JP2012014302A - Monitoring device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a monitoring device for effectively monitoring persons staying in an monitored area with identification of the number of persons that is particularly important in terms of safety management and security.SOLUTION: A monitoring device includes: image obtaining means 1 to obtain images of a monitored area and its surroundings; a number-of-person estimation unit 7 to estimate whether or not the number of persons in the monitored area is 1 or other than 1 among categories of 0, 1, or 2 or more, by analyzing the images inputted from the image obtaining means 1; and a stay reporting unit 8 to identify a stay with the estimated number of persons staying in the monitored area being 1 and activate an alarm, in order to effectively monitoring a stay of one person.

Description

  The present invention relates to a monitoring apparatus that acquires an image in a monitoring area using a monitoring camera or the like, processes the acquired image, and notifies a monitoring person or the like.

  In order to reduce the burden on the surveillance staff who visually check the surveillance camera, the behavior of the person at the time of the accident and abnormal behaviors such as dangerous actions leading to the accident are detected from the captured images and detected by the surveillance staff. Reporting systems are becoming more popular.

  Among the abnormal behaviors described above, a person staying in a specific place is, for example, a situation in which a person cannot move due to an accident or injury in a surveillance system for safety management, or a house in a surveillance system for security purposes. It is an abnormal behavior that has a high degree of importance to lead to situations such as preparing for entry.

  As a technique for detecting stay in the monitoring area, in Patent Document 1, the vicinity of the escalator deck is set as the monitoring area, and the larger the amount of difference between frames of the two-time captured images, the more the background difference of the monitoring area (from the monitoring area). A technique is disclosed that estimates that the larger the area of the difference between the input image and the background image, the greater the degree of congestion due to the greater number of passengers on the deck. In Patent Document 1, the degree of congestion calculated by the above method is divided into two levels, when it is quiet and when it is crowded, and when it is quiet, abnormal behavior of passengers such as sitting on the deck or mischief on the handrail is detected, and congestion is detected. In some cases, it is considered that there is a low possibility that an abnormal behavior that hinders passenger traffic has occurred, and the abnormal behavior is not detected. Instead, the contents of image recognition are switched so as to measure the passenger traffic.

  In general, Non-Patent Document 1 or the like is known as a technique for detecting the behavior and movement of a person by analyzing temporal changes in an acquired image, and in the field of escalators, Non-Patent Document 2 proposes a technique for detecting an abnormal behavior of a person by extracting spatial features and distinguishing and recognizing the movement of an escalator and the movement of a person.

JP 2000-7264 A

G. Bradski and J. Davis., `` Real-time recognition of activity using temporal templates '', Third IEEE Workshop on Applications of Computer Vision (December 1996) Yasuhiro Murai, Hironobu Fujiyoshi, Masato Nusui, “Detection of anomalous human behavior in an escalator scene based on spatio-temporal features”, Information Processing Society of Japan Research Report, CVIM, 164th, pp. 251-258 (September 2008)

  The situation in which there are no other people in the stay, especially those who are staying, is dangerous because, for example, in a surveillance system for safety management, it is not expected that the person involved in the accident will be rescued by the surrounding people at the time of the accident. However, in a monitoring system for security purposes, the crime is not expected to be suppressed by the eyes of other persons, so the risk level is high, and it is necessary for the observer to monitor it with priority.

  Further, in a monitoring system that detects the movement and behavior of a person by image processing as in Non-Patent Document 1 or 2, even if there is a change in the image similar to an abnormal behavior, the person is actually in the monitoring area. There is no need to issue a notification from the monitoring system in a scene that does not exist. This is because an unnecessary report makes the observer feel bothersome.

  From the above, in a monitoring system that detects stay by image recognition, it is desirable to detect stay and to estimate the number of people in the monitoring area at the time of stay based on the classification of 0, 1, 2 or more I found out.

Therefore, if the technique of Patent Document 1 is applied to the estimation of the number of people of 0, 1, 2 or more, the standard that the degree of congestion increases as the area of the background difference disclosed in Patent Document 1 increases, In such a situation that the illumination light is inserted at a low angle and the shadow extends, there is a problem that the area generated in the shadow portion is affected and erroneous determination is made. For example, when a person is on the escalator deck and a long shadow extends from the person, the area of the background difference increases because of the shadow. There is a possibility. In addition, in the situation where the shadow of a structure such as a handrail extends, the area of the background difference increases due to the shadow of the structure such as the handrail even though there is no passenger on the deck. There is a possibility that it is judged as crowded even though there is no one. Therefore, the technique disclosed in Patent Document 1 cannot accurately discriminate between the case where the number of people in the monitoring area is 0 and the case where there are 1 or 2 or more people.
In addition, in the standard disclosed in Patent Document 1, which is the standard that the degree of congestion is higher as the difference area between frames is larger, for example, in a situation where one passenger is going to get up after falling Depending on the movement of the passenger, the area of the difference between the frames is widened, and there is a possibility that it is actually crowded although there is only one passenger and it is quiet. On the other hand, in a situation where a large number of passengers cannot move due to extreme congestion, the area of the difference between frames becomes narrow due to the small movement of passengers. There is a possibility of misjudging that it is. Therefore, even if the technique disclosed in Patent Document 1 is applied to the estimation of the number of persons, it is not possible to distinguish between a case where there are a small number of passengers and a case where there are a large number of passengers. Cannot accurately estimate the situation beyond people.

  From the above, with the technology of Patent Document 1, it is difficult to accurately determine whether the number of people in the monitoring area is 0, 1 or 2 or more.

  On the other hand, Non-Patent Documents 1 and 2 are superior in the human recognition technology distinguished from the movement of the person and the movement of the escalator, but the classification of 0 person, 1 person, 2 persons or more in the monitoring area. There is no idea of estimating the number of people, and attention is paid to the number of people in the monitoring area, and no consideration is given to reducing reporting in a scene where there is no person.

  The present invention has been made in view of such conventional monitoring technology, and the main purpose thereof is to identify the situation of only one person, which is particularly important in terms of safety management and security. Thus, it is to provide a monitoring device that can effectively monitor the inside of a monitoring area by estimating the number of people.

  Another object of the present invention is to provide a device suitable for accurately estimating that the number of persons is one and estimating the number of objects to be detected as 0, 1, 2 or more. However, their purpose will be described in detail in the embodiments described below.

  In order to achieve the main object, the present invention is characterized in that an image acquisition means for acquiring an input image near the monitoring area, and an input image from the image acquisition means are processed to detect in the monitoring area. Number of persons estimation means for estimating whether the number of subjects is 0, 1, or 2 or more, and a retention when the estimated number of persons in the monitoring area is 1 It is provided with the stay alerting | reporting means to alert | report outside. As a result, the number of people in the monitoring area can be distinguished from the case where the number of people is 0 or 2 or more, and it can be specified and notified. Etc. can be urged.

  Here, the above-mentioned “specify stay when the estimated number of people in the monitoring area is one and report to the outside” means that when the number of staying people is one, distinguish it from the case of zero or two or more. If the number of people is other than one, it may not be issued, and if the number of people is other than one, it will be reported, but the content or length of the signal may be changed. It is a concept that includes the case where one person or two or more persons and one person are distinguished by adding a specific signal.

  Another feature of the present invention is a configuration for accurately estimating that the number of detection targets is one, and further estimates the number of detection targets by dividing them into 0, 1, 2 or more. However, the features of the preferred embodiment and its application examples will be described in detail in the embodiments described below.

  According to the present invention, it is possible to specify that the number of people in the monitoring area is one and call attention. Therefore, it is possible to effectively monitor the inside of the monitoring area from the viewpoint of safety management and security.

The functional block diagram of the monitoring apparatus which concerns on one Example of this invention. The equipment block diagram at the time of applying one Example of this invention to an escalator. The figure explaining the long-term change extraction part which concerns on one Example of this invention. The figure explaining the short-term change extraction part which concerns on one Example of this invention. The figure explaining the processing flow of the number-of-people estimation means which concerns on one Example of this invention. The figure which supplements the time relationship of the number estimation means based on one Example of this invention. The figure explaining the extraction method of the movement locus | trajectory of the person which concerns on one Example of this invention. The figure explaining the extraction method of the movement locus | trajectory of the shadow part which concerns on one Example of this invention. The figure explaining the extraction method of the movement locus | trajectory of the person which concerns on one Example of this invention. The figure explaining the extraction method of the movement locus | trajectory of the dense person which concerns on one Example of this invention. The figure explaining the example of a screen display of the alarm means which concerns on one Example of this invention. The figure which supplements the short-term change extraction part which concerns on one Example of this invention. The functional block diagram of the monitoring apparatus which concerns on the other Example of this invention. The apparatus block diagram at the time of applying the other Example of this invention to an escalator. The figure explaining the detection area of the other Example of this invention. The equipment block diagram of the monitoring apparatus which concerns on the further another Example of this invention.

Hereinafter, specific embodiments of a monitoring apparatus according to the present invention will be described with reference to the drawings. In the present embodiment, the case where the present invention is applied to an escalator as an object to be monitored will be described as an example. However, the monitoring device of the present invention is not limited to this, and any arbitrary moving or staying person such as an elevator or the inside or outside of a building. Applicable to location.

  FIG. 1 shows a functional block diagram of a monitoring apparatus according to Embodiment 1 of the present invention, and FIG. 2 shows an apparatus configuration example when applied as an escalator monitoring apparatus. In the first embodiment, the escalator deck is used as a monitoring area, and the movement of the escalator or the stay of a person near the handrail is detected as a behavior that may lead to an accident involving the step or the handrail. It is targeted.

  In FIG. 2, 101 is an escalator, 110 is a deck at a lower landing of the escalator, 103 is a camera that captures the area around the deck 110 from the rear of the deck 110, 104 is an image processing device, and 105 is a control of the escalator 101. The device 106 is an alarm device that warns the passenger of the escalator 101 by voice or light emission from a speaker, 107 is a monitor that performs screen output or voice output to a supervisor, 108 is an image or image processing of the camera 103 A transmission device 109 for remotely transmitting the video of the device 104 and a recording device 109 for continuously recording an input image of the camera 103 or an output image of the image processing device 104. The camera 103 is not limited to a visible light imaging device, and a device that acquires a two-dimensional image of the deck 110 and the vicinity of the deck 110, such as a near infrared imaging device, can be applied.

  Specific functional blocks of the image processing apparatus 104 are shown in FIG. In FIG. 1, an acquired image from the image acquisition unit 1 includes a background creation unit 2, a long-term change extraction unit 3, a short-term change extraction unit 4, a short-term change accumulation unit 5, a trajectory extraction unit 6, a person number estimation unit 7, a stay report. The image is processed by the image processing device 104 including the unit 8 and output to the alarm unit 9 (corresponding to the alarm device 106 in FIG. 2) and the control unit 10 (corresponding to the control device 105 in FIG. 2).

  The image processing apparatus 104 is realized by a predetermined computer housed in a housing, and can be connected to the functions of the camera 103 and the control apparatus 105, the alarm apparatus 106, the monitor 107, the transmission apparatus 108, and the recording apparatus 109. If possible, all or some of the functions may be shared with the control device 105, the alarm device 106, the monitor 107, the transmission device 108, and the signal processing device in the recording device 109.

  In FIG. 1, a detection target information storage unit 11 and a spot information storage unit 12 are storage units provided in the image processing apparatus 104, and include an image acquisition unit 1, a background creation unit 2, a long-term change extraction unit 3, and a short-term change unit. Data necessary for signal processing of each functional block such as the change extracting unit 4, the short-term change accumulating unit 5, the trajectory extracting unit 6, the number estimating unit 7, and the staying reporting unit 8 is stored. The detection target information storage unit 11 holds information such as the height and width in the space of the detection target and the movement speed. The point information storage unit 12 is information on areas on the input image of the camera 103 (for example, information such as which area on the input image corresponds to the deck, which area corresponds to the step, and which area is the detection area). ), Threshold values necessary for image processing for each point of the camera 103, and information on the height, width, and moving speed on the detection target image on the input image of the camera 103.

  In FIG. 1, the image acquisition unit 1 is configured to continuously acquire the input image of each frame from the camera 103, but the same function is achieved even when the recorded image of the camera 103 is played back. it can. The acquired image from the image acquisition unit 1 is processed and stored in the background creation unit 2, the long-term change extraction unit 3, the short-term change extraction unit 4, and the short-term change accumulation unit 5. Means for detecting the stay of the detection target in the.

  Hereinafter, the processing content of each functional block in FIG. 1 will be sequentially described.

  The short-term change extraction unit 4 extracts, as a short-term change, a portion that has changed in a short period on the image from the difference between the input images at two times [T−ΔT, T]. 4A and 4B show examples of input images at time T-ΔT and time T of the short-term change extraction unit 4. 111 is a handrail, 112 is a step, 113 is a person, and an arrow 117 is a person 113. The direction of movement. Between FIG. 4A and FIG. 4B, the person 113 has moved a little from the bottom to the top in the input image by the amount of walking toward the step 112. The short-term change extraction unit 4 extracts a difference (short-term change) in which the brightness has changed by a predetermined threshold value or more from the two-time image in units of pixels. Further, the short-term change extraction unit 4 extracts a difference connected region from the short-term change.

  FIG. 4C shows an example of the result of extracting the difference from FIG. 4A and FIG. 4B as an example of the extraction result of the short-term change extraction unit 4. A difference connection region 133 is a difference connection region due to the movement of the person 113. This connection area 133 corresponds to the outline of movement of the person 113 on the screen between two times [T−ΔT, T]. In the example of FIG. 4C, since the person 113 moves from the bottom of the screen to the top in the input image, the connection region 133 has a thickness corresponding to the amount of movement in the vertical direction. The short-term change extraction unit 4 extracts the connection region 133 from the short-term change as in the example of FIG. 4C, and from the connection region 133, the circumscribed rectangle 135 of the connection region 133 and the representative side 136 of the circumscribed rectangle 135 The representative point 137 of the representative side 136 is extracted.

  In FIG. 4C, when the connection area 133 captures the entire person 113, the base of the circumscribed rectangle 135 is on the deck 110 at the angle of view of the camera 103 that captures the deck 110 obliquely from behind as shown in FIG. Therefore, the representative side 136 is the bottom side of the circumscribed rectangle 135. The width of the representative side 136 corresponds to the width of the person 113. The representative point 137 is the midpoint of the representative side 136 and corresponds to the left and right centers of the feet of the person 113.

  Since the shape of the person 113 is vertically long, when the person moves on the screen from the bottom to the top, the movement of the person 113 in the horizontal direction between the two times [T−ΔT, T] is small. When a difference is taken between input images of T−ΔT, T], the upper and lower portions of the person 113 are connected such that the connected regions of the differences are 133a and 133b on the side surface of the person 113 as shown in FIG. It may be divided into two.

  As shown in FIG. 4D, even when the connected region of the difference of the person 113 is divided, the circumscribed rectangles 135a and 135b, the representative sides 136a and 136b, and the representative points 137a and 137b are separated from the individual connected regions 133a and 133b. If extracted, the coordinates of the feet of the person 113 can be extracted from at least one of them.

  In addition, when the person 113 moves on the screen in the horizontal direction, the connection area 133 may be divided into left and right, but in that case as well, a circumscribed rectangle for each of the divided connection areas, Find representative edges and representative points.

  In addition, since the connection area 133 may be mixed with minute noise such as fluctuations in the ceiling illumination reflected on the deck 110, the short-term change extraction unit 4 may use the vertical direction or the horizontal length of the circumscribed rectangle 135. However, when the size of the person on the input image of the camera 103 is less than the estimated threshold, the circumscribed rectangle 135 is rejected as noise.

  In the above description, the short-term change extraction unit 4 described the method of extracting the difference from the brightness of the same pixel on the input image at two times as the short-term change, but this is only an example, It is also possible to extract feature amounts such as RGB and HSI colors and edge feature amounts from the input images at two times, and extract the connected region 133 from the difference between the feature amounts of the same pixel.

  Further, as a method of extracting the connected region 133, in addition to the method of using the change in brightness or feature amount of the same pixel at two times, a method of extracting a moving part on the input image by the movement vector, the connected region 133 is extracted. May be extracted. Further, the short-term change extraction unit 4 extracts a motion from an input image at a plurality of times of two or more times, such as Motion History Image described in Non-Patent Document 1, in addition to a method of using an input image at two times. You may extract the connection area | region 133 by the method to do.

  The short-term change extraction unit 4 repeats the functions described above at a predetermined processing cycle. The short-term change accumulation unit 5 accumulates the representative points 137 extracted by the short-term change extraction unit 4 for each processing cycle in time series.

  Next, the background creation unit 2 generates a background image in which the person 113 does not exist by updating the background image at a cycle sufficiently slower than the continuation time of the person 113 during the abnormal behavior. As a method of creating a background image by the background creation unit 2, for example, a method in which an image of a frame with a small change in the short-term change extraction unit 4 is collected and the median of each pixel is obtained from the collected image within a predetermined time from the current time. Although possible, it is not limited to this example.

  The long-term change extraction unit 3 extracts a background difference, which is a difference between the input image and the background image, in order to extract a portion that changes between the input image and the background image of each frame as a long-term change. Ask. Further, the long-term change extraction unit 3 obtains a background difference connected region on the deck 110 from the obtained background difference.

  FIG. 3 is a diagram for explaining the long-term change extraction unit 3. FIG. 3A shows an example of a background image held by the background creation unit 2, 111 denotes a handrail, and 112 denotes a step. FIG. 3B is an example of an input image, and reference numeral 113 denotes a person who has fallen over because he failed to step 112. Since the person 113 has fallen near the step 112, there is a possibility that the foot of shoes or pants may be caught in the moving step 112, which is dangerous. FIG. 3C shows a portion where a change in brightness has occurred in the same pixel between the FIG. 3A and FIG. 3B over a predetermined threshold, and shows a long-term difference (background difference) in the input image. Show. In FIG. 3C, 120 is a detection area corresponding to the deck 110, 122 is a difference connection area where the step 112 is moved between the background image and the input image, and 123 is a difference connection area of the fallen person. Is shown.

  The long-term change extraction unit 3 obtains a background difference connected region that is a difference between the input image and the background image in the detection area 120 from the long-term change, and the area of the connected region is the detection target information storage unit 11. A value that is equal to or greater than a threshold value that assumes the size of the person to be held is selected as an effective connection area, and a circumscribed rectangle is obtained from the effective connection area. When the input image is shown in FIG. 3A, the connection area 123 is validated, and a circumscribed rectangle 125 of the connection area 123 is obtained. Further, the long-term change extraction unit 3 retains the center coordinates and vertical and horizontal lengths as information indicating the location of the circumscribed rectangle 125. Further, when a plurality of effective connected areas such as the connected area 123 are extracted, the long-term change extraction unit 3 extracts a plurality of circumscribed rectangles such as the circumscribed rectangle 125.

  Furthermore, the long-term change extraction unit 3 holds the time at which one or more circumscribed rectangles 125 extracted by the method described above are extracted. Further, when the detection of one or more circumscribed rectangles 125 continues at the same place, it is determined that the stagnation is occurring, and information indicating the time when the stagnation is occurring is held as information on the stagnation time. Here, the information on the staying time is information that can specify the time and length of the stay, such as the stay start time and end time, or the stay start time and duration. Including. In addition, the long-term change extraction unit 3 also holds information on the place where the stagnation has occurred.

  In FIG. 3, the difference 122 due to the movement of the step is outside the detection area 120 and thus does not affect the extraction of the circumscribed rectangle 125. Therefore, the range in which the long-term change extraction unit 3 calculates the difference may be limited to the detection area 120.

  In the above description, it has been described that the long-term change extraction unit 3 extracts the connected region 123 from the change in brightness of the same pixel in the background image and the input image as a long-term change. It is also possible to extract a feature quantity such as RGB or HSI or a feature quantity such as an edge feature quantity from each image, and extract the connected region 123 from the difference between the feature quantities of the same pixel. In addition to the method of using a single image as the background image, the brightness value of each pixel described by the probability distribution parameter is used as the background image, and when calculating the background difference, the brightness of each pixel is set. A method of calculating the difference between the background image described by the probability distribution parameter and the input image may be used.

  Next, the number estimating unit 7 estimates the number of people on the detection area 120 by the process flow shown in FIG. The processing contents will be described below with reference to processing steps S1 to S7 in FIG. 5 and a diagram supplementing the time relationship at the time of estimating the number of people shown in FIG. 6 as appropriate. 6A, 6B, and 6C correspond to cases where there are 0, 1, 2, or more people in the detection area, respectively.

  In S <b> 1, the number estimating unit 7 determines whether or not the dwelling has continued for a time τ that is the minimum dwelling time or longer based on the background difference extraction result or the dwelling extraction result in the long-term change extraction unit 3. To do. If the stay does not continue for the time τ or more (no in S1), it is assumed that the person 113 does not stay on the detection area 120 and is estimated to be zero (S5).

  In S1, if the staying continues for time τ or more, the process proceeds to processing step S2. The time relationship at this time is as shown in any one of FIGS. 6A, 6B, and 6C. When the retention start time is time T2, and the time T3 after time T2 is time T3, time T2 From the time T3 to the time T3.

  In S2, the number estimating unit 7 determines whether or not there is a movement locus that moves from the outside of the detection area 120 (the end of the detection area 120) toward the staying location in S1 immediately before time T2.

  Note that the extraction of the movement locus used for the determination in S2 is performed by the locus extraction unit 6. Estimated from the length of the detection area 120 and the minimum moving speed of the person 113, the longest passage time required for the person 113 to pass through the detection area 120 is defined as time π. The trajectory extraction unit 6 searches for the movement trajectory between time T1 and time T2.

  The movement trajectory extraction by the trajectory extraction unit 6 will be described by taking as an example a scene in which the person 113 stays on the deck 110 from time T2 to time T3 in a state of falling down as shown in FIG. When FIG. 3A is the background image and FIG. 3B is the input image, the long-term change extraction unit 3 determines the connection region 123 and the connection region 123 of the person 113 from the long-term change (background difference). A circumscribed rectangle 125 is extracted. 7A and 7B show that the person 113 moves from the outside of the deck 110 in the direction of the arrow 117 from the time Ta between the time T2 and the time T1 to the time T2. It shows the situation of walking to the vicinity.

  7A to 7B, the short-term change extraction unit 4 extracts short-term changes for each operation cycle, and the short-term change accumulation unit 5 outputs the short-term change extraction unit 4 for each operation cycle. A time series of (representative points 137) is accumulated. The trajectory extraction unit 6 extracts a movement trajectory by searching for corresponding points from the time series of the representative points 137 accumulated in the short-term change accumulation unit 5 retroactively from the time T2. As a method for searching for corresponding points, the nearest point of the representative point 137 at two adjacent times is searched, and corresponding points whose coordinates on the image of the two nearest representative points 137 are within an allowable error range. Although the method of concatenating can be considered, it is not limited to this. FIG. 7C illustrates an example in which the trajectory extraction unit 6 extracts the movement trajectory 153 of the representative point 137 between FIG. 7A and FIG.

  In the above description, the method in which the locus extraction unit 6 extracts the movement locus 153 from the time series of the representative point 137 has been described. However, the locus extraction unit 6 may obtain the movement locus 153 from data other than the representative point 137. . As an example, in Non-Patent Document 1, by accumulating changes (for example, inter-frame differences) of the person 113 at two times [T−ΔT, T] as in the connection region 133, the object is present during a predetermined time. A method of specifying the moved area is described, and the movement locus 153 can be obtained using this method, and is not limited to the above-described embodiment. In this case, the short-term change accumulation unit 5 may accumulate a time series of short-term changes (for example, interframe differences) instead of accumulating the time series of the representative points 137.

  For example, as shown in FIG. 7, in the period from time Ta to time T2 after time T1, the person 113 walks from outside the deck 110 to the vicinity of step 112, and in the period from time T2 to time T3, in FIG. As shown, in the scene where the person 113 stays in the vicinity of step 112, the time relationship shown in FIG. 6B is established, and from the outside of the detection area 120 as shown in FIGS. 3 and 7 during the period from time T1 to time T2. Since the movement locus 153 extends to the circumscribed rectangle 125, the determination in S2 is yes. In FIG. 6B, the generation period 202 of the movement locus 153 extending from the outside of the detection area 120 to the staying portion exists between the time Ta and the time T2.

  On the other hand, as shown in FIGS. 8A and 8B, in a scene where the shadow 115 continues to extend on the deck 110 from the handrail 111 between the time T2 and the time T3, even if going back from the time T2 to the time T1. The trajectory extraction unit 6 extracts the slight movement trajectory 155 shown in FIG. 8C by the fluctuation of the shadow 115 due to the fluctuation of the illumination condition near the escalator 101, or does not extract the movement trajectory at all. Therefore, the determination in S2 is no, and the number estimating unit 7 estimates the number of people staying in the deck 110 as zero (S5). In this case, the time relationship shown in FIG.

  If the determination in S2 is yes, the number-of-people estimation unit 7 considers that at least one person 113 has entered the deck 120 from time T1 to time T2 and stayed between time T2 and time T3. , Go to S3.

  Next, in S3, the number estimating unit 7 determines whether there are two or more movement trajectories 153 extending from the outside of the detection area 120 to any circumscribed rectangle 125 between time T1 and time T2. If there are two or more, it is considered that two or more persons entered between time T1 and time T2 and stayed between time T2 and time T3, and the number estimating unit 7 determines the number of persons staying in the deck 110. It is estimated that there are two or more people (S7). If the determination in S3 is no, the process proceeds to S4.

  Next, in S <b> 4, the number estimating unit 7 determines whether or not the movement locus (second person) has entered the detection area 120 from outside the detection area 120 between time T <b> 2 and time T <b> 3. FIGS. 9A and 9B show an example in which the second person 114 enters the deck 110 from outside the deck 110 between time T2 and time T3. In S4, the trajectory extraction unit 6 searches for a movement trajectory from the time series of the representative point 137 between time T2 and time T3 as in S2, and detects the detection area from outside the detection area 120 as shown in FIG. 9C. If at least one movement locus 154 of the second person 114 entering 120 is found, the determination in S4 is yes, and in addition to the person 113 entering between time T1 and time T2, in the period from time T2 to time T3. Assuming that the second person 114 or a larger number of persons have entered, the number estimating unit 7 estimates the number of persons as two or more (S7). The time relationship in this case is as shown in FIG. In FIG. 6C, 203 is a generation period of the movement locus 153 of the person 113, and 204 is an occurrence period of the movement locus 154 of the person 114.

  On the contrary, when the determination of S4 is no, it is considered that no new person 114 has entered between time T2 and time T3, and the number estimating unit 7 estimates the number of persons staying in the deck 110 as one ( S6).

  FIG. 6C shows an example in the case of two or more persons in the detection area 120, but the present invention is not limited to this, and other patterns exist. For example, as described in S3, when two or more persons enter between time T1 and time T2 and stay between time T2 and time T3, the generation period of the movement locus in FIG. Of 203 and 204, only 203 may exist (however, there is a movement trajectory for two people). Even in this case, it is possible to determine that there are two or more people by S3.

  Next, what kind of determination is performed in the case where the escalator 101 is very crowded will be described with reference to FIG. FIGS. 10A and 10B show that the escalator 101 is very crowded, and the crowd of people 143 repeatedly moves on and stops on the deck 110 between time T2 and time T3. The example of the scene which is slowly moving toward step 112 pulsatingly in the direction is shown. In the scenes shown in FIGS. 10A and 10B, the crowd of people 143 continues to exist in a wide area on the deck 110, and therefore, as shown in FIG. The circumscribed rectangle 175 in the detection area 120 of the connected area 173 extends over a wide range of the detection area 120, and the determination of S1 in FIG.

  Next, S2 will be described. First, since the crowd of people 143 continues pulsatingly on the deck 110 from time T1 to time T2 as in FIGS. 10 (a) and 10 (b), between time T1 and time T2. In addition, the short-term change extraction unit 4 extracts a connection region 133 and a representative point 137 (not shown). In such a case, the movement trajectory extracted from the crowd of people 143 often shows that the connected region 133 appears intermittently as a result of the person in the crowd of people 143 moving and stopping repeatedly. ) Of movement trajectories 163a, 163b, 163c. In addition, when the people in the dense people 143 move at the same time, the connected region 133 may be gathered around the detection area 120 and the representative point 137 cannot be extracted. In S <b> 2, the circumscribed rectangle 175 is detected in the detection area 120 if the movement trajectory entering the detection area 120 from the outside of the detection area 120, such as the movement trajectory 163 a, can be captured according to the movement state of the person near the boundary of the detection area 120. , The movement locus 163a extends to the circumscribed rectangle 175, and the determination is yes and the process proceeds to S3. On the contrary, depending on the movement state of the person near the boundary of the detection area 120, if the movement locus is only near the center of the detection area in S2 or there is no movement locus at S2, the determination of S2 is No, the number estimating unit 7 estimates zero (S5).

  If the determination in S2 is yes, then in S3, if there are two or more movement trajectories extending from outside the detection area 120 such as the movement trajectory 163a between time T1 and time T2, the determination is yes and the number of people is estimated. Part 7 estimates that there are two or more people (S7). If the determination in S3 is no, then in S4, if at least one movement locus extending from outside the detection area 120 such as the movement locus 163a exists between time T2 and time T3, the determination is yes and the number of people is estimated. Part 7 estimates that there are two or more people (S7). Here, if the time τ, which is the minimum continuation time of the stay, is set to be sufficiently long, an object that extends from the outside of the detection area 120 such as the movement locus 163a is reliably extracted, and the determination in S4 is prevented from being no. Is possible.

  As described above, the number of persons estimation unit 7 may not erroneously determine that there are 0 people in S2 with respect to the crowd 143 of people, but does not determine that the person is a subject to be reported in S4. . Note that the person estimation unit 7 erroneously determines that the number of persons 143 is 0 in S2, so that the time π that is the longest passage time is the same as the time τ that is the minimum duration of residence. If it takes longer, it can be suppressed.

  The time τ that is the minimum duration of the stay is estimated in advance. For example, a time during which it can be determined that the person 113 stays in an abnormal behavior such as falling down rather than a normal behavior such as when the person 113 stops and looks at the clock in S1 is time τ1, and in S3, the movement trajectory from the crowd of people 143 Assuming that the time τ2 is a time sufficiently long to extract 163a reliably, the time τ is the maximum value max (τ1, τ2) of the time τ1 and the time τ2. The time τ stores a value in the point information storage unit 12. The time π, which is the longest passage time, is obtained by dividing the length in the space of the deck 110 by the lowest moving speed in the space of the person 113 or by dividing the length in the image of the detection area 120 by the minimum in the image of the person 113. It can be calculated by dividing by the moving speed of.

  Supplementing the trajectory extraction unit 6, even when the connected region of the difference of the person 113 is divided into two or more like 133a and 133b as shown in FIG. 4D, the trajectory extracting unit 6 is connected to the connected regions 133a and 133b. By connecting the nearest points from the time-series data of the representative points 137a and 137b, the movement trajectory 153 can be extracted as in the case where the connected region of the person 113 is not divided as shown in FIG. At this time, two or more movement trajectories 153 may be extracted from the time series of the plurality of representative points 137a and 137b, but the movement trajectory 153 whose distance is within an allowable error considering the size of the person is obtained. By integrating them into one, it is possible to extract one movement trajectory from one person 113.

  Next, the staying notification unit 8 is limited to the case where the number estimating unit 7 determines that one person 113 is staying in the detection area 120, that is, no one or two or more people. It is specified that it is staying at the time, and a warning is output to at least one of the alarm unit 9 and the control unit 10.

  When the alarm unit 9 receives the notification from the stay alarm unit 8, the alarm unit 9 alerts the staying person by blinking the light of the alarm device 106 or outputting sound from the speaker, or to the people around the escalator 101. Make a rescue request. Further, when the alarm unit 9 receives the notification from the staying alarm unit 8, the alarm unit 9 displays the input image of the camera 103 on the monitor 107, transmits the image from the transmission device 108 to the external device, or the recording device 109. To control recording.

  Here, the alarm unit 9 may add a screen effect that blinks the input image of the camera 103 at a predetermined cycle so that the attention of the monitoring staff is attracted to the occurrence of the stay. The alarm unit 9 may draw and display the circumscribed rectangle 125 and the movement locus 153 on the input image displayed on the screen of the monitor 107. By watching the movement trajectory 153, the monitor can easily grasp where the person 113 entered from the deck 110 and how the person 113 moved on the deck 110.

  When drawing the screen, the alarm unit 9 outputs the drawn image or the drawn image and the input image of the camera 103 to the monitor 107, the transmission device 108, or the recording device 109. In the device configuration of FIG. 2, the alarm unit 9 can realize the function if there is at least one of the alarm device 106, the monitor 107, the transmission device 108, and the recording device 109.

  Finally, when the control unit 10 receives the notification from the alarm notification unit 8, the control unit 10 reduces the operation speed of the escalator 101, or slowly stops the escalator 101, thereby preventing an accident of a staying person, Reduce damage at the time of occurrence.

  With the functional configuration described above, in the first embodiment, while staying on the deck 110 is detected, the number of people on the deck 110 is estimated by classification of 0, 1, 2 or more, and the number of people on the deck 110 is determined. It is possible to report only when one person is staying, or to give a high importance to staying when the number of persons on the deck 110 is one. In addition, the present Example 1 demonstrates the same effect, when the number estimation part 7 estimates a number of persons in the division | segmentation of 0 person, 1 person, 2 or more by methods other than the processing flow of FIG.

  In the first embodiment, when S3 and S4 are omitted in the processing flow of FIG. 5 of the number estimating unit 7, 0 (S5), 1 (S6), or 2 or more (from the determination result of S1 and S2) The number of people can be estimated in two categories of S7). In the above-mentioned classification of 0 person and 1 person or more than 2 persons, when staying is continued (yes in S1), whether staying is due to other than the person 113 such as the shadow 115 as shown in FIG. 8 (S5), It is possible to determine whether it is due to the person 113 (S6 or S7). In the estimation of the number of people in two categories of 0 person and 1 person or more than 2 persons, the stay alarming unit 8 for stays other than the person 113 such as the shadow 115 is used in the monitoring device that reports stays. This has the effect of deterring notifications. It should be noted that the same effect is exhibited when the number estimating unit 7 estimates the number of persons in two categories of 0 person and 1 person or 2 persons or more by a method other than the flow of FIG.

  Further, in the first embodiment, when staying is continued in the processing flow of FIG. 5 of the number estimating unit 7 (yes in S1), when S2 is omitted and the process proceeds to S3 when S1 is yes, S3 is yes. The number of people is estimated in two categories: 0 (S5) or 1 (S6), and 2 or more (S7), when S3 is no, S4 is yes, and other cases Can do. In the estimation of the number of people in two categories of 0 or 1 and 2 or more, whether the stay is due to a crowd of people 143 as shown in FIG. 10 when the stay is continued (yes in S1) (S7), It is possible to determine whether this is not the case (S5 or S6). In the above-mentioned classification of 0 persons or 1 person and 2 or more persons, in a monitoring device that reports a stay due to abnormal behavior, in a situation where there are many users of the escalator 101 such as a crowded person 143 and there is no abnormal event. , There is an effect of preventing the staying notification unit 8 from reporting. It should be noted that the same effect is exhibited when the number estimating unit 7 estimates the number of people in two categories of 0 person or 1 person and 2 or more persons by a method other than the processing flow of FIG.

  In the first embodiment, when the stagnation occurs, the person estimation unit 7 determines whether the movement locus 153 from the outside of the monitoring area to the stagnation location, the position of the movement locus 153 and the long-term change, or the position information of the circumscribed rectangle 125. A behavior determination unit (not shown) that determines the type of behavior of the staying detection target (person 113) is provided. The stay notification unit 8 may perform control such as drawing on the screen output of the alarm unit 9 according to the result determined by the behavior determination unit. In FIG. 11A, the movement locus 153 extends to the vicinity of step 112, and the circumscribed rectangle 153 is in front of the deck 110 from the vicinity of step 112, so that the person 113 behaves in the vicinity of step 112. In this case, an example in which a fallen character string 311a is output near the step on the screen output 301 in a scene determined by the staying notification unit 8 is shown. Further, in FIG. 11B, since the movement locus 153 extends to the vicinity of the handrail 11 and the circumscribed rectangle 125 is applied to the handrail 111, the person 113 behaves mischievously on the handrail 111. The example in which the character string 311b of handrail mischief is output on the screen output 301 in the scene determined by the stay alarm unit 8 is shown. Reference numeral 312 is an example of a caution statement that alerts the monitor that there are no people around and is dangerous. The text of the character string 312 (dangerous alone) is merely an example, and other text may be used. 11A and 11B, components other than the character string 311a and the character string 311b, the circumscribed rectangle 125, the movement locus 153, and the character string 312 are not essential and may be omitted.

  Further, in the first embodiment, the staying alerting unit 8 adds importance to the alerting, and adds a high importance when the number of people estimating unit 7 estimates 1 person, and the number of people estimating unit 7 is two or more. May be issued with a low importance added. When the stay alarm unit 8 adds importance to the alarm, the alarm unit 9 instructs the alarm device 106 to output the sound of the speaker or the light blinking pattern according to the high or low importance. Or, when the degree of importance is low, the audio output of the speaker and the blinking pattern of the light are not output. In addition, when the staying alerting unit 8 adds importance to the alerting, the alarming unit 9 controls the alarming device 106 to control the operation according to high or low importance of the control device 105, or important. The operation of the escalator 101 is controlled only when the degree is high.

  When the staying alerting unit 8 adds importance to the alerting, the alarm unit 9 switches the contents of the character string 312 on the screen output 301 according to the significance or draws only when the significance is high. It's okay.

  In the first embodiment, the short-term change extraction unit 4 determines whether the person 113 who has extracted the connection region 133 is located inside the deck 110 depending on whether the representative side 136 of the connection region 133 is included in the detection area 120 or not. You may determine whether you are outside or outside. As an example, FIG. 12A shows a scene where a long shadow person 113p passes by the side of the deck 110 in the direction of the arrow 117p, and FIG. 12B shows a short-term change extraction in the scene of FIG. The representative points 137p at the centers of the representative sides 136p and 136p of the circumscribed rectangles 135p and 135p of the connected regions 133p and 133p extracted by the part 4 are shown, respectively. 12 (c) shows a scene where the person 113q passes by the front of the deck 110 in the direction of the arrow 117q. FIG. 12 (d) shows a connection extracted by the short-term change extraction unit 4 in the scene of FIG. 12 (c). The representative points 137q at the centers of the representative sides 136q and 136q of the circumscribed rectangles 135q and 135q of the regions 133q and 133q are shown, respectively. The representative side 136p and the representative side 136q both greatly protrude from the detection area 120 because the person 113p and the person 113q are outside the deck 110. On the other hand, when the person 113 is on the deck 110 as shown in FIG. 4C, the representative side 136 is on the deck 110. Therefore, the short-term change extraction unit 4 determines whether the person 113 from which the connected area 133 is extracted is inside or outside the deck 110 depending on whether the representative side 136 is within the detection area 120 or not. Is possible.

  In the first embodiment, when the representative points 137p and 137q are accumulated in the short-term change accumulating unit 5 by the trajectory extracting unit 6 when the representative points 137 extracted from the persons existing outside the deck are accumulated, the trajectory extracting unit 6 decrements the deck 110. An incorrect movement locus 153 is extracted from the outside person 113. Therefore, if the short-term change extraction unit 4 rejects the representative point 137 when the representative side 136 protrudes from the detection area 120, the representative point 137 of the person 113 outside the deck 110 is accumulated in the short-term change accumulation unit 5. Can be prevented. When the short-term change accumulating unit 5 accumulates the short-term change itself, the short-term change extracting unit 4 rejects the short-term change corresponding to the person existing outside the deck 110 (outside the detection area 120). Accumulation in the change accumulation unit 5 is prevented.

  In the first embodiment, if the camera 103 can capture the vicinity of the deck 110 in the field of view, the installation conditions for capturing from near the deck 110 as illustrated in 103b in FIG. 2 or the oblique front of the deck 110 as illustrated in 103c. The installation conditions for capturing the vicinity of the deck 110 may be applied.

  In addition, under the installation conditions that are captured from directly above the deck 110 as in the camera 103b, the side surface of the person 113 is hardly visible on the input image, so the length of the person 113 in the vertical direction is significantly different from the input image of the camera 103. . Therefore, when the monitoring device is applied at a plurality of points with different angles of view like the cameras 103, 103b, and 103c, the size of the person 113 on the input image, the moving speed, and the size of the detection area 120 are different for each point. Therefore, the point information storage unit 12 may store information for each angle of view in advance and switch the information according to the angle of view where the camera 103 is installed. Alternatively, the monitor 107 is provided with a user interface to receive information about the angle of view of the camera 103 from the monitor, and the information in the point information storage unit 12 is switched or fine-tuned according to the information about the angle of view received from the monitor. It may be.

  In the first embodiment, the movement of the movable part as in step 112 is learned in advance with the background model based on the spatio-temporal feature shown in Non-Patent Document 2, and the short-term change extraction unit 4 as shown in FIG. The connection region 132 in FIG. 3 and the connection region 122 by the long-term change extraction unit 3 as shown in FIG. 3C are determined as a steady change in step 112, and are discarded as unnecessary information and extracted as a difference. Can be suppressed. When the dynamic background model based on the spatiotemporal gradient is applied to the short-term change extraction unit 4, the movement trajectory 153 is extracted even when the person 113 flows backward in step 112, the escalator 112 goes down, and the person 113 starts from step 112. The movement trajectory 153 can be extracted when entering the deck 110.

  Further, the long-term change extraction unit 3 can extract a circumscribed rectangle 125 including a portion where the connection region 123 protrudes from the deck 110. If the circumscribed rectangle 125 including a portion protruding from the deck 110 can be extracted, the circumscribed rectangle 125 can cover the whole body of the person 113 even when the person 113 protrudes from the deck 110 as shown in FIG. Extraction can be made easy to see the drawing of the alarm unit 9.

  FIG. 13 shows a functional block diagram of Embodiment 2 of the present invention. In FIG. 13, the image acquisition unit 1, the trajectory extraction unit 6, the number of people estimation unit 7, the staying notification unit 8, the alarm unit 9, the control unit 10, the detection target information storage unit 11, and the point information storage unit 12 are the same as those in the first embodiment. Perform the same function. Moreover, the example of an apparatus structure at the time of applying to an escalator is shown in FIG.

  In FIG. 14, a sensor 203 is a sensor that can detect the presence or absence of a person 113 based on some measurement principle and simultaneously output information related to the location of the person 113 near the deck 110, and sense at least a part of the deck 110. A signal is output in a predetermined operation cycle.

  In FIG. 13, the sensing unit 23 receives a signal from the sensor 203 at a predetermined capture period and outputs the signal to at least one of the long-term change extraction unit 3 or the short-term change extraction unit 4. The measurement principle of the sensor 203 is to extract an object having a predetermined height from the ground surface as a person 113 from distance measurement information and installation information by a laser radar, a millimeter wave radar or the like, and at the same time, determine the location of the person 113 from the distance information. Examples include a method for identifying and a method for identifying a place by detecting the person 113 from the temperature corresponding to the body temperature with a directivity thermal sensor, but are not limited to this example. The sensor 203 includes pattern recognition for detecting the person 113 from one input image and specifying the position, such as face recognition. When the sensor 203 performs pattern recognition, an input image of the camera 103 may be used as an input for signal processing.

  In the functional configuration of FIG. 13, the sensing unit 23 outputs detection information of the person 113 detected by the sensor 203 to the long-term change extraction unit 3 as a long-term change or outputs a short-term change to the short-term change accumulation unit 5. Do at least one of them.

  The long-term change extraction unit 3 performs at least one of extraction of long-term changes by the method of the first embodiment and reception of long-term changes from the sensing unit 23. In the second embodiment, the long-term change extraction unit 3 provides the same background creation unit 2 as that of the first embodiment when extracting the long-term change by the method of the first embodiment.

  The short-term change accumulation unit 5 performs at least one of a short-term change by the method of the first embodiment and a short-term change reception from the sensing unit 23. In the second embodiment, when the short-term change extraction unit 4 extracts short-term changes by the method of the first embodiment, the same short-term change extraction unit 3 as that of the first embodiment is provided.

  As described above, in the second embodiment, in addition to the long-term change and short-term change of the person 113 extracted by the method of the first embodiment, at least the long-term change or short-term change of the person 113 from the signal of the sensor 203. One can be extracted. For example, when the camera 103 is a visible light camera, long-term changes and short-term changes of the person 113 can be stably extracted in the configuration of the first embodiment if external light is strongly reflected at a specific location on the deck 110. However, in the second embodiment, if a sensor such as a laser radar that is not easily affected by external light is used as the sensor 203, at least a part of the range on the deck 110 is changed over a long period or a short period. This improves the overall reliability of the monitoring device.

  Further, in the second embodiment, a high-precision sensor 203 can be used to capture the vicinity of a specific place where an accident such as a boundary between the deck 110 and the step 112 is likely to occur as in the detection area 190 of FIG. The long-term change of the detection area 190 can be reliably captured, and the overall reliability of the monitoring device can be improved. Further, if the sensor 203 that captures the detection area 190 is a distance measuring sensor, the person 113 may be tilted as shown in FIG. 3B or standing at the position shown in FIG. 7B. Thus, the behavior of the person 113 can be determined.

  FIG. 16 shows a device configuration of the monitoring device according to the third embodiment of the present invention. In FIG. 16, the door 381 provided in the passage 383, the authentication device 382 provided in the door 381, the monitoring area 310 provided in the vicinity of the door 381 and the authentication device 382, and the vicinity of the monitoring area 310 are captured in the field of view. A control device 105 and an alarm device 106, an image processing device 104, a monitor 107, a transmission device 108, and a recording device 109 provided beside the camera 103 and the door 381 are shown. FIG. 16 shows a person 113 traveling in the passage 383.

  In FIG. 16, the door 381 controls the passage of the person 113 by opening and closing, and the authentication device 382 uses a predetermined authentication means such as a card key or fingerprint authentication to open the door 381 to the person 113 permitted to pass by the authentication means. Allow to open. The control device 105 controls the opening and closing of the door 381, and the alarm device 106 includes a light or a speaker that can issue an alarm toward a person near the monitoring area 310. The devices 104, 107, 108, and 109 perform the same functions as the devices having the same numbers in the first embodiment shown in FIG. In the third embodiment, a sensor 203 having a detection area 390 at a specific portion in the monitoring area 310 may be provided.

  A person who stays in the monitoring area 310 for an extremely long time waits for a person with work or permission to pass through the authentication device 382 or the door 381 even though the door 381 is not permitted to pass. There is a possibility that a person with permission may attempt to pass the door 381 illegally at the moment when the door 381 is opened, and it should be detected as an abnormal behavior. Further, when there is only one person in the monitoring area 310, there is no surrounding person witnessing the passage of the unauthorized door 381, and the degree of danger is high.

  Therefore, in the third embodiment, when there is no sensor 203, the monitoring area 310 has the functional configuration of the first embodiment shown in FIG. 1, and when the sensor 203 has the functional configuration of the second embodiment shown in FIG. In addition to detecting staying above, the number of people on the monitoring area 310 is estimated by classification of 0, 1, 2 or more, and only when the number of people in the monitoring area 310 is staying alone is reported. Or by giving a high importance to staying when the number of people in the monitoring area 310 is one, and alerting by specifying that it is staying when one person, It can also be effectively monitored from the aspect of safety management and security.

DESCRIPTION OF SYMBOLS 1 ... Image acquisition part 2 ... Background creation part 3 ... Long-term change extraction part 4 ... Short-term change extraction part 5 ... Short-term change accumulation part 6 ... Trajectory extraction part 7 ... Number of persons Estimating unit 8 ... Stalling reporting unit 9 ... Alarm unit 10 ... Control unit 11 ... Detection target information storage unit 12 ... Point information storage unit 101 ... Escalators 103, 103a, 103b ..Camera 110 ... Pickup deck 113 ... People

Claims (13)

Image acquisition means for acquiring an input image near the monitoring area;
Processing the input image from the image acquisition means, and estimating the number of people to be detected in the monitoring area is 0, 1, 2, or more than one person, and a number estimation means for estimating whether or not
A monitoring device comprising: a staying alerting means for specifying staying when the estimated number of persons in the monitoring area is one and reporting to the outside. Image acquisition means for acquiring an input image near the monitoring area;
A stay detection means for detecting stay of a detection target in a monitoring area using an input image from the image acquisition means;
A movement trajectory extracting means for extracting a trajectory in which the detection target moves in the monitoring area;
Using the staying information from the staying detection means and the movement locus information from the movement locus extraction means, the number of people to be detected in the monitoring area is 0 person, 1 person, or 1 person out of 2 or more categories. A means for estimating the number of people other than a person,
A monitoring device comprising: a staying alerting means for specifying staying when the estimated number of persons in the monitoring area is one and reporting to the outside. Image acquisition means for acquiring an input image near the monitoring area;
A background creation unit that creates a background image, and a long-term change extraction unit that extracts a portion changing between the input image and the background image as a long-term change and extracts a staying place and timing from the long-term change A short-term change extraction unit that extracts a short-term change on the image from the input image at two or more times as a short-term change, and information extracted from the time series of the short-term change or the short-term change From a short-term change accumulation unit to accumulate, a stay detection means for detecting stay of a detection target in the monitoring area,
A movement trajectory extracting means for extracting a movement trajectory from the time series of the short-term change or the accumulation of time series of information extracted from the short-term change;
Estimating the number of people to be detected in the monitoring area from the time and place at which the stay and the movement trajectory occurred, whether the number of people to be detected is 0, 1, or 2 or more. Means,
A monitoring device comprising: a staying alerting means for specifying staying when the estimated number of persons in the monitoring area is one and reporting to the outside.   The monitoring device according to any one of claims 1 to 3, wherein alarm means for alarming by at least one of a screen output and a sound output or a device in the vicinity of the monitoring area is controlled in accordance with an output from the stay alarm means A monitoring device comprising at least one of means.   The monitoring apparatus according to claim 2 or 3, wherein the number of persons estimating means is set in the monitoring area when no movement trajectory is detected from the outside of the monitoring area to the staying place immediately before the staying detection is continued. A monitoring apparatus characterized in that the number of persons is estimated to be zero and the remaining number of persons is estimated to be one or more.   The monitoring device according to claim 2 or 3, wherein the person estimation means detects a movement trajectory from the outside of the monitoring area to the location of the stay immediately before the stay detection is continued, and in the monitor area. The monitoring is performed when one or more movement trajectories that have entered the monitoring area from outside the monitoring area are detected while the staying detection is continued, or immediately before the staying is continued in the monitoring area. When two or more movement trajectories that have entered the monitoring area from outside the area are detected, the number of detection targets in the monitoring area is estimated to be two or more, and the remaining number of people is 0 or 1 A monitoring device characterized by estimating a person. The monitoring apparatus according to claim 3, wherein the information extracted from the short-term change is a representative point of a representative side of a circumscribed rectangle of the connected region extracted from the short-term change.   The monitoring apparatus according to claim 3, wherein the short-term change extraction unit includes a detection target in which the short-term change is extracted from the position of the short-term change or the information extracted from the short-term change in the monitoring area. Whether or not the monitoring area is outside, and when outside the monitoring area, the corresponding short-term change or the information extracted from the corresponding short-term change is rejected. .   5. The monitoring apparatus according to claim 4, wherein the alarm means outputs a movement trajectory from the outside of the monitoring area to a staying location on a screen in response to an output from the staying alerting means.   The monitoring apparatus according to claim 2 or 3, wherein when the stagnation occurs, the number-of-persons estimation unit moves from the outside of the monitoring area to a stagnation place, or the position of the movement locus and the long-term change. Alternatively, a monitoring device comprising behavior determining means for determining the type of the behavior of the detection target staying from the position information of the circumscribed rectangle.   The monitoring apparatus according to any one of claims 1 to 3, further comprising a sensor that captures at least a part of the monitoring area separately from the image acquisition means, wherein the number of people estimation means includes the input image from the image acquisition means and the image acquisition means. A monitoring apparatus that estimates the number of detection targets in the monitoring area using area information from a sensor.   4. The monitoring apparatus according to claim 3, wherein a sensor that captures at least a part of the monitoring area is provided separately from the image acquisition unit, and the sensor extracts a long-term change or a short-term change related to a detection target in the monitoring area. Then, replace it with a long-term change by the background creation unit, replace it with a short-term change by the short-term change extraction unit, make up for a long-term change by the background creation unit, or replace a short-term change by the short-term change extraction unit A monitoring device characterized by supplementing or performing at least one of them.   The monitoring apparatus according to claim 1, wherein the escalator is monitored by setting a monitoring area of the image acquisition means in a deck portion of the escalator.

Images