JP5907162B2 - Mobile object attribute estimation system, mobile object attribute estimation method, and program - Google Patents

Description

  The present invention relates to a moving object attribute estimation system, a moving object attribute estimation method, and a moving object attribute estimation program that estimate an attribute of a moving object.

  In recent years, an ID (Identification) is attached to a moving body such as a person or an object to manage the moving body. For example, in offices and factories, IC cards (Integrated Circuit Cards) in which employee numbers are recorded as IDs are distributed to employees. The IC card is used to authenticate the employee at the security gate, thereby preventing the suspicious person from entering. In this example, the mobile object is managed by the ID for ensuring security.

  In addition to the ID assigned to the mobile object, it is conceivable to detect the attribute of the mobile object and manage the mobile object in association with the attribute and the ID. As an example of the attribute of the moving object, for example, an appearance feature of the moving object can be cited. As described above, by managing the mobile object by associating the attribute with the ID, for example, when a third party obtains an IC card illegally and tries to pass through the security gate, it is detected from the third party. By comparing the attribute and the attribute of the true owner of the IC card, unauthorized use of the IC card by a third party can be detected.

  Thus, for example, Patent Document 1 discloses a method of associating the ID and attribute of a moving object. In the method described in Patent Literature 1, an ID of an RF tag possessed by a user is acquired in a predetermined area. Further, the situation of the predetermined area is photographed, the shadow of the moving object is extracted by image processing from the image obtained by photographing, and a unique ID is given to the shadow. Then, the time when the ID is acquired from the RF tag is compared with the time when the shadow is obtained by the image processing, and when the time is close, the ID obtained from the RF is associated with the ID assigned to the shadow. That is, in the method described in Patent Document 1, the shadow of the moving object is used as an attribute, and the ID of the moving object obtained at the time when the moving object enters the predetermined area corresponds to the shadow ID of the moving object. Attached.

Japanese Patent Laying-Open No. 2004-350166 (paragraphs 0014-0033)

  When managing a mobile object for the purpose of ensuring security, security may not be sufficiently ensured only with an ID assigned to the mobile object. For example, if a third party illegally obtains an employee's ID card, the third party can impersonate the employee and pass through the security gate.

  As described above, in addition to the ID assigned to the mobile object, the security effect can be enhanced by detecting the attribute of the mobile object and managing the mobile object in association with the attribute and the ID. A method for associating the ID and attribute of a moving object is described in Patent Document 1.

  However, in the method described in Patent Document 1, the ID of the moving object obtained when the moving object enters the predetermined area is associated with the shadow ID of the moving object. Therefore, the opportunity of associating the ID of the moving object with the ID of the attribute (shadow of the moving object) is only once when the moving object enters the predetermined area.

  Further, in the method described in Patent Document 1, when a large number of moving objects enter at the same time, the IDs of the large number of moving objects and the IDs of the shadows assigned to the shadows of the large number of moving objects are obtained. For this reason, there are a large number of shadow ID candidates to be associated with one mobile object ID, and it is difficult to accurately associate the mobile object ID and the shadow ID one-on-one.

  Therefore, an object of the present invention is to provide a mobile object attribute estimation system, a mobile object attribute estimation method, and a mobile object attribute estimation program that can appropriately estimate an attribute corresponding to a detected mobile object.

  The mobile object attribute estimation system according to the present invention includes an ID of a mobile object, state information indicating the state of the mobile object, ID information including the ID and the detection time of the state information, an attribute of the mobile object, and the mobile object. Attribute information including the state information representing the state of the body and the attribute and the detection time of the state information are input, ID information is sequentially selected, and a predetermined time based on the detection time included in the selected ID information Attribute information including detection time belonging to the band, and the attribute information determined to be similar to the status information included in the selected ID information is selected as attribute information related to the selected ID information. For each attribute value included in the attribute information selection means and each attribute information determined to be related to the ID information, a likelihood derivation for obtaining a likelihood corresponding to the attribute of the mobile object identified by the ID included in the ID information. And an attribute selection unit that selects a value of an attribute having the maximum likelihood for one ID and associates the ID with the value of the selected attribute for each ID. To do.

  The mobile object attribute estimation method according to the present invention includes a mobile object ID, state information indicating the state of the mobile object, ID information including the ID and the detection time of the state information, an attribute of the mobile object, Attribute information including the state information representing the state of the mobile object and the attribute and the detection time of the state information is input, ID information is sequentially selected, and a predetermined time is based on the detection time included in the selected ID information. Attribute information including detection times belonging to the time zone, and the attribute information determined to be similar to the state information included in the selected ID information is attribute information related to the selected ID information. For each attribute value selected and determined to be related to the ID information, a likelihood corresponding to the attribute of the mobile object identified by the ID included in the ID information is obtained. Degree Select a value for the attribute that is a large, a process of associating the value of the selected attribute and ID, and characterized by performing for each ID.

  The mobile object attribute estimation program according to the present invention includes a mobile object ID, state information indicating the state of the mobile object, ID information including the ID and the detection time of the state information, an attribute of the mobile object, A moving object attribute estimation program mounted on a computer to which state information representing the state of the moving object and attribute information including the attribute and the detection time of the state information are input. Attribute information including detection time belonging to a predetermined time zone with reference to the detection time included in the selected ID information, and the state information is similar to the state information included in the selected ID information Attribute information selection processing for selecting attribute information determined as attribute information related to the selected ID information, for each attribute value included in each attribute information determined to be related to ID information, A likelihood derivation process for obtaining a likelihood corresponding to the attribute of the moving object identified by the ID included in the D information, and a value of the attribute having the maximum likelihood for one ID, and An attribute selection process is performed in which the process of associating the selected attribute value with each ID is executed.

  According to the present invention, it is possible to appropriately estimate an attribute corresponding to a detected moving object.

It is a block diagram which shows the example of the moving body attribute estimation system of the 1st Embodiment of this invention. It is a block diagram which shows the structural example of the mobile body attribute estimation part 3. FIG. It is a flowchart which shows the example of process progress of the moving body attribute estimation system of 1st Embodiment. 6 is a schematic diagram showing detection positions of a moving object ID “i1” and attributes z1, z2, z3, and z4 in a detection region 50. FIG. It is explanatory drawing which shows the detection time of ID "i1" shown in FIG. 4, and attributes z1, z2, z3, and z4. It is explanatory drawing which shows the example of the calculation result of the Euclidean distance of the position coordinate contained in ID information I1, and each position coordinate contained in attribute information Z1, Z2, Z3. It is a schematic diagram which shows the detection position of the attribute contained in the relevant information selected as attribute information relevant to ID information I1. It is explanatory drawing which shows the value of the attribute contained in the attribute information selected as a thing relevant to ID information containing ID "i1" detected at each time shown in FIG. It is explanatory drawing which shows the example of the score regarding each attribute. It is explanatory drawing which shows the score after updating with respect to the score shown in FIG. It is a block diagram which shows the example of the moving body attribute estimation system of the 2nd Embodiment of this invention. It is explanatory drawing which shows the example of a change of the detection frequency by the conversion model in case an attribute classification is age. It is a flowchart which shows the example of a process progress of the mobile body attribute estimation part 3a in 2nd Embodiment. It is a block diagram which shows the example of the moving body attribute estimation system of the 3rd Embodiment of this invention. It is a flowchart which shows the example of the process progress of the mobile body attribute estimation part 3b in 3rd Embodiment. It is a block diagram which shows the example of the minimum structure of the moving body attribute estimation system of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Embodiment 1. FIG.
FIG. 1 is a block diagram illustrating an example of a moving object attribute estimation system according to the first embodiment of this invention. The moving object attribute estimation system of the present invention includes an ID information input unit 1, an attribute information input unit 2, a moving object attribute estimation unit 3, and an estimation result output unit 4. In the following description, an area for detecting the ID and attribute of a moving object is referred to as a detection area 50.

  The ID information input unit 1 detects the ID information of each moving body P that moves in the detection area 50 and inputs it to the moving body attribute estimation unit 3. Further, the attribute information input unit 2 detects the attribute information of each moving body P and inputs it to the moving body attribute estimation unit 3. The contents of the ID information and attribute information will be described later. The moving body attribute estimation unit 3 uses the ID information input from the ID information input unit 1 and the attribute information input from the attribute information input unit 2 to calculate the likelihood that each ID information corresponds to each attribute information. By estimating the maximum likelihood, the attribute of each moving object is estimated.

  The kind of each moving body P is not specifically limited, A human, an animal, or a thing may be sufficient.

  The ID information input unit 1 is a device that detects the ID information of each mobile object P and inputs it to the mobile object attribute estimation unit 3. The ID information is information including at least a set of the ID of the moving object, the state information indicating the state of the moving object P, and the detection time of the ID of the moving object. Therefore, the “ID information” and the “ID” itself of the moving object are distinguished from each other, and the “ID” itself of the moving object is included in the ID information. The state information is information indicating a state of the moving body P that can be detected from the moving body P, and a state that changes with the passage of time. Examples of such state information of the moving body P include information indicating the position, moving speed, moving direction, and the like of the moving body P. In addition, when the moving body is a person and the movement state of the person is adopted as the state of the moving body, for example, a state in which the movement state is labeled such as “stop”, “walking”, “running”, etc. It may be used as information. The state information may represent information related to one type of state (for example, only the position) or may represent information related to a plurality of types of states (for example, position and action).

  When the ID information input unit 1 tries to detect the ID of the moving object P and the ID cannot be detected, the ID and the state of the moving object are set to “none” in the ID information, and the time is included in the ID information. The ID information may be input to the moving object attribute estimation unit 3. Even if the ID information input unit 1 does not input ID information to the moving object attribute estimation unit 3, the moving object attribute estimation unit 3 determines that no ID of the moving object P is detected at that time. Good.

  The ID information input unit 1 may be any device that can detect the ID unique to the moving object P and the state of the moving object P and specify the detection time. For example, when the position of the moving body is adopted as the state of the moving body P, the moving body P has an active RFID (Radio Frequency IDentification) tag, and the identification information of the tag is used as the ID of the moving body P. An active RFID reader may be used as 1. Alternatively, the moving body P may own an IC card, the identification information of the IC card may be used as the ID of the moving body P, and the IC card reader may be used as the ID information input unit 1. Further, the mobile unit P owns a wireless local area network (LAN) terminal, uses the MAC address (Media Access Control address) of the wireless LAN terminal as the ID of the mobile unit P, and uses the access point as the ID information input unit 1. Also good. Further, when a unique barcode is printed for each moving body P, a barcode reader may be used as the ID information input unit 1. When the moving body P is a person, biological information such as a person's face, retina, fingerprint, and vein is used as the ID of the moving body P, and these biological information reading devices are used as the ID information input unit 1. May be. Moreover, you may use together ID information input part 1 from which a detection target differs like a face authentication apparatus and a RFID reader.

  For example, when the state of the moving body P is expressed by operations such as stop, walking, and running, a mobile terminal equipped with an acceleration sensor owned by the moving body P is used as the ID information input unit 1 and the ID of the moving body P is used. Alternatively, an identification number unique to the mobile terminal may be used.

  Although one ID information input unit 1 is illustrated in FIG. 1, a plurality of ID information input units 1 may be provided.

  The attribute information input unit 2 is a device that detects the attribute information of each moving object P and inputs it to the moving object attribute estimation unit 3. The attribute information is information including a set of at least an attribute detected from the moving object P, state information indicating the state of the moving object P, and a detection time of the attribute of the moving object. Therefore, the “attribute information” and the “attribute” of the moving object are distinguished, and the “attribute” of the moving object is included in the attribute information. Further, since the state information indicating the state of the moving body P has already been described, the description thereof will be omitted. The attribute information may include an attribute type representing the type of the detected attribute. However, if there is only one type of attribute information to be detected, the attribute type may not be included in the attribute information.

  The attribute of the moving object P means information that can be detected from the moving object P and that hardly changes over time. Examples of the attributes of the moving object P include, for example, the color, shape, size, weight, and the like of the moving object P. Further, when the moving body P is a person, the age, sex, preference, etc. of the person may be used as attributes. The attribute type is information for identifying the type of attribute of the moving object such as color, shape, size, weight, and the like.

  Although one attribute information input unit 2 is illustrated in FIG. 1, a plurality of attribute information input units 2 may be provided.

  When a plurality of attribute information input units 2 are provided, a plurality of attribute information input units 2 having different attribute types to be detected may be provided. Alternatively, a plurality of attribute information input units 2 for detecting a common attribute type may be provided. When providing a plurality of attribute information input units 2 for detecting a common attribute type, the attribute type may not be included in the attribute information. If the attribute type is different for each attribute information input unit 2, the attribute type is included in the attribute information.

  If the attribute information input unit 2 tries to detect the attribute of the moving object P and the attribute cannot be detected, the attribute information input unit 2 does not need to input the attribute information to the moving object attribute estimation unit 3. Then, the moving object attribute estimation unit 3 may determine that no attribute of the moving object P is detected at that time because no attribute information is input.

  The attribute information input unit 2 may be any device that can detect the attribute of the moving object P and the state of the moving object P and specify the detection time. For example, when position coordinates are used as the state of the moving object P, the attribute information input unit 2 may be realized by a moving object tracking system using a camera, a floor pressure sensor, and a laser range finder. In this case, the color, shape, size, etc. of the moving object P are detected from the camera as attributes, the weight of the moving object P is detected from the floor pressure sensor, and the shape of the moving object P is detected from the laser range finder. The size may be detected as an attribute. When the attribute information input unit 2 is realized in such a manner, the mobile object P does not need to hold a device necessary for detecting its own attribute. In addition, the attribute information input unit 2 may detect the attribute of the moving body in a manner held by the moving body. For example, the attribute information input unit 2 may be realized by a portable terminal capable of wireless communication that includes a storage device that records the attributes of the moving object P and a positioning sensor such as GPS (Global Positioning System).

  Note that the attribute information input unit 2 does not need to detect an ID unique to the moving object. Even when the attribute information input unit 2 is a mobile terminal or the like held by the mobile object, the ID of the mobile object need not be detected.

  Further, the state of the moving body P detected by the attribute information input unit 2 and the state of the moving body P detected by the ID information input unit 1 are the same type of information. For example, when the attribute information input unit 2 detects the position as the state of the moving body P, the ID information input unit 1 also detects the position as the state of the moving body P.

  If the state of the moving body P detected by the attribute information input unit 2 and the state of the mobile body P detected by the ID information input unit 1 are the same type of information and can be compared, the attribute information input unit 2 The accuracy of the state of the moving body P to be detected may be different from the accuracy of the state of the moving body P detected by the ID information input unit 1. For example, it is assumed that the attribute information input unit 2 can detect the position coordinates of the moving body P represented by two-dimensional coordinates within an error of 1 m. Further, it is assumed that the ID information input unit 1 can detect the position coordinates of the moving body P represented by two-dimensional coordinates within an error of 5 m. In this case, although the position detection accuracy differs between the attribute information input unit 2 and the ID information input unit 1, it is possible to compare the proximity of the moving object P using the mutual position detection results. Therefore, the detection accuracy may be different between the attribute information input unit 2 and the ID information input unit 1 as described above.

  In addition, as long as the attribute information input unit 2 and the ID information input unit 1 detect the same type of information as the state of the moving body P, there may be a plurality of types of states of the moving body P to be detected. . For example, both the attribute information input unit 2 and the ID information input unit 1 may detect the position and movement of the moving body P as the state of the moving body P.

  The detection of ID information by the ID information input unit 1 and the detection of attribute information by the attribute information input unit 2 are performed at the same time. Alternatively, the ID information input unit 1 and the attribute information input unit 2 detect ID information and attribute information asynchronously and input them to the mobile object attribute estimation unit 3, and the mobile object attribute estimation unit 3 receives the input ID information. And attribute information may be buffered for a certain period of time, and the attribute estimation process of the moving object may be performed using ID information and attribute information accumulated in the buffer every certain period of time. Alternatively, if the detection time of the information cannot be synchronized between the ID information input unit 1 and the attribute information input unit 2, the mobile object attribute estimation unit 3 is input within a certain period. The same detection time may be set for the ID information and the attribute information.

  The moving body attribute estimation unit 3 is input using the ID information of the plurality of moving bodies P input from the ID information input unit 1 and the attribute information of the plurality of moving bodies P input from the attribute information input unit 2. Each attribute information related to each ID information is selected. Here, the fact that the ID information and the attribute information are related means that the state information of the mobile body included in the ID information is similar to the state information of the mobile body included in the attribute information. For example, when the similarity between the state information of the mobile object included in the ID information and the state information of the mobile object included in the attribute information is calculated, a condition that the similarity is equal to or greater than a predetermined threshold (or threshold value) The condition that it is greater than the threshold) or the condition that it is less than or equal to the threshold (or the condition that it is less than the threshold). When the similarity is calculated so that the similarity is higher as the similarity is higher, a condition that the similarity is equal to or higher than the threshold (or a condition that the similarity is higher than the threshold) may be employed. In addition, when the similarity is calculated so that the similarity becomes higher as the similarity becomes higher, a condition that the similarity is equal to or lower than a threshold (or a condition that the similarity is lower than the threshold) may be employed. Here, the case where it is determined whether or not the state information is similar using a threshold is illustrated, but it may be determined whether or not the state information is similar by another method.

  The moving object attribute estimation unit 3 integrates the detection results of the same attribute values obtained with the passage of time for each attribute information related to the focused ID information, and determines the detection frequency obtained as a result. , And the likelihood that each attribute candidate is associated with the ID. When attention is paid to one piece of ID information, each attribute indicated by each piece of attribute information related to the ID information is set as an attribute candidate for an ID included in the focused ID information. The moving object attribute estimation unit 3 compares the likelihood calculated using the attribute candidate of the ID of interest, and determines that the attribute value having the maximum likelihood is the attribute corresponding to the ID. . This determination process can be said to be a process of estimating the attribute of the moving object identified by the ID.

  In addition, integrating the detection results of the same attribute is a predetermined value related to the number of times that the attribute value (for example, “male” or “female” in gender) is determined to be related to the focused ID. Is to perform the operation. For example, for each attribute value, the number of times determined to be related to the focused ID may be counted, and the count result may be used as the detection frequency. Alternatively, when a value of an attribute (for example, “male” or “female”) is associated with the focused ID once, an error according to a normal distribution is generated with respect to the numerical value “once” The calculated value may be calculated as a score, the sum of scores for each time determined to be related to the ID may be calculated, and the calculation result may be used as the detection frequency. Or the product of the score for every time linked | related with ID is calculated, and it is good also considering the calculation result as a detection frequency. In the following description, in order to simplify the description, the number of times associated with one ID is counted for each attribute value, and the count result is used as a detection frequency.

  The estimation result output unit 4 is an output device that outputs an attribute determined by the mobile object attribute estimation unit 3 to correspond to the ID (that is, an attribute estimated for the mobile object specified from the ID). The output mode of the attribute is not particularly limited. For example, when displaying the correspondence between the ID and the attribute, a display device may be used as the estimation result output unit 4. Hereinafter, a case where the estimation result output unit 4 displays the correspondence relationship between the ID and the attribute is taken as an example.

  FIG. 2 is a block diagram illustrating a configuration example of the moving object attribute estimation unit 3. The ID information input unit 1, the attribute information input unit 2, and the estimation result output unit 4 are also illustrated. The moving object attribute estimation unit 3 includes attribute candidate selection means 31, attribute likelihood calculation means 32, attribute estimation means 33, and attribute likelihood storage means 34.

  The attribute candidate selection unit 31 uses the ID information input from the ID information input unit 1 and the attribute information input from the attribute information input unit 2, and uses the state information included in the ID information and the state information included in the attribute information. Based on the similarity, each attribute information related to each ID information is selected. A method for evaluating the similarity between the state information included in the ID information and the state information included in the attribute information may be determined in advance according to the type of the state to be detected. For example, when detecting the position of a moving body as a state, the Euclidean distance between the position coordinates included in the ID information of interest as the state information and the position coordinates included in the state information in the attribute information is represented by the similarity of the state information. Calculate as If the Euclidean distance is within a predetermined threshold, the attribute candidate selection unit 31 selects the attribute information as attribute information related to the ID information of interest.

  Further, the attribute candidate selection unit 31 determines that the ID of the state information when the similarity between the state information does not satisfy a predetermined evaluation criterion (for example, a threshold) in any combination of the focused ID information and any attribute information. It may be determined that there is no attribute information related to the information, and the attribute information related to the ID information may not be selected.

  Further, the attribute candidate selection means 31 determines whether or not it is related to the ID information based on the detection time of the ID information when evaluating the similarity between the state information in the combination of the ID information and the attribute information. The attribute information to be determined may be limited. For example, only attribute information detected within n seconds before and after the detection time of ID information may be a target for determination as to whether or not it is related to ID information.

  Further, when evaluating the similarity between the state information included in the ID information and the state information included in the attribute information, if the accuracy of the state information included in the ID information and the accuracy of the state information included in the attribute information are different, An evaluation method considering the accuracy of the state information may be used. For example, the case where the position of a moving body is detected as a state will be described as an example. It is assumed that the ID information input unit 1 can detect the position coordinates of the moving body P represented by two-dimensional coordinates within an error of 5 m. In addition, it is assumed that the attribute information input unit 2 can detect the position coordinates of the moving body P represented by two-dimensional coordinates within an error of 1 m. At this time, if the region within a radius of 5 m from the two-dimensional coordinates detected by the ID information input unit 1 and the region within a radius of 1 m from the two-dimensional coordinates detected by the attribute information input unit 2 are partially overlapped, The position coordinates (state information) may be determined to be similar, and the attribute information may be determined to be attribute information related to the ID information.

  In addition, from the viewpoint of enabling the attribute candidate selection unit 31 to appropriately specify a combination of ID information and attribute information having high similarity between state information, values of values that can be detected as state information from individual mobile objects A wider range is preferred. By widening the range of values that can be detected as state information, it is possible to make it difficult for state information to overlap between different mobile objects. For example, when the ID information input unit 1 and the attribute information input unit 2 detect the movement of the moving object as the state information, “stop” is detected rather than only one of the two types of “stop” and “move”. , “Walking”, “moving by bicycle”, “moving by car”, and “moving by train” are preferably detected. Thus, by expanding the value that can be detected as the state information, the attribute candidate selection unit 31 can select ID information and attribute information with higher similarity.

  The attribute likelihood calculation means 32 integrates the detection results of the same attribute obtained over time for all attribute candidates selected for each ID information by the attribute candidate selection means 31, thereby Calculate the value detection frequency. Then, the detection frequency obtained for each attribute value is defined as the likelihood of each attribute value. For example, the attribute type is the color of the moving object, and each attribute information includes a value obtained by labeling the representative color of the moving object with “red”, “blue”, “yellow”, etc. Suppose that The attribute likelihood calculating means 32 refers to the value of the attribute included in each attribute information determined to be related to one focused ID, and calculates the appearance count of attribute information including “red”, “blue”. The number of appearances of the attribute information including the number of appearances of the attribute information including “yellow” is counted. In this example, the count result is the detection frequency of each attribute (red, blue, yellow, etc.). Then, the attribute likelihood calculating means 32 sets the detection frequency as the likelihood of the attribute of the moving object indicated by one focused ID. For example, if the detection frequency of “red” is “a”, the likelihood that the attribute of the moving object is red is a, and if the detection frequency of “blue” is “b”, the attribute of the moving object is The likelihood of being blue is b. If the detection frequency is high, the likelihood is large, and if the detection frequency is low, the likelihood is small.

  Alternatively, the attribute information input unit 2 may represent the representative color of the moving object by a value in the RGB color space or HSB (Hue, Saturation, Brightness) color space, and may be included in the attribute information. Then, the attribute likelihood calculating means 32 may generate a color histogram with an error distribution added with reference to the representative color, and add the frequency for each class in the color histogram generated from each attribute candidate. In this example, the addition result of the frequency of each class becomes the detection frequency of each attribute (individual class in the color histogram). Then, the attribute likelihood calculating unit 32 sets the detection frequency as the likelihood of the attribute of the moving object indicated by the ID of interest.

  When the attribute information includes a plurality of attribute types, the attribute likelihood calculating unit 32 performs the same process for each attribute type.

  Further, the attribute likelihood calculating means 32 has individual attribute values (“red”, “blue”, “yellow” in the above example) for each ID included in each ID information input from the ID information input unit 1. Etc.).

  The attribute likelihood storage unit 34 is a storage device that stores each likelihood determined by the attribute likelihood calculation unit 32.

  The attribute estimation means 33 uses the likelihood determined by the attribute likelihood calculation means 32 to select the attribute value having the highest likelihood for each ID of the moving object, and associates it with the ID. Then, the correlation between the ID and the attribute value is output to the estimation result output unit 4. The value of the attribute associated with a certain ID means that it is estimated as the attribute of the moving object identified by that ID.

  In the first embodiment, the attribute candidate selection unit 31, the attribute likelihood calculation unit 32, and the attribute estimation unit 33 are realized by, for example, a CPU of a computer that operates according to a moving object attribute estimation program. In this case, a program storage device (not shown) of the computer stores the moving object attribute estimation program, and the CPU reads the program, and according to the program, attribute candidate selection means 31, attribute likelihood calculation means 32, attribute estimation means. It is sufficient to operate as 33. Moreover, the attribute candidate selection means 31, the attribute likelihood calculation means 32, and the attribute estimation means 33 may each be implemented by different hardware.

Next, the operation will be described.
FIG. 3 is a flowchart illustrating an example of processing progress of the moving object attribute estimation system according to the first embodiment.

  The attribute information input unit 2 detects the attribute and state of the moving object, and inputs the attribute information including the attribute and state information of the moving object and the detection time thereof to the attribute candidate selecting unit 31. The attribute candidate selection means 31 acquires the attribute information (step S1). Here, in order to simplify the description, the attribute information input unit 2 detects only one type of attribute, and the attribute information does not include the attribute type as an example.

  Further, the ID information input unit 1 detects the ID and state of the moving object, and inputs the ID information including the ID and state information of the moving object and their detection times to the attribute candidate selecting unit 31. The attribute candidate selection means 31 acquires the ID information (step S2).

  Next, the attribute candidate selection means 31 determines whether or not the process of selecting attribute information related to the ID information has been completed for each ID information acquired in step S2 (step S3). When unselected ID information exists (Yes in step S3), the attribute candidate selecting unit 31 selects one unselected ID information (step S4).

  Next, the attribute candidate selection means 31 selects the attribute information relevant to ID information selected at step S4 from each attribute information acquired at step S1 (step S5). In this example, the attribute candidate selection means 31 associates attribute information including attributes and state information detected within a predetermined time before and after that time with reference to the detection time included in the selected ID information. It is a judgment target of whether or not to do. And the attribute candidate selection means 31 is the attribute information relevant to the ID information among the attribute information to be determined, the attribute information whose similarity between the state information satisfies the criteria with the selected ID information Select as.

  The attribute included in the selected attribute information is an attribute candidate for the ID included in the selected ID information.

  Evaluation of the relationship between the ID information and the attribute information in step S5 will be described using a specific example. In this example, it is assumed that the attribute information input unit 2 detects the position of the moving body (more specifically, the position coordinates) as the state of the moving body. Further, the case where the Euclidean distance between the position coordinates included in the ID information and the attribute information is used as the similarity of the state information will be described as an example.

  FIG. 4 is a schematic diagram illustrating detection positions of the moving object ID “i1” and the attributes z1, z2, z3, and z4 in the detection region 50. Hereinafter, ID information including ID “i1” will be referred to as “I1”. Similarly, attribute information including the attribute z1 is denoted as “Z1”. Similarly, attribute information including other attributes z2, z3, and z4 is written as “Z2”, “Z3”, and “Z4” using capital letters. Similarly, in the following description, IDs and attributes themselves are indicated by lowercase letters, and ID information and attribute information are indicated by uppercase letters. FIG. 5 is an explanatory diagram showing detection times of the ID “i1” and the attributes z1, z2, z3, and z4 shown in FIG. In FIG. 5, the attribute z1 is detected at time t1, the ID “i1” and the attributes z2 and z3 are detected at time t2, and the attribute z4 is detected at time t5. In this example, the ID information input unit 1 and the attribute information input unit 2 regularly try to detect IDs and attributes, and the intervals between the times are constant at times t1 to t5. Let T be the interval. Here, it is assumed that the attribute candidate selection unit 31 selects the ID information I1.

  In step S5, the attribute candidate selection unit 31 calculates a difference in detection time between the selected ID information I1 and each attribute information, and determines whether or not there is an association with the ID information I1. Refine attribute information. In this example, it is assumed that the attribute candidate selection unit 31 narrows down the attribute information including the detection times belonging to the time zone before and after the detection time included in the ID information. Therefore, the attribute candidate selection unit 31 uses the detection time t2 included in the ID information I1 as a reference, and sets the attribute information Z1, Z2, and Z3 including the attributes detected during the times t1 to t3 as relevance to the ID information. Narrowing down as the determination target of presence / absence of the attribute information Z4 is excluded.

  Next, the attribute candidate selection unit 31 calculates the Euclidean distance between the position coordinates included in the ID information I1 and the position coordinates included in the attribute information Z1, Z2, and Z3 as determination targets. FIG. 6 shows an example of a calculation result of the Euclidean distance between the position coordinates included in the ID information I1 and the position coordinates included in the attribute information Z1, Z2, and Z3. FIG. 6 shows that the distance between the position coordinates included in the ID information I1 and the position coordinates included in the attribute information Z1 is 1. Similarly, the distance is 1.5 for the combination of ID information I1 and attribute information Z2, and the distance is 3 for the combination of ID information I1 and attribute information Z3. Each distance is a similarity between state information (position coordinates) in a set of ID information and attribute information. It is assumed that a condition for determining that attribute information is high and attribute information is related to ID information is set to a threshold value “2” or less. Then, the attribute candidate selection unit 31 selects the attribute information Z1 and Z2 as the attribute information related to the ID information I1, and excludes the attribute information Z3. In this way, the attribute candidate selection unit 31 selects attribute information that satisfies a predetermined criterion for the selected ID information as attribute information related to the ID information I1.

  After step S5, the attribute likelihood calculating means 32 updates the likelihood related to the ID included in the ID information selected in step S4 using the attribute information selected in step S5 (step S6). In step S6, the likelihood (detection frequency) stored in the attribute likelihood storage means 34 is read for each possible value of the attribute regarding the ID included in the selected ID information, and the likelihood may be updated. The attribute likelihood calculating unit 32 stores the updated likelihood in the attribute likelihood storage unit 34.

  The likelihood update process in step S6 will be described using a specific example. In this example, in order to simplify the description, a case where the color of a moving object is detected as an attribute will be described as an example. More specifically, a case where the detected attribute is expressed based on the hue (Hue) of the HSB color space is taken as an example. The hue is expressed by a value in the range of 0 to 360. Here, this range is divided every 30 and the attribute is represented by a range of hues divided into 12 stages.

  FIG. 7 is a schematic diagram illustrating a detection position of an attribute included in related information selected as attribute information related to the ID information I1. In this example, it is assumed that ID “i1” is detected at times t2, t5, t8, and t10. FIG. 8 is an explanatory diagram showing attribute values included in the attribute information selected as related to the ID information including the ID “i1” detected at each time shown in FIG. 7 and 8 indicate that the attribute information related to the ID information I1 including t2 as the detection time includes the attributes z1 and z2. Similarly, the attribute information related to the ID information I1 including t5 as the detection time indicates that the attributes z5, z6, and z7 are included. The attribute information related to the ID information I1 including t8 as the detection time indicates that the attributes z10 and z13 are included. The attribute information related to the ID information I1 including t10 as the detection time indicates that the attributes z15 and z16 are included. In this example, the current time is t10.

  The attribute likelihood calculating means 32 holds, as a detection frequency, how many hues have been detected in the past and how many times it has been detected in the past in order to obtain the likelihood associated with the ID (i1 in this example). However, this detection frequency is calculated with respect to the number of times detected during the past fixed period. Therefore, even if an attribute is detected before a certain period in the past, the detection is not reflected in the detection frequency. Therefore, the number of detections before a certain period with the current time as a reference is excluded from the detection frequency. Here, in order to simplify the description, it is assumed that no attribute is detected before a certain period with the current time t10 as a reference.

  In the example shown in FIG. 8, color information (120, 0, 0) in the HSB color space is detected as the attribute z1 at time t2. In this attribute z1, since the hue value is 120, when the attribute likelihood calculating unit 32 selects the attribute information Z1 as the attribute information related to the ID information I1 including the time t2 by detecting z1. The score “1” corresponding to the number of detections for one time is added to the hue range “90 to 120” to which the hue value 120 belongs. Similarly, the attribute likelihood calculating means 32 adds the score “1” to the hue range to which the hue value belongs for the other attributes z2, z5, z6, z7, z10, z13, z15, and z16.

  At time t10, it is assumed that the score distribution illustrated in FIG. 9 is obtained as a result of adding scores based on the attributes detected at times t2 to t8. Then, it is assumed that information representing the score distribution illustrated in FIG. 9 is stored in the attribute likelihood storage unit 34. In FIG. 9, the horizontal axis represents 12 types of hue stages determined by dividing the range of 0 to 360 that can be taken by the hue into 30. In addition, the vertical axis shown in FIG. 9 represents the detection frequency of each stage of the hue. That is, the score distribution shown in FIG. 9 represents the detection frequency of individual attributes (individual ranges of 12 divided hues) related to i1. In the example shown in FIG. 9, the hue values in the range of “30 to 60” and the range of “90 to 120” are detected most. In this state, the attribute corresponding to the ID “i1” is uniquely set. Cannot be determined.

  When the process proceeds to step S6 based on the ID information and attribute information at time t10, the attribute likelihood calculating unit 32 refers to the score distribution (detection frequency of each attribute) regarding i1 shown in FIG. Then, the attribute likelihood calculating unit 32 adds 1 to each of the hue value ranges to which the hue values of the attributes z15 and z16 detected at time t10 belong in the distribution of scores shown in FIG. That is, since the hue value of attribute z15 is 120, score 1 is added to the range of “90 to 120”, and since the hue value of attribute z16 is 120, score 1 is further added to the range of “90 to 120”. to add. As a result, the detection frequency of individual attributes (individual ranges of 12 divided hues) relating to i1 is as shown in FIG. The meanings of the horizontal axis and the vertical axis shown in FIG. 10 are the same as those of the horizontal axis and the vertical axis in FIG. In FIG. 10, most hue values in the range of “90 to 120” are detected. The attribute likelihood calculating unit 32 stores the score updated in this way (see FIG. 10) in the attribute likelihood storing unit 34. Specifically, the ID of the moving object and the likelihood obtained for each attribute value with respect to the ID are stored. When attributes are detected for a plurality of attribute types, the likelihood of each value of the attribute may be stored for each attribute type with respect to one ID. In this case, it is also stored which attribute type the likelihood relates to.

  Thus, the attribute likelihood calculating means 32 is based on the value of the attribute related to the input ID every time the ID information and the attribute information are input from the ID information input unit 1 and the attribute information input unit 2. The detection frequency (score) of the attribute value is updated.

  Further, the detection frequency obtained in step S6 is the likelihood of each attribute candidate regarding the ID of interest.

  When step S6 ends, the moving object attribute estimation unit 3 repeats the processes after step S3. If it is determined that there is no unselected ID information (No in step S3), the attribute estimation unit 33 is updated in step S6 for each ID included in each ID information acquired from the ID information input unit 1. The latest likelihood (detection frequency counted for each attribute) is referred to, and the attribute with the highest likelihood is selected and associated with the ID (step S7). By this process, the attribute estimation means 33 estimates the attribute of the moving body identified by the ID for each ID. Moreover, the attribute estimation means 33 displays the correspondence between each ID and the attribute on the estimation result output unit 4.

  For example, it is assumed that the detection frequency of each attribute (each range of hue value) shown in FIG. 10 is obtained for ID “i1” based on the ID and attribute detection results at the current time t10. In this case, since the range of the hue value with the highest detection frequency is “90 to 120”, the attribute estimation unit 33 associates the range of the hue value with “90 to 120” to the ID “i1”. This means that the hue value of the moving object identified by “i1” is estimated to be in the range of “90 to 120”.

  According to the present embodiment, the attribute candidate selection unit 31 includes the ID information of the plurality of mobile bodies P input from the ID information input unit 1 and the attribute information of the plurality of mobile bodies P input from the attribute information input unit 2. Are used to select each attribute information related to each input ID information. Then, the attribute likelihood calculating means 32 integrates the detection results of the same attribute value obtained with the passage of time, calculates the detection frequency obtained as a result, and the likelihood that the attribute candidate corresponds to the ID. To do. And the attribute estimation means 33 estimates the attribute of the moving body identified by ID by selecting the value of the attribute with the highest likelihood for every ID. Therefore, an opportunity to associate the ID and the attribute can be obtained for each detection time of the ID and the attribute. As a result, it is possible to appropriately estimate the attribute corresponding to the detected moving object. Moreover, even if there are a large number of moving objects, the attributes corresponding to the moving objects can be estimated appropriately.

  Compared with the present invention and the method described in Patent Document 1, in the method described in Patent Document 1, the opportunity to associate the ID of the moving object with the ID of the shadow is when the moving object enters the predetermined area. Only. Therefore, when a large number of moving bodies enter a predetermined area at the same time, it becomes difficult to associate the ID of the moving body with the shadow ID.

  On the other hand, in the present invention, an opportunity for associating the ID and attribute of the moving object is obtained every time the ID and attribute are detected. And ID and an attribute are matched for every ID of a moving body based on the detection frequency (likelihood) of the value of an attribute. Therefore, it is possible to appropriately estimate the attribute corresponding to the moving object.

Embodiment 2. FIG.
FIG. 11 is a block diagram illustrating an example of the moving object attribute estimation system according to the second embodiment of this invention. The same components as those of the first embodiment are denoted by the same reference numerals as those in FIG. The mobile object attribute estimation system of the second embodiment includes an ID information input unit 1, an attribute information input unit 2, a mobile object attribute estimation unit 3a, and an estimation result output unit 4. The ID information input unit 1, the attribute information input unit 2, and the estimation result output unit 4 are the same as those elements in the first embodiment, and a description thereof is omitted.

  The moving object attribute estimation unit 3a includes attribute candidate selection means 31, attribute likelihood calculation means 32a, attribute estimation means 33, attribute likelihood storage means 34a, and attribute model storage means 35. The attribute candidate selection means 31 and the attribute estimation means 33 are the same as those elements in the first embodiment, and a description thereof is omitted.

  The attribute likelihood storage unit 34a includes the ID of the mobile body obtained from the ID information, the likelihood obtained for each possible value of the attribute, the attribute type of the attribute, and the opportunity for obtaining the likelihood. The detected ID time is stored in association with each other. Regarding the ID, the likelihood obtained for each possible value of the attribute is, for example, the detection frequency of each value of the attribute as illustrated in FIG. 9 and FIG.

  The attribute model storage unit 35 is a storage device that stores an attribute type and a model of change in detection frequency over time defined in advance. The detection frequency change model determines whether or not the likelihood (detection frequency) obtained for each possible value of the attribute with respect to each ID is changed over time. It is information that defines how to change the detection frequency when it is satisfied. The change model of the detection frequency may be determined in advance according to the attribute type.

  For example, it is assumed that the moving body is a human and the attribute information input unit 2 detects gender as an attribute. Human gender does not change over time. Therefore, a change model is defined in which the detection frequency obtained for each possible value (“male” and “female”) of each ID is not changed over time, and is stored in the attribute model storage unit 35. Just keep it.

  Further, for example, it is assumed that the moving body is a human and the attribute information input unit 2 detects age as an attribute. The age of a human increases by one year for every year. Therefore, the detection frequency obtained for each value (each age such as “31 years old”, “32 years old”, etc.) that can be attributed to each ID is 1 when one year has passed since the detection frequency was obtained. A change model of changing to the detection frequency of the attribute value increased by the year (that is, the age one year old) may be determined and stored in the attribute model storage means 35. FIG. 12 is an explanatory diagram illustrating an example of a change in detection frequency by the conversion model when the attribute type is age. Assume that the attribute likelihood storage means 34a stores the detection frequency for each age exemplified in FIG. When this detection frequency is changed by the above-described change model, each detection frequency becomes the detection frequency of the age one year old when one year has elapsed since the detection frequency illustrated in FIG. As described above, the detection frequency of each attribute value is changed. For example, the detection frequency indicating the number of times the age is detected as 31 years old is changed to the detection frequency detected as the age 32 (see FIG. 12B). The same applies to the detection frequencies of other ages.

  Further, for example, it is assumed that the moving body is a human and the attribute information input unit 2 detects the color of clothes as an attribute. The color of clothes is likely to change every day, and it is impossible to predict what clothes will change when one day passes. Therefore, a change that the detection frequency obtained for each value that can be taken by the attribute for each ID (for example, the color of clothes such as “dark blue” and “gray”) is initialized to 0 once a day at a predetermined time. A model may be determined and stored in the attribute model storage means 35.

  For each ID input from the ID information input unit 1, the attribute likelihood calculating unit 32a calculates a detection frequency for each attribute value and a detection time of the ID that triggered the detection frequency. It reads out from the attribute likelihood storage means 34a, and reads out a change model of the detection frequency determined for the attribute type. Then, the attribute likelihood calculating unit 32a determines whether or not to change the detection frequency for each attribute value based on the change model, and further changes the attribute value for each attribute value according to the change model. Change the detection frequency. This means that the detection frequency for each attribute value is corrected to a value suitable for the current time according to the passage of time from the time when the detection frequency for each attribute value is obtained to the current time. Further, when the attribute likelihood calculating unit 32a determines that the detection frequency for each attribute value is not changed based on the change model, the attribute likelihood calculating unit 32a calculates the detection frequency for each attribute value read from the attribute likelihood storage unit 34a. Do not change.

  The attribute likelihood calculating means 32a refers to the attribute value included in the attribute information determined to be related to the focused ID information. And the process which increases the score which shows the detection frequency of the newly detected attribute value with respect to the detection frequency for every value of the attribute after said process using a change model is performed. That is, an update process is performed on the detection frequency (likelihood) for each attribute value processed according to the change model based on the newly detected attribute. In the “process according to the change model”, the detection frequency may not be changed.

  In the second embodiment, the attribute candidate selection unit 31, the attribute likelihood calculation unit 32a, and the attribute estimation unit 33 are realized by a CPU of a computer that operates according to a moving object attribute estimation program, for example. In this case, a program storage device (not shown) of the computer stores the moving object attribute estimation program, and the CPU reads the program, and according to the program, attribute candidate selection means 31, attribute likelihood calculation means 32a and attribute estimation means. It is sufficient to operate as 33. Moreover, the attribute candidate selection means 31, the attribute likelihood calculation means 32a, and the attribute estimation means 33 may each be implement | achieved by another hardware. Note that the attribute likelihood storage unit 34a and the attribute model storage unit 35 may be provided outside the mobile object attribute estimation unit 3a. Further, the attribute likelihood storage unit 34a and the attribute model storage unit 35 may be realized by the same storage device.

  FIG. 13 is a flowchart illustrating an example of processing progress of the moving object attribute estimation unit 3a according to the second embodiment. However, the same processes as those in the first embodiment are denoted by the same reference numerals as those in FIG. 3, and detailed description thereof is omitted.

  The processing up to step S4 is the same as in the first embodiment. After step S4, the attribute likelihood calculating means 32a determines the detection frequency (likelihood) for each attribute value, the attribute type of the attribute, and the detection frequency for the ID included in the ID information selected in step S4. The detection time of the ID that triggered the request is read from the attribute likelihood storage unit 34a (step S41).

  Then, the attribute likelihood calculating unit 32a reads out the change model corresponding to the attribute type read out in step S41 from the attribute model storage unit 35. Then, the detection frequency for each attribute value read in step S41 and the detection time of the ID that triggers the detection frequency are compared with the change model, and for each attribute value read in step S41. It is determined whether or not the detection frequency is changed. If it is determined to be changed, the detection frequency for each attribute value is changed according to the change model (step S42). If it is determined not to be changed, the detection frequency for each attribute value is not changed (step S42).

  A specific example of the process of step S42 is shown. For example, it is assumed that the attribute type acquired from the attribute likelihood storage unit 34a is “moving body color”. As a change model related to “moving body color”, it is assumed that the detection frequency for each attribute value is set to 0 every 24 hours, with 3:00 am as a boundary. In this case, the detection time of the ID included in the ID information selected in step S4 is before 3:00 am, and the time read in step S41 (the trigger for obtaining the detection frequency read from the attribute likelihood storage means 34a) The attribute likelihood calculation means 32a does not change the detection frequency for each attribute value read in step S41. If the detection time of the ID included in the ID information selected in step S4 is after 3 am and the time read out in step S41 is before 3 am on the same day, the attribute likelihood calculating means 32a The detection frequency for each attribute value read in step S41 is initialized to 0.

  The above change model is an example, and the change model is not limited to the above example.

  After the process of step S42, the same process as step S5 in the first embodiment is performed.

  Subsequently, the detection frequency (likelihood) after the processing in step S42 is updated using the attribute information selected in step S5 (step S61). The operation of the update process is the same as that in step S6 in the first embodiment. However, it differs from the first embodiment in that the update target is the detection frequency (likelihood) after the processing in step S42.

  After step S61, the processes after step S3 are repeated. The other points are the same as in the first embodiment.

  In the second embodiment, when calculating the likelihood at the latest time using the likelihood calculated in the past, the likelihood of the likelihood calculated in the past using the detection frequency change model predetermined for each attribute type is used. By estimating the value at the current time (step S42), the attribute corresponding to the ID can be accurately estimated even when the ID attribute estimation interval is not constant.

Embodiment 3. FIG.
FIG. 14 is a block diagram illustrating an example of the moving object attribute estimation system according to the third embodiment of this invention. The same components as those of the first embodiment are denoted by the same reference numerals as those in FIG. The mobile object attribute estimation system according to the third embodiment includes an ID information input unit 1, an attribute information input unit 2, a mobile object attribute estimation unit 3 b, and an estimation result output unit 4. The ID information input unit 1, the attribute information input unit 2, and the estimation result output unit 4 are the same as those elements in the first embodiment, and a description thereof is omitted.

  The moving object attribute estimation unit 3b includes attribute candidate selection means 31, attribute likelihood calculation means 32b, attribute estimation means 33, and attribute likelihood storage means 34b. The attribute candidate selection means 31 and the attribute estimation means 33 are the same as those elements in the first embodiment, and a description thereof is omitted.

  The attribute likelihood calculating means 32b calculates the likelihood that each attribute value corresponds to each ID for all attribute candidates selected for each ID information by the attribute candidate selecting means 31. Here, the attribute likelihood calculation means 32b calculates the likelihood for each time zone based on the ID information and the attribute information obtained from the ID information input unit 1 and the attribute information input unit 2 during that time zone. The likelihood obtained for each time zone is referred to as a first likelihood. Further, the attribute likelihood calculating means 32b, when newly calculating the first likelihood, calculates the first likelihood calculated for the ID of interest during the past fixed time from the present time, for each attribute. Integrate by value. The likelihood obtained by this integration process is referred to as a second likelihood. In the following description, each of the above time zones is represented by the time at the end point of the time zone.

  When calculating the first likelihood, the attribute likelihood calculating means 32b does not depend on the detection frequency of the attribute candidate selected by the attribute candidate selecting means 31, and the first is determined by the presence or absence of the detection result of each attribute value. Calculate the likelihood. For example, using “moving body color” as the attribute type, the attribute value “red” is detected once as an attribute candidate that may correspond to the ID of a moving body detected at a certain time, and the attribute value “black” "Is detected five times. In this case, for example, the attribute likelihood calculating unit 32b may set the first likelihood of the attribute value “red” and the first likelihood of “black” to “1” for the ID of the moving object. Good. In addition, regarding the ID of the moving object, the first likelihood of other attribute values (other colors) not detected may be set to “0”. Here, the case where the first likelihood is expressed by binary values of “0” or “1” is illustrated, but a value that generates an error distribution based on the detected attribute value is used as the first likelihood. It may be used.

  It can be said that the first likelihood is a likelihood determined depending on whether or not the attribute value is included in the attribute information determined to be related to the ID information for each attribute value.

  Further, when the attribute likelihood calculating unit 32b calculates the first likelihood for each attribute value for the focused ID, the time likelihood (in this example, the time zone for which the first likelihood is calculated) And the ID and the first likelihood calculated for each attribute value are stored in the attribute likelihood calculating means 34b.

  When the attribute likelihood calculation unit 32b calculates the first likelihood for each attribute value for the focused ID, the attribute likelihood calculation unit 34b stores the first likelihood as described above. The first likelihood for each attribute value calculated during the past certain period from that time is read. This first likelihood is related to the ID of interest. And the attribute likelihood calculation means 32b integrates those 1st likelihood for every attribute value, and calculates a 2nd likelihood. An example of the integration process will be described later.

  The attribute estimation unit 33 selects an attribute value having the maximum second likelihood for the focused ID, and associates the attribute value with the ID. That is, in the present embodiment, the attribute value having the highest second likelihood is estimated as the attribute value of the focused ID. Then, the attribute estimation unit 33 causes the estimation result output unit 4 to output the correspondence between the ID and the attribute value.

  The attribute likelihood storage unit 34b includes the ID of the moving body obtained from the ID information, the likelihood (first likelihood) obtained for each possible value of the attribute with respect to the ID, the attribute type of the attribute, The time indicating the time zone for which the likelihood is to be obtained (in this example, the time that is the end point of the above time zone) is stored in association with each other. In the present embodiment, the attribute storage unit 34b calculates not only each first likelihood for each ID in the latest time zone calculated by the attribute likelihood calculation unit 32b but also each ID for each ID in the past time zone. Each likelihood of 1 is also stored.

  In the third embodiment, the attribute candidate selection unit 31, the attribute likelihood calculation unit 32b, and the attribute estimation unit 33 are realized by, for example, a CPU of a computer that operates according to a moving object attribute estimation program. In this case, a program storage device (not shown) of the computer stores the moving object attribute estimation program, and the CPU reads the program, and according to the program, attribute candidate selection means 31, attribute likelihood calculation means 32b, and attribute estimation means. It is sufficient to operate as 33. Moreover, the attribute candidate selection means 31, the attribute likelihood calculation means 32b, and the attribute estimation means 33 may each be implemented by different hardware. Note that the attribute likelihood storage unit 34b may be provided outside the mobile object attribute estimation unit 3b.

  FIG. 15 is a flowchart illustrating an example of processing progress of the moving object attribute estimation unit 3b according to the third embodiment. However, the same processes as those in the first embodiment are denoted by the same reference numerals as those in FIG. 3, and detailed description thereof is omitted.

  The processing up to step S5 is the same as in the first embodiment. After step S5, the attribute likelihood calculating means 32b calculates the first likelihood based on whether or not the attribute value is detected for the ID included in the ID information selected in step S4. For example, the first likelihood is set to “1” for the attribute value included in the attribute information determined to be related to the selected ID information, and the first likelihood is set to “0” for the other attributes. It may be determined. Note that this first likelihood determination method is an example, and the first likelihood may be obtained by another method. The attribute likelihood calculating means 32b includes the ID included in the ID information selected in step S4, the first likelihood defined for each attribute value for the ID, the attribute type of the attribute, and the first likelihood. The attribute likelihood storage means 34b stores the time representing the time zone for which the degree is to be obtained (the time zone in which the ID information and the attribute information used to calculate the first likelihood are collected) (step S71).

  Next, the attribute likelihood calculating means 32b obtains the first likelihood for each attribute value obtained with respect to the ID selected in step S4, and the first likelihood obtained during a certain past time from the current time. , Read from the attribute likelihood storage means 34b (step S72).

  Further, the attribute likelihood calculating unit 32b integrates the first likelihoods for a certain past time from the current time acquired in step S72 for each attribute value (step S73). In step S73, the attribute likelihood calculation unit 32b performs, for example, a process of integrating the first likelihood for a certain past time for each attribute value as the integration process, and the integration result is set as the second likelihood. do it. However, the calculation in the integration process in step S73 is not limited to the above example. For example, the attribute likelihood calculating unit 32b may perform a process of adding the first likelihood for a certain past time for each attribute value, and the addition result may be used as the second likelihood.

  As a result of step S73, the second likelihood is obtained for each attribute value with respect to the focused ID.

  After step S73, the processes after step S3 are repeated. When the process proceeds to step S7, the attribute estimation unit 33 selects an attribute value having the maximum second likelihood for one ID, and performs a process of associating the ID with the attribute value. Do it every time.

  The other points are the same as in the first embodiment.

  In the third embodiment, the attribute likelihood calculating means 32b uses the ID likelihood detected for each time zone and the first likelihood for each attribute value calculated by the attribute information for the focused ID as the attribute likelihood. It memorize | stores in the memory | storage means 34b. Then, the attribute likelihood calculating unit 32b calculates the first likelihood (first likelihood related to the focused ID) for each attribute value calculated from the attribute likelihood storage unit 34 during the past fixed time from the current time. ) And integrate for each attribute value. Then, the attribute estimation unit 33 selects the attribute value having the maximum second likelihood for each ID, and associates the attribute value with the ID. By such processing, even if there is a bias in the detection tendency of attribute values that may be associated with a certain ID, the influence of such a bias is reduced and the mobile object identified by the ID is identified. The attribute value can be estimated appropriately.

  For example, it is assumed that the attribute type is “color of a moving object” and the true attribute value of a certain moving object is “red”. Here, at time t1, it is assumed that the attribute value “red” is detected once and the attribute value “black” is detected five times from the attribute information determined to be related to the ID information including the ID of the moving object. . Furthermore, it is assumed that the attribute value “red” is detected once and the attribute value “blue” is detected twice from the attribute information determined to be related to the ID information including the ID of the moving object at time t2. If the attribute value of the moving object is simply estimated based on the detection frequency of each attribute value during these periods, the detection number of “red” is 2, the detection number of “black” is 5, and “blue”. When the number of detections of 2 is 2, “black” is selected. However, the true attribute value of a certain moving body should always be detected as an attribute candidate even if time passes. In the third embodiment, for each ID, the first likelihood is determined for each attribute value based on whether or not the attribute value is detected. Then, the attribute value is estimated based on the second likelihood obtained by integrating the first likelihood obtained in the past certain time. Therefore, it is possible to estimate an attribute value that is continuously detected over time as an attribute value of the moving object. To explain using the above example, in the third embodiment, the first likelihoods of “red”, “black”, and “blue” at time t1 are, for example, “1”, “1”, “ 0 ". The first likelihoods of “red”, “black”, and “blue” at time t2 are, for example, “1”, “0”, and “1”, respectively. Taking the case where the first likelihood is integrated by integration as an example, the integration result (that is, the second likelihood) of the first likelihood regarding “red”, “black”, and “blue” is “1”. ”,“ 0 ”, and“ 0 ”, and“ red ”that is the true attribute value of the moving object is appropriately selected.

  Next, the minimum configuration of the present invention will be described. FIG. 16 is a block diagram showing an example of the minimum configuration of the moving object attribute estimation system of the present invention. The moving object attribute estimation system of the present invention includes attribute information selection means 91, likelihood derivation means 92, and attribute selection means 93.

  The attribute information selection unit 91 (for example, the attribute candidate selection unit 31) includes a mobile object ID, state information indicating the state of the mobile object, ID information including the ID and the detection time of the state information, and the mobile object. , Attribute information including the state information indicating the state of the mobile object, and the detection time of the attribute and the state information are input, ID information is sequentially selected, and the detection time included in the selected ID information is set. Attribute information including detection time belonging to a predetermined time zone as a reference, and attribute information determined to be similar to the state information included in the selected ID information is related to the selected ID information To select as attribute information.

  The likelihood deriving unit 92 (for example, attribute likelihood calculating unit 32, 32a) is identified by the ID included in the ID information for each attribute value included in each attribute information determined to be related to the ID information. The likelihood corresponding to the attribute of the moving object is obtained.

  The attribute selection unit 93 (for example, the attribute estimation unit 33) selects a value of the attribute having the maximum likelihood for one ID, and performs a process of associating the ID with the value of the selected attribute. To do.

  With the configuration as described above, it is possible to appropriately estimate the attribute corresponding to the detected moving object.

  A part or all of the above-described embodiment can be described as in the following supplementary notes, but is not limited thereto.

(Supplementary note 1) ID of a moving object, state information indicating the state of the moving object, ID information including the ID and the detection time of the state information, an attribute of the moving object, and a state indicating the state of the moving object Information and attribute information including the detection time of the attribute and status information are input, ID information is sequentially selected, and detection times belonging to a predetermined time zone based on the detection time included in the selected ID information Attribute information selection means for selecting, as attribute information related to the selected ID information, attribute information that is attribute information that is included and is determined to be similar to the state information included in the selected ID information A likelihood deriving unit for obtaining a likelihood corresponding to the attribute of the moving object identified by the ID included in the ID information for each attribute value included in each attribute information determined to be related to the ID information; I A mobile object attribute estimation comprising: attribute selection means for selecting, for each ID, a value of an attribute having a maximum likelihood and associating the ID with the value of the selected attribute system.

(Supplementary Note 2) Attribute likelihood storage means for storing the likelihood obtained for each attribute value with respect to each ID and the likelihood obtained for each attribute value with respect to each ID Change model storage means for storing a change model of likelihood that defines how to be corrected as time elapses from the time to the current time, and the likelihood derivation means newly adds an ID to the attribute information selection means When information is input, the likelihood calculated for each attribute value is read from the attribute likelihood storage means, the read likelihood is corrected according to the change model, the corrected likelihood, and the ID information The mobile object attribute estimation system according to appendix 1, wherein the likelihood of the attribute value related to the ID included in the ID information is updated based on the attribute value included in each attribute information determined to be related to the ID.

(Supplementary Note 3) The likelihood deriving means obtains a detection frequency for each attribute value with respect to the ID included in the ID information based on the attribute value included in each attribute information determined to be related to the ID information. The mobile object attribute estimation system according to Supplementary Note 1 or Supplementary Note 2, wherein the detection frequency for each attribute value is a likelihood for each attribute value.

(Supplementary Note 4) The likelihood deriving unit obtains a detection frequency for each attribute value with respect to the ID included in the ID information based on the attribute value included in each attribute information determined to be related to the ID information. The detection frequency for each attribute value is set as the likelihood for each attribute value, and the ID and the likelihood for each attribute value related to the ID are stored in the attribute likelihood storage means to select attribute information. When the ID information is newly input to the means, the likelihood obtained for each attribute value is read from the attribute likelihood storage means, the read likelihood is corrected according to the change model, and the corrected likelihood And obtaining the detection frequency for each attribute value with respect to the ID included in the ID information based on the attribute value included in the attribute information determined to be related to the ID information. The moving body attribute estimation system according to attachment 2 to be updated.

(Additional remark 5) It has the attribute likelihood memory | storage means which memorize | stores the likelihood calculated | required for every attribute value regarding each ID, and the likelihood derivation means is attribute information determined to relate to ID information for every attribute value The first likelihood related to the ID included in the ID information is determined according to whether or not the attribute value is included in the ID information, and the first likelihood determined for each attribute value regarding the ID is The first likelihood for each attribute value related to the ID, which is stored in the attribute likelihood storage means and determined during the past fixed time, is read from the attribute likelihood storage means, and the attribute likelihood is determined for each attribute value. The second likelihood is calculated by integrating the first likelihoods of the past certain time read from the degree storage means, and the attribute selection means has the second likelihood maximum for one ID. The attribute value is selected and the ID is associated with the selected attribute value. Mobile attribute estimation system according to Note 1, and performs for each ID.

(Supplementary note 6) ID information including the ID of the moving object, state information indicating the state of the moving object, the ID and the detection time of the state information, the attribute of the moving object, and the state indicating the state of the moving object Information and attribute information including the detection time of the attribute and status information are input, ID information is sequentially selected, and detection times belonging to a predetermined time zone based on the detection time included in the selected ID information Attribute information that is determined to be similar to the state information included in the selected ID information, and is selected as attribute information related to the selected ID information and related to the ID information. Then, the likelihood corresponding to the attribute of the moving object identified by the ID included in the ID information is obtained for each attribute value included in each determined attribute information, and the likelihood becomes maximum for one ID. Attributes Select a value, the processing for associating the value of the selected attribute to the ID, mobile attribute estimating method characterized by performing for each ID.

(Supplementary Note 7) The likelihood obtained for each attribute value with respect to each ID is stored, and the likelihood obtained for each attribute value with respect to each ID is represented as the time elapsed from the time when the likelihood was obtained to the current time. Accordingly, the likelihood change model that defines how to modify the data is stored in advance, and when the ID information is newly input, the likelihood obtained for each attribute value with respect to the ID included in the ID information. The degree of the attribute value related to the ID included in the ID information is corrected based on the likelihood after the correction and the attribute value included in each attribute information determined to be related to the ID information. The moving body attribute estimation method according to attachment 6, wherein the likelihood is updated.

(Additional remark 8) Based on the value of the attribute included in each attribute information determined to be related to the ID information, the detection frequency is calculated for each attribute value with respect to the ID included in the ID information. The mobile object attribute estimation method according to appendix 6 or appendix 7, wherein the detection frequency is a likelihood for each attribute value.

(Additional remark 9) The 1st likelihood regarding ID contained in the said ID information according to whether it is the value of the attribute contained in the attribute information determined to be relevant to ID information for every attribute value The first likelihood determined for each attribute value with respect to the ID is stored in the attribute likelihood storage means, and the first likelihood for each attribute value regarding the ID determined during the past fixed time Is calculated from the attribute likelihood storage means, and for each attribute value, the second likelihood is calculated by integrating the first likelihood for a certain period of time read from the attribute likelihood storage means, The mobile object attribute estimation according to appendix 6, wherein a process is performed for each ID by selecting an attribute value having the maximum second likelihood with respect to one ID and associating the ID with the selected attribute value. Method.

(Additional remark 10) ID information including ID of a moving body, the state information showing the state of the said moving body, the detection time of said ID and state information, the attribute of a moving body, and the state showing the state of the said moving body Information, and attribute information including the detection time of the attribute and state information are input to a computer, and the ID attribute information is selected by sequentially selecting ID information in the computer. Attribute information including a detection time belonging to a predetermined time zone based on the detection time included in the information, and determining that the state information is similar to the state information included in the selected ID information, Attribute information selection processing to be selected as attribute information related to the selected ID information, and for each attribute value included in each attribute information determined to be related to ID information, I included in the ID information A likelihood derivation process for obtaining a likelihood corresponding to the attribute of the moving object identified by the above, and a value of the attribute having the maximum likelihood for one ID is selected, and the ID and the value of the selected attribute are selected A moving object attribute estimation program for executing an attribute selection process for performing the process of associating with each ID.

(Supplementary note 11) Attribute likelihood storage means for storing the likelihood obtained for each attribute value for each ID, and the likelihood obtained for each attribute value for each ID, the likelihood was obtained When the ID information is newly input to the computer including the change model storage unit that stores the change model of the likelihood that defines how to correct with the passage of time from the time to the current time, In the likelihood derivation process, the likelihood obtained for each attribute value is read from the attribute likelihood storage means, the read likelihood is corrected according to the change model, and the corrected likelihood and the ID information are The mobile object attribute estimation program according to appendix 10, wherein the likelihood of the attribute value related to the ID included in the ID information is updated based on the attribute value included in each attribute information determined to be related.

(Supplementary Note 12) Based on the attribute value included in each attribute information determined to be related to the ID information in the likelihood derivation process, the computer detects for each attribute value the ID included in the ID information. The mobile object attribute estimation program according to appendix 10 or appendix 11, wherein the frequency is obtained and the detection frequency for each attribute value is set as the likelihood for each attribute value.

(Supplementary note 13) In a computer having attribute likelihood storage means for storing the likelihood obtained for each attribute value with respect to each ID, it is determined in the likelihood derivation process that the attribute value is related to the ID information. The first likelihood related to the ID included in the ID information is determined according to whether or not the attribute value is included in the attribute information. The likelihood is stored in the attribute likelihood storage means, and the first likelihood for each attribute value related to the ID, which is determined during the past fixed time, is read from the attribute likelihood storage means, and for each attribute value In addition, the second likelihood is calculated by integrating the first likelihood for a certain past time read from the attribute likelihood storage means, and the second likelihood is obtained for one ID in the attribute selection process. Select the attribute value with the highest degree and select the ID The process of associating the attribute value, the mobile attribute estimation program according to Note 10 for causing each ID.

(Additional remark 14) ID information including ID of a moving body, the state information showing the state of the said moving body, the detection time of said ID and state information, the attribute of a moving body, and the state showing the state of the said moving body Information and attribute information including the detection time of the attribute and status information are input, ID information is sequentially selected, and detection times belonging to a predetermined time zone based on the detection time included in the selected ID information An attribute information selection unit that selects attribute information that is determined to be similar to the state information included in the selected ID information, as attribute information related to the selected ID information. A likelihood deriving unit for obtaining a likelihood corresponding to the attribute of the moving object identified by the ID included in the ID information for each attribute value included in each attribute information determined to be related to the ID information; ID A mobile object attribute comprising: an attribute selection unit that selects a value of an attribute having the maximum likelihood and associates the ID with the value of the selected attribute for each ID. Estimation system.

(Supplementary Note 15) An attribute likelihood storage unit that stores the likelihood obtained for each attribute value with respect to each ID, and the likelihood obtained for each attribute value with respect to each ID. And a change model storage unit for storing a change model of likelihood that defines how to correct with the passage of time from time to the current time, and the likelihood derivation unit newly adds an ID to the attribute information selection unit When information is input, the likelihood calculated for each attribute value is read from the attribute likelihood storage unit, the read likelihood is corrected according to the change model, the corrected likelihood, and the ID information 15. The moving object attribute estimation system according to appendix 14, wherein the likelihood of the attribute value related to the ID included in the ID information is updated based on the attribute value included in each attribute information determined to be related to the ID.

(Additional remark 16) A likelihood derivation | leading-out part calculates | requires detection frequency for every value of each attribute regarding ID contained in the said ID information based on the value of the attribute contained in each attribute information determined to be relevant to ID information. The mobile object attribute estimation system according to appendix 14 or appendix 15, wherein the detection frequency for each attribute value is the likelihood for each attribute value.

(Additional remark 17) A likelihood derivation | leading-out part calculates | requires detection frequency for every value of each attribute regarding ID contained in the said ID information based on the value of the attribute contained in each attribute information determined to be relevant to ID information. The detection frequency for each attribute value is set as the likelihood for each attribute value, and the ID and the likelihood for each attribute value related to the ID are stored in the attribute likelihood storage unit to select attribute information. When the ID information is newly input to the unit, the likelihood obtained for each attribute value is read from the attribute likelihood storage unit, the read likelihood is corrected according to the change model, and the corrected likelihood And obtaining the detection frequency for each attribute value with respect to the ID included in the ID information based on the attribute value included in the attribute information determined to be related to the ID information. The moving object attribute estimation system according to supplementary note 15 to be updated.

(Supplementary Note 18) An attribute likelihood storage unit that stores the likelihood obtained for each attribute value with respect to each ID, and the likelihood deriving unit is attribute information determined to be related to the ID information for each attribute value The first likelihood related to the ID included in the ID information is determined according to whether or not the attribute value is included in the ID information, and the first likelihood determined for each attribute value regarding the ID is A first likelihood for each attribute value related to the ID, which is stored in the attribute likelihood storage unit and determined during the past certain period of time, is read from the attribute likelihood storage unit, and for each attribute value, the attribute likelihood The second likelihood is calculated by integrating the first likelihoods of the past fixed time read from the degree storage unit, and the attribute selection unit has the second likelihood maximum for one ID. The process of selecting the value of the attribute being selected and associating the ID with the value of the selected attribute Mobile attribute estimation system according to note 14 performed.

  This application claims the priority on the basis of the JP Patent application 2011-068551 for which it applied on March 25, 2011, and takes in those the indications of all here.

  Although the present invention has been described with reference to the embodiments, the present invention is not limited to the above-described embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

Industrial applicability

The present invention is suitably applied to a mobile object attribute estimation system that estimates attribute information of a mobile object given an ID.
For example, the present invention can be applied to a case where an employee number unique to each person working in an office or a factory is stored in association with physical characteristics such as gender, age, height, and clothes of each person. It is. As a result, when a third party attempts to pass the security gate by unauthorized use of an IC card with an employee number recorded, the physical characteristics of the unauthorized user differ from those of the true owner. Can be used for security purposes such as outputting alerts.
In addition, for example, the present invention can be applied to a case where a shopping cart with an ID is provided in a shopping center and the physical characteristics of a customer associated with the cart ID are estimated. As a result, it is possible to collect attributes such as the sex and age of the purchaser, and it can be used for marketing purposes such as measuring flow lines according to attributes such as sex and age.

DESCRIPTION OF SYMBOLS 1 ID information input part 2 Attribute information input part 3, 3a Mobile body attribute estimation part 4 Estimation result output part 31 Attribute candidate selection means 32, 32a Attribute likelihood calculation means 33 Attribute estimation means 34, 34a Attribute likelihood storage means 35 Attributes Model storage means

Claims (10)

ID of the moving object, state information indicating the state of the moving object, ID information including the ID and the detection time of the state information, attributes of the moving object, state information indicating the state of the moving object, Attribute information including the attribute and the detection time of the status information is input, the ID information is sequentially selected, and the attribute information includes the detection time belonging to a predetermined time zone based on the detection time included in the selected ID information. Attribute information selection means for selecting, as attribute information related to the selected ID information, attribute information determined that the state information is similar to the state information included in the selected ID information;
Likelihood derivation means for obtaining a likelihood corresponding to the attribute of the moving object identified by the ID included in the ID information, for each attribute value included in each attribute information determined to be related to the ID information;
An attribute selection unit that selects, for each ID, an attribute value having a maximum likelihood and associates the ID with the selected attribute value for each ID. Body attribute estimation system. Attribute likelihood storage means for storing the likelihood obtained for each attribute value for each ID;
Stores a likelihood change model that defines how the likelihood obtained for each attribute value for each ID is to be corrected as time elapses from the time the likelihood is obtained until the current time. And a change model storage means for
Likelihood derivation means
When the ID information is newly input to the attribute information selection means, the likelihood obtained for each attribute value is read from the attribute likelihood storage means, the read likelihood is corrected according to the change model, and after the correction The likelihood of the attribute value related to the ID included in the ID information is updated based on the likelihood of the ID and the attribute value included in each attribute information determined to be related to the ID information. Mobile object attribute estimation system. Likelihood derivation means
Based on the value of the attribute included in each attribute information determined to be related to the ID information, the detection frequency for each attribute value is obtained for the ID included in the ID information, and the detection frequency for each attribute value is calculated. The mobile object attribute estimation system according to claim 1, wherein the likelihood is set for each attribute value. Likelihood derivation means
Based on the value of the attribute included in each attribute information determined to be related to the ID information, the detection frequency for each attribute value is obtained for the ID included in the ID information, and the detection frequency for each attribute value is calculated. The likelihood for each attribute value is stored in the attribute likelihood storage means with the ID and the likelihood for each attribute value related to the ID,
When the ID information is newly input to the attribute information selection means, the likelihood obtained for each attribute value is read from the attribute likelihood storage means, the read likelihood is corrected according to the change model, and after the correction And obtaining the detection frequency for each attribute value with respect to the ID included in the ID information based on the likelihood of the ID and the attribute value included in each attribute information determined to be related to the ID information, The mobile object attribute estimation system according to claim 2, wherein the likelihood is updated. Attribute likelihood storage means for storing the likelihood obtained for each attribute value for each ID,
Likelihood derivation means
For each attribute value, a first likelihood related to the ID included in the ID information is determined according to whether the attribute value is included in the attribute information determined to be related to the ID information, Storing the first likelihood defined for each attribute value with respect to the ID in the attribute likelihood storage means;
Read from the attribute likelihood storage means a first likelihood for each attribute value related to the ID, determined during the past fixed time,
For each attribute value, the second likelihood is calculated by integrating the first likelihood for the past certain period of time read from the attribute likelihood storage means,
The attribute selection unit selects, for each ID, a value of an attribute having the maximum second likelihood, and performs a process of associating the ID with the value of the selected attribute for each ID. The moving body attribute estimation system described. ID of the moving object, state information indicating the state of the moving object, ID information including the ID and the detection time of the state information, attributes of the moving object, state information indicating the state of the moving object, Attribute information including the attribute and the detection time of the status information is input, the ID information is sequentially selected, and the attribute information includes the detection time belonging to a predetermined time zone based on the detection time included in the selected ID information. And selecting attribute information determined that the state information is similar to the state information included in the selected ID information as attribute information related to the selected ID information,
For each attribute value included in each attribute information determined to be related to the ID information, a likelihood corresponding to the attribute of the moving object identified by the ID included in the ID information is obtained.
A mobile object attribute estimation method characterized in that, for one ID, a value of an attribute having the maximum likelihood is selected, and the process of associating the ID with the value of the selected attribute is performed for each ID. Store the likelihood obtained for each attribute value for each ID,
A likelihood change model that prescribes how to modify the likelihood obtained for each attribute value for each ID with the passage of time from the time the likelihood is obtained to the current time is stored in advance. And
When the ID information is newly input, the likelihood obtained for each attribute value with respect to the ID included in the ID information is corrected according to the change model, and the corrected likelihood is related to the ID information. The mobile object attribute estimation method according to claim 6, wherein the likelihood of the attribute value relating to the ID included in the ID information is updated based on the attribute value included in each determined attribute information. Based on the attribute value included in each attribute information determined to be related to the ID information, with respect to the ID included in the ID information, the detection frequency is obtained for each attribute value, and the detection frequency for each attribute value is The mobile object attribute estimation method according to claim 6 or 7, wherein the likelihood is set for each attribute value. For each attribute value, a first likelihood related to the ID included in the ID information is determined according to whether the attribute value is included in the attribute information determined to be related to the ID information, Storing the first likelihood determined for each attribute value with respect to the ID in the attribute likelihood storage means;
Read from the attribute likelihood storage means a first likelihood for each attribute value related to the ID, determined during the past fixed time,
For each attribute value, the second likelihood is calculated by integrating the first likelihood for the past certain period of time read from the attribute likelihood storage means,
The mobile object attribute according to claim 6, wherein, for one ID, the value of the attribute having the maximum second likelihood is selected, and the process of associating the ID with the value of the selected attribute is performed for each ID. Estimation method. ID of the moving object, state information indicating the state of the moving object, ID information including the ID and the detection time of the state information, attributes of the moving object, state information indicating the state of the moving object, A moving object attribute estimation program installed in a computer to which attribute information including attribute and state information detection time is input,
In the computer,
ID information is sequentially selected, and is attribute information including a detection time belonging to a predetermined time zone based on a detection time included in the selected ID information, and the state information is included in the selected ID information Attribute information selection processing for selecting attribute information determined to be similar as attribute information related to the selected ID information;
A likelihood deriving process for obtaining a likelihood corresponding to the attribute of the moving object identified by the ID included in the ID information, for each attribute value included in each attribute information determined to be related to the ID information; and
Mobile attribute estimation for executing attribute selection processing for each ID that selects the value of the attribute having the maximum likelihood for one ID and associates the ID with the value of the selected attribute program.

Images