JP2010176561A - Security system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a security system constructible by an inexpensive camera system without needing a zooming camera or the like. <P>SOLUTION: The security system has a model data storage section 120 storing an ID given to each of persons permitted to stay in a space for security assurance, and storing posture information model data related to the posture information of the person corresponding to the ID; an imaging device 110 imaging the space and acquiring imaging data; a posture information estimating means for extracting the motion of the person from the imaging data acquired by the imaging means, and estimating the posture information of the person; a probability computing means for computing the probability that the person is the person given the ID from the attitude information estimated by the attitude estimating means and the attitude information model data stored in the storage means; and a first determining means for determining whether the person is the person given the ID based on the probability computed by the probability computing means. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a security system that automatically performs monitoring in a space where security is desired.

Conventionally, various security systems for unattended monitoring of a space where security is desired have been proposed. For example, in Patent Document 1 (Japanese Patent Laid-Open No. 2005-295469), an intruder is accurately detected in a system that monitors a predetermined area, and accurate angle control and zooming control are performed even in an unattended monitoring state. A monitoring system capable of recording an accurate video of an intruder and issuing an alarm with a low false alarm rate is disclosed.
JP 2005-295469 A

  However, the security system described in Patent Document 1 requires a zoom-up camera and the like, and there is a problem that a large cost is required when constructing the security system.

  In order to solve the above problem, the invention according to claim 1 stores an ID given to each person permitted to stay in a space to be secured, and also stores the ID of the person corresponding to the ID. Storage means for storing posture information model data related to posture information, imaging means for imaging the space and acquiring imaging data, and extracting a person's movement from the imaging data acquired by the imaging means, and posture information of the person The probability of calculating the probability that the person is given an ID from the attitude information estimation means for estimating the position, the attitude information estimated by the attitude estimation means and the attitude information model data stored in the storage means A first determining unit that determines whether the person is an ID-assigned person based on the probability calculated by the probability calculating unit; A security system and having a.

  According to a second aspect of the present invention, in the security system according to the first aspect, when the first determination unit determines that the person is not a person who has been given an ID, the first determination unit It has the 2nd determination means which determines whether the said person is a person to whom ID was given by a method different from the determination method.

  According to the security system of the first aspect of the present invention, it is possible to construct a security system with an inexpensive camera system without requiring a zoom-up camera or the like.

  Further, according to the security system of the first aspect of the present invention, even when the person to be identified wears sunglasses or the like, the security system is not disabled.

In addition, according to the security system of the second aspect of the present invention, the second determination for determining whether or not the person is given a ID by a method different from the determination method by the first determination means. With the means, the first determination means can be backed up.

It is a figure explaining the concept of the security system which concerns on embodiment of this invention. It is a block diagram of the outline | summary of the security system which concerns on embodiment of this invention. It is a figure which shows the concept of the Bayesian network used with the security system which concerns on embodiment of this invention. It is a figure which shows the data acquired by the imaging device of the security system which concerns on embodiment of this invention. It is a figure which shows the structure of the operation | movement data recognized from the imaging data etc. of the security system which concerns on embodiment of this invention. It is a figure explaining the data memorized by the model data storage part of the security system concerning the embodiment of the present invention. It is a figure which shows the data structure of the attitude | position information model data in the security system which concerns on embodiment of this invention. It is a figure explaining the data memorized by the registration data storage part of the security system concerning the embodiment of the present invention. It is a figure which shows the network structure of the attitude | position information model used with the security system which concerns on embodiment of this invention. It is a figure which shows the state which physiological state Ph (5) takes in a Bayesian network structure. It is a figure which shows the flowchart of the process by the security system which concerns on embodiment of this invention. It is a figure which shows the flowchart of the 1st person determination process subroutine in the security system which concerns on embodiment of this invention. It is a figure which shows the flowchart of the attitude | position recognition process subroutine in the security system which concerns on embodiment of this invention. It is a figure which shows the flowchart of the probability _Pid calculation process subroutine in the security system which concerns on embodiment of this invention. It is a figure which shows the flowchart of the 2nd person determination processing subroutine in the security system which concerns on embodiment of this invention. It is a figure which shows the flowchart of the learning process subroutine in the security system which concerns on embodiment of this invention. It is a figure which shows the flowchart of the subroutine of the learning process of the personal model data in the security system which concerns on embodiment of this invention.

Hereinafter, preferred embodiments of the security system of the present invention will be described with reference to the drawings. FIG. 1 is a diagram for explaining the concept of a security system according to an embodiment of the present invention, and FIG. 2 is a block diagram of an outline of the security system according to an embodiment of the present invention. In FIG. 1, R is a space for which security is secured in the present system, and A, B, C, and D represent persons staying in such a space R. In FIGS. 1 and 2, 100 is a main control unit, 110 is an imaging device, 120 is a model data storage unit, 130 is a registration data storage unit, 140 is a communication unit, and 200 is a security management center.

The security system according to the present embodiment operates so as to identify each person staying in the space R in order to ensure the security of the space R in FIG. Here, each person who is allowed to stay in the space R is previously registered with a unique ID number. Furthermore, the system determines the person by converting the habit of movement (action) for each person corresponding to this ID into data. In the present embodiment, the determination that the system determines the person based on the habit of how to move for each person is the first person determination. Further, in this embodiment, data that records the habit of movement is referred to as posture information model data related to posture information, and this data is created based on a model based on a Bayesian network as described later. The security system according to the present embodiment has second person determination means for performing person determination without being based on the posture information model data. In the present embodiment, the second person determination unit will be described based on an example in which face image data registered in advance for each person and imaged captured data are compared by image processing. However, in the security system of the present invention, other methods can be used as the second person determination means.

  Here, an outline of a Bayesian network used for person determination based on posture information model data in the security system of the present invention will be described. FIG. 3 is a diagram showing the concept of a Bayesian network used in the security system according to the embodiment of the present invention. A Bayesian network is a model that represents a probabilistic causal relationship in a graph structure. Each node X1, X2, X3, and X4 shown in FIG. 3 represents a random variable, the presence or absence of a causal relationship between them is a directed graph, and the causal relationship between nodes of the values that the random variable can take is a conditional probability. Represented by The fact that there is a stochastic relationship between the node X1 and the node X2 is indicated by an arrow from the node X1 to the node X2. The node X2 is affected by the nodes X1 and X3, and the node X4 is affected by the node X3. The probability table taking into account such a causal relationship is a conditional probability table shown in FIG. It is. On the other hand, the nodes X1 and X3 are parent nodes that are not affected by any node, and the probability table at this time is like the prior probability table shown in the figure. In the case shown in FIG. 3, from the structure of the graph, the joint probabilities of all random variables are P (X1, X2, X3, X4) = P (X1) P (X2 | X1, X3) P (X3) P (X4 | X3). Inference using a Bayesian network is performed by obtaining peripheral posterior probabilities by marginalizing this connection probability. As for the probability calculation method based on such a Bayesian network, the inventors have application number 2008-90755 (invention name “security system”) and application number 2008-90777 (invention name “security system and driving support”). Since the method is described in detail in “Method”), the contents of these applications are also incorporated herein.

  Now, the security system according to the present embodiment operates to identify each person staying in the space R in order to ensure the security of the space R. For this purpose, the security system shoots a moving image of the space R. An imaging device 110 is provided. Although FIG. 1 shows an example in which four imaging devices 110 are provided, the number of such imaging devices 110 to be installed is not limited to this. Therefore, in the security system of the present invention, n imaging data from the n imaging devices 110 can be acquired using two or more n imaging devices 110. FIG. 4 is a diagram showing data acquired by the imaging apparatus of the security system according to the embodiment of the present invention.

  From the imaging data acquired by the n imaging devices 110 as shown in FIG. 4, information (posture information including position information and the like) related to the person's movement is extracted and estimated, and the movement method for each person is determined. Created posture information model data related to posture information, such as habit, or a person recognized by imaging data based on the extracted / estimated information and previously registered posture information model data Or to identify.

FIG. 5 is a diagram showing a structure of operation data recognized from image data and the like of the security system according to the embodiment of the present invention. As shown in FIG.
It can be broadly classified into “position information”, “posture information”, and “unseen information”. Among these data, “position information” and “posture information” are estimated from the imaging data acquired by the imaging device 110. “Position information” is information relating to the location of the target person, and is obtained from the imaging data. Also, under “Posture Information”, “Right Hand Position”, “Left Hand Position”, “Right Elbow Position”, “Left Elbow Position”, “Face Direction”, “Right Foot Position”, “Left Foot Position” There are data such as “position”, “right knee position”, “left knee position”, “waist position”, “body orientation”, etc., which are estimated from the imaging data. In addition, “psychological state” and “behavioral variable” are information that cannot be recognized from imaging data and the like in “non-apparent information”, and in this embodiment “psychological state” is “calm”, “tension”, The distributed amount data takes five states of “fatigue”, “excitement”, and “others”, and the “behavior variable” is continuous amount data obtained by quantifying the behavior pattern of the person.

  The main control unit 100 is a general-purpose information processing mechanism that includes a CPU, a ROM that holds programs that run on the CPU, and a RAM that is a work area of the CPU. The main control unit 100 cooperates and operates with each component connected to the main control unit 100 shown in the block diagram. The main control unit 100 executes various control processes in the security system of the present invention based on programs and data stored and held in storage means such as a ROM in the main control unit 100.

  A model data storage unit 120 is connected to the main control unit 100 as described above, and the main control unit 100 refers to posture information model data of individual persons registered and recorded in the model data storage unit 120. Or newly registering posture information model data or updating already registered posture information model data.

  Next, data stored in the model data storage unit 120 will be described. FIG. 6 is a diagram for explaining data stored in the model data storage unit of the security system according to the embodiment of the present invention. FIG. 7 shows the data structure of the posture information model data in the security system according to the embodiment of the present invention. FIG. As shown in FIG. 6, the model data storage unit 120 has a unique registration ID (id = 1, 2, 3,...) Assigned to each person who is allowed to stay in the space R that is to be secured. .., M) and the posture information model data of each person corresponding to the registered ID. FIG. 7 is a diagram for explaining the data structure of each posture information model data in more detail. As shown in FIG. 7, the data in the posture information model data can be classified into two types of data: outline, discrete data, and continuous data. “Psychological state” data belongs to the former, and this data is held in the form of a conditional probability table. The latter includes “current position”, “right hand position”, “left hand position”, “right elbow position”, “left elbow position”, “head position”, “face orientation”, “right foot position” ”,“ Left foot position ”,“ Right knee position ”,“ Left knee position ”,“ Waist position ”,“ Body orientation ”,“ Behavior variables ”data, and these data are Gaussian distribution Are held in the form of average, variance, and weight.

Further, a registration data storage unit 130 is connected to the main control unit 100, and the main control unit 100 refers to face image data for matching individual persons registered and recorded in the registration data storage unit 130. Or newly registering face image data for matching, or updating already registered face image data. FIG. 8 is a diagram for explaining data stored in the registration data storage unit of the security system according to the embodiment of the present invention. As shown in FIG. 8, the registration data storage unit 130 has a unique registration ID (id = 1, 2, 3,...) Assigned to each person who is allowed to stay in the space R to be secured. .., M) and face image data obtained by photographing each person's face corresponding to the registered ID. Such face image data stored in the registered data storage unit 130 is used when executing a second person determination unit that is a person determination not based on the posture information model data.

  The communication unit 140 is connected to the main control unit 100. For example, when it is determined that a person who is not permitted to stay in the space R is staying, the security management center is connected via the communication unit 140. 200 is notified / reported. It is not always necessary to provide such a communication unit 140 for reporting, for example, an alert display unit is simply provided, and a person who is not permitted to stay in the space R is staying. When it is determined to be a thing, only the alert display may be performed by the display unit.

  Next, a specific configuration of the posture information model used in the security system of the present invention will be described. FIG. 9 is a diagram showing a network structure of the posture information model used in the security system according to the embodiment of the present invention. The posture information model used in the security system of the present invention includes (a) the structure of the Bayesian network shown in FIG. 9, (b) a conditional probability table (psychological state data in FIG. 7), a prior probability table, and (c) a Gaussian distribution. Parameter (data of each item belonging to continuous data in FIG. 7). Of these three, the structure of the Bayesian network (a) is commonly used for all registered ID persons. On the other hand, (b) and (c) are stored as model data for each registered ID by learning or updated as appropriate.

  In the structure of the Bayesian network of FIG. 9, □ indicates a discrete node, ◯ indicates a continuous node, and the numbers in the discrete nodes indicate the number of states that each node takes. For example, the “physiological state Ph (5)” includes five values of Ph = 0 (normal), Ph = 1 (tension), Ph = 2 (fatigue), Ph = 3 (excitement), and Ph = 4 (others). It takes a state.

  In the network structure of FIG. 9, the current position (PO), right hand position (RH),... Face orientation (FA) nodes shown in the upper row and the right foot position (shown in the lower row) RF), right foot position (LF),... Body orientation (BD) are parent nodes, and these nodes affect the nodes of physiological state Ph (2) and the action variable DB. I think. In the network structure, it is considered that the node of the physiological state Ph (2) affects the node of the behavior variable DB. That is, physiological state data is a node affected by posture information data such as a person's position information data and right hand position, and a behavior variable is a node affected by position information data, posture information data, and physiological state data. The causal relationship of this network structure is determined by the Bayesian network.

  Here, the current position (PO), the right hand position (RH),... Face orientation (FA) nodes shown in the upper part of the network structure, and the right foot position (RF) shown in the lower part. , Left foot position (LF),... Body direction (BD) nodes correspond to posture information model data stored in the security system as shown in FIG. These continuous data are held in the form of “discrete”, “average”, and “variance” of Gaussian distribution, and these are replaced with predetermined probabilities.

The physiological state Ph (5) node in FIG. 9 serves as a kind of tolerance, and it has been found that if this node does not exist, the security system of the present invention may not operate effectively. This is because the behavior variable DB is unique even if the state (driving state) of each node of the current position (PO), right hand position (RH), ... body direction (BD) is uniquely determined. This is due to the fact that it is not determined. In the security system of the present invention, the introduction of the physiological state Ph (5) node makes it possible to handle causality including uncertain elements. The physiological state Ph (5) node is very important when the security system of the present invention is executed by a computer program.

  FIG. 10 is a diagram showing a state taken by the physiological state Ph (5) in the Bayesian network structure. As shown in FIG. 10, in this embodiment, the physiological state Ph (5) is Ph = 0 (normal), Ph = 1 (tension), Ph = 2 (fatigue), Ph = 3 (excitement), Ph. In the security system of the present invention, the physiological state Ph can be set to an appropriate natural number of 5 or more.

  Next, an outline of a flow of monitoring processing and learning processing by the security system of the present invention will be described. FIG. 11 is a diagram showing a flowchart of processing by the security system according to the embodiment of the present invention. Note that such a flowchart operates when the entry of a person into the space R that is to be secured is confirmed.

  In FIG. 11, when the process by the security system of the present invention is started in step S100, the process proceeds to step S101, and a first person determination process subroutine is executed. This subroutine of the first person determination process is to perform a person determination process based on the posture information model data.

  Here, a subroutine of the first person determination process will be described. FIG. 12 is a flowchart of the first person determination processing subroutine in the security system according to the embodiment of the present invention. When the first person determination process subroutine is started in step S200, the process proceeds to step S201, where the posture recognition process subroutine is executed. This subroutine is for estimating the posture of a person who has been imaged.

  Here, a subroutine for the posture recognition process will be described. FIG. 13 is a flowchart of a posture recognition processing subroutine in the security system according to the embodiment of the present invention. In FIG. 13, when the posture recognition processing subroutine is started in step S300, the process proceeds to step S301, and image data (background image data) when there is no motion prepared in advance is read. Such background image data is acquired in advance at the moment when the image does not change for a predetermined time. In step S302, the current imaging data is read by the imaging device 110. In the next step S303, a difference between the imaging data and the background image data is created. In step S304, the position (current position) of the non-background object (that is, a person) in the space is determined from the difference information from the plurality of imaging devices 110. Estimation is performed. In step S305, posture estimation is performed, and posture information (right hand position, left hand position, right elbow position,...) Is acquired. Here, as a technique for posture estimation, for example, a technique disclosed in Tomomoto Murakami et al. “Estimation of a person posture including self-occlusion by a distance image” Information Processing Society of Japan 65th National Convention can be used. In step S306, the process returns to the original routine.

Returning to FIG. 12, steps S202 to S205 are loops for calculating the probability for each id of the registration ID. From this loop, when all the ids have been calculated, we can exit and proceed to the next step. In this loop, in step S203, the posture information model data corresponding to the corresponding id is read from the model data storage unit 120. In step S204, a subroutine for calculating the probability P _id from the posture information model data is executed. The When the loop is exited, the maximum _one of P _{1 to} P _M in step S206 is set as P _max . In step S207, it is determined whether or not P _max is equal to or greater than a predetermined value. If the determination result is YES, the process proceeds to step S208, the registration ID corresponding to P _max is set as a return value, and the determination result is NO. If there is, the process proceeds to step S209, and the corresponding value is set as not applicable. In step S210, the process returns to the main routine.

Here, a subroutine for calculating the probability P _id from the posture information model data will be described. FIG. 14 is a flowchart of a probability _Pid calculation processing subroutine in the security system according to the embodiment of the present invention. In FIG. 14, when the subroutine of the probability P _id calculation process is started in step S400, the process proceeds to step S401, where the posture information model data with ID = id is acquired. In step S402, the coincidence probability calculation is performed from this model data. P _id is calculated by the following equation (1).

なお、ＲＨはノードを示し、ｒｈはノードの値を示す。また、Ｐ（ＩＤ＝ｉｄ｜ＲＨ＝ｒｈ）とは、ノードＲＨがｒｈであるときの事後確率といい、ノードＲＨのパラメーターがｒｈであるときに、ＩＤがｉｄである確率を示すものである。また、Ｐ_DBは式（２）によって求められるものである。

Note that RH indicates a node, and rh indicates a value of the node. P (ID = id | RH = rh) is a posterior probability when the node RH is rh, and indicates a probability that the ID is id when the parameter of the node RH is rh. . Further, P _DB is obtained by the equation (2).

また、ノードＰＯ、ＲＨ、・・・ＢＤ、ＤＢは連続ノードであり、これらに係る量は正規分布関数Ｎの平均μ、分散Σ、重みから式（３）のように求めることができる。

Nodes PO, RH,..., BD, DB are continuous nodes, and the quantities relating to these can be obtained from the mean μ, variance Σ, and weight of the normal distribution function N as shown in Expression (3).

ここで、Ｐ_a（ＰＯ）はＰＯの親ノードを示す。ステップＳ４０２に続く、ステップＳ４
０３では、元のルーチンにリターンする。

Here, P _a (PO) indicates a parent node of PO. Following step S402, step S4
In 03, the process returns to the original routine.

For more information on how to derive a predetermined probability using a Bayesian network,
○ Jensen, F.M. V. "Bayesian networks and decision graphics", Springer, 2001.
O David C.I. MacKay, “Information Theory, Information and Learning Algorithms”, Cambridge University Press, 2003.
It shall be incorporated with reference to the description.

  Returning to the main flowchart shown in FIG. 11, in step S102, it is determined whether there is a registration ID as a return value. If the decision result in the step S102 is YES, it is found that the person having the registration ID exists in the space R, so the process proceeds to the step S109 and the process is ended. When the determination result in step S102 is NO, the process proceeds to step S103, and in step S103, a subroutine for second person determination processing is executed. In this subroutine, it is determined whether or not the person is given a ID by a method different from the determination method by the first determination process. Such a second determination process serves as a backup for the first determination process. Moreover, if it is determined in the second determination process that the target person is a person having a registration ID, it may be an opportunity to learn posture information model data.

  As described above, according to the processing such as step S101 and step S102 related to the first determination processing, it is sufficient that the position information and posture information of the target person can be acquired. Since expensive cameras and high-resolution cameras are not required, a security system can be constructed at a low cost. In addition, as described above, the first determination process performs the person determination based on the posture information model data obtained by converting the character of the person's motion into data. For example, when the person to be identified wears sunglasses or the like. However, the security system will not be disabled.

  Here, a subroutine of the second person determination process will be described. FIG. 15 is a view showing a flowchart of a second person determination processing subroutine in the security system according to the embodiment of the present invention. In FIG. 15, when the second person determination processing subroutine is started in step S500, in step S501, imaging data of the face of the target person is acquired by the imaging apparatus. At this time, it is preferable to use imaging data acquired with high accuracy among a plurality of imaging devices. Steps S502 to S505 are a loop for making a determination for each registered ID, and exit from the loop upon completion of checking all registered IDs from id = 1 to id = M. In step S503 in the loop, face image data is read from the registered data storage unit 130, and in step S504, matching processing between the face image data and the imaging data is performed. In step S504, a face authentication processing technique such as general template matching can be used. If the check loop is exited in step S505, the process proceeds to step S506. If there is a match, the registration ID is set as a return value, and if no match is found, no registration ID is set as the return value. In step S507, the process returns to the original routine.

  Returning to the main flowchart shown in FIG. 11, in step S104, it is determined whether there is a registration ID as a return value. If the determination result in step S104 is NO, it means that the person could not be specified by the two types of person determination processing, and therefore the process proceeds to step S107, and the security management center 200 is notified via the communication unit 140. Take measures that you want to set in advance, such as The person extracted in this step is a person such as C in FIG.

  If the decision result in the step S104 is YES, it has been found that the person having the registration ID exists in the space R. Therefore, no action is taken to notify the person decision process based on the posture information model data (first The process proceeds to step S105 in order to identify the cause of the failure of the determination process 1).

  In step S105, it is determined whether or not the data of the corresponding registration ID exists in the model data storage unit 120. If the determination in step S105 is NO, the system side has not yet learned the posture information model data corresponding to the ID (step S108), so a model data learning process is performed in step S106. The person extracted in this step is a person such as D in FIG.

  On the other hand, if the determination in step S105 is YES, it is possible that the action of the person with the ID has changed due to, for example, injury, etc., the process proceeds to step S106 to perform model data learning processing, and the model data storage unit 120 Update the data. In step S106, a learning process subroutine is executed, and in step S109, the process ends.

  Next, the learning processing subroutine will be described. FIG. 16 is a flowchart of a learning processing subroutine in the security system according to the embodiment of the present invention. When the learning processing subroutine is started in step S600, the posture recognition processing subroutine is executed in step S601 to acquire position information, posture information, and the like. In a subsequent step S602, a personal model data learning process subroutine is executed to calculate posture information model data from the position information and posture information, and the process returns in step S603.

Next, a subroutine for learning process of personal model data will be described. FIG. 17 is a flowchart of a subroutine for learning process of personal model data in the security system according to the embodiment of the present invention. In FIG. 17, when the personal model data learning processing subroutine starts in step S700, the process proceeds to step S701. If it exists, the model data of the person corresponding to the ID is taken in. In step S702, the position information and orientation are acquired. The model data is obtained by calculating the mixed normal distribution probability from the information. For the creation of this model data, Arthur Demster, Nan Laird, and Donald Rubin. “Maximum likelihood from completed d
ata via the EM algorithm ". Journal of the Royal Statistical Society, Series B, 39 (1): 1-38, 1977.

  In step S703, new attitude information model data is stored in the model data storage unit 120, and in step S704, the process returns to the original routine.

  According to the security system configured as described above, it is possible to construct a security system with an inexpensive camera system without requiring a zoom-up camera or the like.

  Moreover, according to the security system having the above-described configuration, even when a person to be identified wears sunglasses, the security system is not disabled.

  In addition, according to the security system having the above-described configuration, the second determination unit that determines whether or not the person is given an ID by a method different from the determination method by the first determination unit is provided. The first determination means can be backed up.

DESCRIPTION OF SYMBOLS 100 ... Main control part 110 ... Imaging device 120 ... Model data storage part 130 ... Registration data storage part 140 ... Communication part 200 ... Security management center

Claims (2)

A storage unit that stores an ID assigned to each person permitted to stay in a space to be secured, and stores posture information model data related to posture information of the person according to the ID;
Imaging means for imaging the space and acquiring imaging data;
Posture information estimation means for extracting the movement of a person from the imaging data acquired by the imaging means and estimating posture information of the person;
A probability calculating means for calculating the probability that the person is given an ID from the attitude information estimated by the attitude estimating means and the attitude information model data stored in the storage means;
A security system, comprising: first determination means for determining whether or not the person is given a ID based on the probability calculated by the probability calculation means. When the first determination means determines that the person is not an ID-assigned person, whether or not the person is an ID-assigned person by a method different from the determination method by the first determination means The security system according to claim 1, further comprising second determination means for determining whether or not.

Images