus model and the feature point. By comparing with the locus 82 of the feature points from the tracking unit 58, it is determined that the person in the human region 72 is taking the characteristic behavior of the lost child 20 when their similarity is high.

Further, the behavior detection unit 60 may determine the characteristic behavior of the lost child 20 without using the lost child locus model. For example, as shown in FIG. 6A, when a long feature point locus 82 can be obtained with a small number of frames or when the amplitude of the feature point locus 82 is large, it can be determined that a person is running. This determination is made, for example, by comparing a predetermined length threshold value with the length of the feature point locus 82, or by comparing a predetermined amplitude threshold value with the amplitude of the feature point locus 82. It can be realized by doing. Further, for example, when the locus 82 of the upper feature point of the human region 72 changes so as to swing left and right as shown in FIG. 6B, it can be determined that the person is swinging. .. Further, for example, when the loci 82 of the feature points of a plurality of human regions 72 appear in the input image as shown in FIG. 6C, the loci 82 of the feature points of one human region 72 are the features of a large number of human regions 72. When extending in the direction opposite to the locus 82 of the point, it can be determined that one person is moving in a direction different from that of many people. Further, for example, as shown in FIG. 6D, when the extending direction of the locus 82 of the feature point of the human region 72 is reversed many times during a small number of frames, a person moves back and forth. It can be determined that there is. Further, for example, when the locus 82 of the upper feature point of the human region 72 changes so as to sway up and down as shown in FIG. 6E, it can be determined that the child is crying and crying.

Next, in the lost child determination unit 62 of FIG. 2, the size detection unit 54 determines that the person in the human area 72 is a child, and the behavior detection unit 60 detects the characteristic behavior of the lost child in the person area 72. In this case, it is determined that the lost child 20 is reflected in the input image 70.

When the lost child 20 is detected by the lost child determination unit 62, the lost child detection device 12 identifies the color of the clothes of the lost child 20 from the color of the human area 72 of the input image 70, and pattern matching or the like with respect to the human area 72. By performing the above, the types of clothes of the lost child 20 and the like are specified, and they are displayed on the display as the characteristics of the lost child 20. Further, the lost child detecting device 12 receives information on the approximate height of the lost child 20 from the size detecting unit 54 and displays it on the display as a feature of the lost child 20. Further, the position information of the image pickup apparatus 14 that has captured the input image in which the lost child 20 is detected is displayed on the display. This is an image pickup device in which a table in which the identification information and the position information of each image pickup device 14 are associated with each other is stored in advance in the lost child detection device 12, and when the lost child detection device 12 receives an input image, the table is imaged. This can be realized by receiving the identification information of the 14 together and collating the identification information with the above table to obtain the position information of the image pickup apparatus 14.

Next, the operation and effect of the lost child detection device 12 described above will be described.

As described above, the first motion vector detection unit 50 detects the motion vector for all the pixels of the frame image (performs a dense optical flow), whereas the second motion vector detection unit 56 detects the motion vector. Detects motion vectors only for feature points (performs sparse optical flow). Therefore, the processing load of the second motion vector detection unit 56 is much lighter than that of the first motion vector detection unit 50.

According to the lost child detection device 12 described above, the motion vector is not detected for the entire area of all the frames of the input image, but as shown in FIG. 3, the first motion vector detection unit 50 is a frame. A motion vector is detected for the entire area of the image, the feature point extraction unit 52 extracts the feature points from the detection result, and the second motion vector detection unit 56 detects the subsequent frame images for a plurality of frame images. Since the motion vector is detected only at the feature points, the processing load of the entire device can be greatly suppressed. According to the lost child detection device 12 described above, real-time processing can be realized even for an input image having a high frame rate or a frame image having a large number of pixels. Further, even when the processing capacity of the processor 30 is low, the lost child can be detected.

As shown in FIG. 3, the reason why the first motion vector detection unit 50 periodically detects the motion vector for the entire area of the frame image is when a new person appears in the input image 70. This is because it is necessary to detect the human area 72 for that person and extract the feature points. In addition, when the color of the feature point changes due to a change in the way the light hits a person, it may not be possible to track the feature point, and at this time, it is necessary to extract the feature point again. ..

Next, another embodiment of the lost child detection device will be described. In the lost child detection device 12 described above, the first motion vector detection unit 50 detects the motion vector in the entire region of the input image, but in the lost child detection device of another embodiment, the first motion vector detection unit 50 detects the motion vector. , The motion vector is detected only in the area where the person in the input image is framed in (the area where the person starts to appear). FIG. 7 is a diagram showing an example of a region 90 (predetermined region) in which the first motion vector detection unit 50 detects a motion vector in this embodiment. In the figure, the region 90 is an input image. It is near the end of 70.

The range to be imaged by the image pickup device 14 is predetermined, and in the input image 70 of each image pickup device 14, the area 90 (the area where the person frames in) where the person starts to appear can be grasped in advance. The lost child detection device stores in advance a table in which the identification information of each image pickup device 14 and the area 90 are associated with each other, and when the lost child detection device receives the input image, the identification information of the image pickup device 14 that captures the input image is also stored. The area 90 can be grasped by receiving them together and collating the identification information with the above table. Then, the first motion vector detection unit 50 detects the motion vector only for the region 90 of the frame image, and the feature point extraction unit 52 extracts the feature points from the detection result, and the second motion vector. The detection unit 56 detects motion vectors of feature points for a plurality of subsequent frame images.

According to this another embodiment, since the first motion vector detection unit 50 detects the motion vector with respect to the predetermined region 90 of the frame image, the first motion vector detection unit 50 detects the entire frame image. Compared with the case of detecting a motion vector for a region, the processing load of the device can be reduced.

The area 90 in the input image 70 where a person starts to appear changes depending on the area in which the image pickup device 14 takes an image. Therefore, for example, when there is a corridor or an elevator exit near the center of the input image 70, as shown in FIG. The area 90 is set near the center of the input image 70. Further, although not shown, for example, when a descending escalator or stairs is reflected near the upper end of the input image 70, the area 90 may be set near the upper end of the input image 70.

In this other embodiment, the first motion vector detection unit 50 may detect the motion vector for the region 90 of all the frame images as shown in FIG. According to this configuration, when a person starts to appear in the input image 70, the human area 72 can be quickly detected and the feature points thereof can be extracted. For example, when the color of the feature point changes and the feature point cannot be tracked, the second motion vector detection unit 56 can immediately acquire the new feature point, and the new feature point can be tracked. Tracking will be possible soon.

In each of the above-described embodiments, the image pickup device 14 and the lost child detection device have separate configurations, but they may be integrated. For example, the image pickup device 14 may have the function of the lost child detection device described above.

In each of the above-described embodiments, the lost child detection device is assumed to process the moving image sent from the image pickup device 14 in real time, but the lost child detection device temporarily processes the moving image of the image pickup device 14. It may be recorded in, read out, and processed.

Further, in each of the above-described embodiments, the processor 30 of the lost child detection device is assumed to be a general-purpose CPU that operates according to a program, but a dedicated processor (for example, GPU: Graphics Processing Unit, ASIC: Application Specific Integrated Circuit). , FPGA: Field Programmable Gate Array, etc.) may be included.

10 Lost child detection system, 12 Lost child detection device, 14 Imaging device, 16 Network, 20 Lost child, 30 Processor, 32 Input section, 34 Display section, 36 ROM, 38 RAM, 40 Hard disk drive, 42 Communication section, 46 Bus, 50th 1 motion vector detection unit, 52 feature point extraction unit, 54 size detection unit, 56 second motion vector detection unit, 58 feature point tracking unit, 60 behavior detection unit, 62 lost child determination unit, 70 input image, 72 people Region, 74 background region, 80 feature points, 82 feature point trajectories, 90 regions.

Claims (8)

A first motion vector detection unit that detects a motion vector for an input image composed of a plurality of frame images and detects a human region, which is a region in which a person is reflected, and
A feature point extraction unit that extracts a portion having a relatively large motion vector as a feature point near the end of the human region, and a feature point extraction unit.
From the distribution of the feature points, a size detection unit that detects the size of the person reflected in the input image, and
A second motion vector detection unit that detects the motion vector at the feature point, and
A feature point tracking unit that obtains a locus of the feature point from a motion vector at the feature point,
A behavior detection unit that detects the characteristic behavior of a lost child based on the locus of the feature points,
When a person is determined to be a child from the size of the person detected by the size detection unit, and the behavior detection unit detects the characteristic behavior of the lost child for the child, the child is regarded as a lost child. It is equipped with a lost child judgment unit for judgment.
The first motion vector detection unit detects a motion vector for a certain frame image, and the second motion vector detection unit detects a motion vector over a plurality of subsequent frame images.
A lost child detection device characterized by this. The lost child detection device according to claim 1.
The first motion vector detection unit detects a motion vector for the entire region of the input image.
A lost child detection device characterized by this. The lost child detection device according to claim 1.
The first motion vector detection unit detects a motion vector in a predetermined region where a person starts to appear in the input image.
A lost child detection device characterized by this. The lost child detection device according to claim 3.
The predetermined area is an edge area of the input image.
A lost child detection device characterized by this. The lost child detection device according to claim 3 or 4.
The first motion vector detection unit also detects a motion vector in the predetermined region of a plurality of frame images following the certain frame image.
A lost child detection device characterized by this. The lost child detection device according to any one of claims 1 to 5.
The characteristic behaviors of the lost child are that it moves faster than the people around it, that it is moving in a different direction from those around it, that it swings around, that it moves back and forth in a narrow area, and that it moves back and forth in a narrow area. , Including at least one of crying
A lost child detection device characterized by this. The first detection step of detecting a motion vector for an input image consisting of a plurality of frame images to detect a human area, which is an area in which a person is reflected, and
An extraction step of extracting a portion having a relatively large motion vector as a feature point near the end of the human region, and an extraction step.
A size detection step that detects the size of a person reflected in the input image from the distribution of the feature points, and
The second detection step of detecting the motion vector at the feature point, and
A feature point tracking step for obtaining a locus of the feature point from a motion vector at the feature point,
A behavior detection step that detects the characteristic behavior of a lost child based on the locus of the feature points, and
When a person is determined to be a child from the size of the person detected by the size detection step, and the characteristic behavior of the lost child is detected for the child by the behavior detection step, the child is regarded as a lost child. With a lost child judgment step to judge,
In the first detection step, a motion vector is detected for a certain frame image, and in the second detection step, a motion vector is detected over a plurality of subsequent frame images.
A lost child detection method characterized by this. A first motion vector detecting means, which detects a motion vector for an input image composed of a plurality of frame images and detects a human region, which is a region in which a person is reflected.
A feature point extraction means for extracting a portion having a relatively large motion vector near the end of the human region as a feature point.
A size detecting means for detecting the size of a person reflected in the input image from the distribution of the feature points.
A second motion vector detecting means for detecting a motion vector at the feature point,
A feature point tracking means for obtaining a locus of the feature point from a motion vector at the feature point.
A behavior detection means for detecting a characteristic behavior of a lost child based on the locus of the feature points, and a behavior detection means.
When a person is determined to be a child from the size of the person detected by the size detecting means, and the characteristic behavior of the lost child is detected for the child by the behavior detecting means, the child is regarded as a lost child. Lost child judgment means to judge,
It is a program that makes a computer function as
The first motion vector detecting means detects a motion vector for a certain frame image, and the second motion vector detecting means detects a motion vector over a plurality of subsequent frame images.
program.

Images