JP2023090067A - Action recognition system - Google Patents

Abstract

To provide an action recognition system that accurately detects and classifies action information for an individual person in a dwelling space where multiple persons live.SOLUTION: An action recognition system 1000 includes: a sensor group containing a plurality of presence sensors 100 for sensing a person and a plurality of person recognition sensors 200 which senses a person and is capable of identifying a person; and an action estimation device 300 that recognizes an action of an identified person on the basis of sensing information received from the sensor group. The action estimation device 300 includes: a storage unit 350 that stores movement information related to a route through which a person moves from one region where at least one sensor in the sensor group is installed to another region where at least another sensor in the sensor group is installed; and a calculation unit 340 that performs time series analysis of information sensed by the sensor group, creates candidate information for a movement route through which the person to be recognized can move on the basis of a result of the analysis and the movement information, and estimates the action of the person identified by the plurality of human recognition sensors on the basis of the candidate information for the movement route and the information sensed by the plurality of human recognition sensors.SELECTED DRAWING: Figure 1

Description

The present invention is a technology related to a system that recognizes the actions of people living in a living environment by combining detection information from sensors such as home appliances and human sensors installed in the living environment and sensors that detect and identify people. be.

Japanese Unexamined Patent Application Publication No. 2002-100002 discloses a conventional technology that utilizes operation information of home appliances and sensor information in order to control equipment according to the state of a user living in a living environment and to grasp the activity state of the user. Patent Document 1 describes a remote care system for collective housing that implements and operates remote care using white goods used for general households. A detection unit that detects the behavior of existing residents is provided, and based on the detection information of the detection unit provided in the home appliance, home appliances installed in a plurality of living spaces are remotely controlled and remotely monitored based on the detection information. Techniques are disclosed.

JP 2017-011698 A

In a system for confirming the safety of residents from a remote location, since multiple residents share a living space in a normal household, the behavior information for determining the safety of residents includes the individual behavior of the residents. is required to detect and identify On the other hand, in order to individualize the behavioral information of residents, it is necessary to install multiple sensors such as images and sounds in various places in the living space. create challenges.

For example, in a two-family house, if a safety confirmation system that identifies behavioral information based only on motion sensors and home appliance operation information is introduced, after multiple residents share their behavior in shared spaces such as the dining room and living room, one person stays in the common space and the remaining people move to another living space, it becomes impossible to distinguish and identify which resident's behavior the value derived from the detection information of the human sensor. For this reason, if a safety confirmation system that identifies behavioral information based only on motion sensors and household appliance operation information is introduced in a living space where two or more people live in a shared living environment, individual behavior of residents cannot be detected. cannot be detected and identified. On the other hand, in order to solve the above problems, when multiple sensors such as images and sounds are installed in the living environment, problems arise in terms of ensuring privacy, introduction costs, operating costs, etc. .

The present invention is made to solve the above-mentioned problems, and even if the number of human recognition sensors installed is limited in consideration of privacy, it is possible to detect the presence of individuals in a living space where multiple people live. The object is to provide an action recognition system that can detect and classify action information with high accuracy.

In order to solve the above problems, the action recognition system of the present invention provides a sensor group including a plurality of human sensors that detect people, and a plurality of human recognition sensors that detect and specify the people. and a behavior estimation device for recognizing the behavior of the person identified by the human recognition sensor based on the detection information received from the sensor group, wherein the behavior estimation device is provided with at least one of the sensors. A storage unit that stores at least movement information related to a route taken by a person from an area to another area where at least one other sensor group is installed; , movement information, and moving route candidate information along which a person to be recognized can move is created, and a plurality of persons are recognized based on the moving route candidate information and information detected by a plurality of person recognition sensors. and a computing unit that recognizes the behavior of the person identified by the sensor.

According to the present invention, by connecting a plurality of human detection sensors and human recognition sensors provided in the living environment with a behavior estimation device provided in the living environment, the behavior of each individual resident living in the target living environment can be detected. The estimating unit that estimates makes it possible to discriminate the behavior of the inhabitants. Furthermore, in order to create human behavior information using human sensors and detection information from home appliances, it is possible to reduce the number of sensors that recognize people. of action recognition system can be provided.
Further features related to the present invention will become apparent from the description of the specification and the accompanying drawings. Further, problems, configurations and effects other than those described above will be clarified by the following description of the embodiments.

Schematic diagram showing the configuration of the action recognition system of the present invention. The figure showing the installation example which installs several sensors in living environment. FIG. 4 is a diagram for explaining various types of information used in the action recognition system of the present invention; FIG. 2 is a flow diagram showing a procedure for processing information detected by a sensor in the action recognition system of the present invention; FIG. 4 is a diagram for explaining action detection information and personal action detection information used in the action recognition system of the present invention; FIG. 4 is a flow diagram showing a process of specifying personal movement information based on information detected by a sensor in the action recognition system of the present invention; FIG. 4 is a diagram for explaining moving route candidate information;

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 shows an outline of the configuration of the action recognition system of the present invention. The action recognition system 1000 of the present invention includes a plurality of human sensors 100 (inexpensive, simple human sensors and home appliances) that detect people, and a plurality of human recognition sensors 200 (image, voice, ToF) that detect and identify people. , an acceleration sensor, etc.), and a behavior estimation device 300 that identifies and recognizes a person and his/her behavior based on the detection information of the sensor group. In addition, in this embodiment, the behavior recognition system 1000 further includes a remote management unit 400 that can receive information from the behavior estimation device 300 and remotely confirm the safety and condition of the residents to be detected. there is

The plurality of human sensors 100 includes a power supply unit 110 for activating the human sensor 100, a communication unit 120 for transmitting detection information of the human sensor 100 and a unique ID of the human sensor 100 to the behavior estimation device 300, a human It has a storage unit 130 for storing unique information of the sensor 100 , a detection unit 140 for detecting a person, and a control unit 150 for controlling the operation of the human sensor 100 .

In addition, the plurality of human recognition sensors 200 also include a power supply unit 210 for activating the human recognition sensor 200, a communication unit 220 for transmitting detection information of the human recognition sensor 200 and the unique ID of the sensor to the behavior estimation device 300, a person It has a storage unit 230 for storing unique information of the recognition sensor 200 , a detection unit 240 for detecting and specifying a person, and a control unit 250 for controlling the operation of the person recognition sensor 200 .

Further, action estimation device 300 has communication section 310 , external communication section 320 , control section 330 , calculation section 340 , storage section 350 , timer 360 , and external power source 370 . Note that the behavior estimation device 300 is configured by a server, a PC, or the like.

The communication unit 310 receives detection information and sensor unique IDs transmitted from the plurality of human sensors 100 and the plurality of human recognition sensors 200 . The external communication unit 320 receives and transmits detection information and analysis information to/from an external network other than the sensor network constructed by the plurality of human sensors 100 and the plurality of human recognition sensors 200 and the action estimation device 300 . The control unit 330 analyzes information from sensor networks, external networks, etc., and controls devices based on the analysis results. Based on the input information from the external communication unit 320, the detection information of the control unit 330, the plurality of human sensors 100 and the plurality of human recognition sensors 200, the spatial movement information described later, the residence information, etc. behavior and determine the safety of residents. The storage unit 350 stores movement route information obtained by determining the behavior of each individual resident determined by the calculation unit 340 . The timer 360 evaluates the time when detection information from the plurality of human sensors 100 and the plurality of human recognition sensors 200 is received. In addition, the remote management unit 400 can receive and transmit information regarding the safety of residents analyzed by the behavior estimation device 300 via the external communication unit 320 .

By constructing the action recognition system with the configuration described above, the detection information of the inhabitants detected by the plurality of human sensors 100 and the plurality of human recognition sensors 200 installed in the living environment and the unique IDs of the sensors can be detected by the action estimation device. 300. The computing unit 340 of the behavior estimation device 300 that has received the detection information and the sensor ID manages the reception time using the timer 360, and if the sensor that transmitted the detection information is the human sensor 100, the sensor ID and the sensor installation From the information, the room in which the sensor reacted (where the action occurred) and the time can be determined by the calculation unit 340 and stored in the storage unit 350 as action detection information.

Similarly, the calculation unit 340 recognizes the type of sensor from the unique ID of the sensor, and if the sensor that has transmitted the detection information is the human recognition sensor 200, the calculation unit 340 compares it with the resident information to identify the detection target. Residents can be specified and stored in the storage unit 350 as individual detection information.

The computing unit 340 identifies the movement information of the individual by analyzing the movement route of the individual using the individual detection information and behavior detection information stored in the storage unit 350, the residence information and the spatial movement information. The storage unit 350 can store the movement route of the identified individual. By monitoring this saved movement route, if detection information that diverges from the individual's daily movement information is detected, it can be determined that some abnormality has occurred in the individual to be detected, and the abnormality information can be communicated externally. It can be transmitted to the remote management unit 400 through the unit 320 .

FIG. 2 is a diagram showing an overview when the action recognition system 1000 configured as shown in FIG. 1 is introduced into a living environment including a living space 2000. As shown in FIG. The human sensor 100 includes a human sensor that detects the presence or absence of a person, a door open/close sensor that detects opening and closing of a door, an illuminance sensor that detects the brightness of a room, and the like. These sensors simply detect the presence or absence of a person, and do not specify personal information. Therefore, there is no restriction on the installation location as long as it is within the living space 2000 so as not to violate the privacy of the residents living in the detection target. In FIG. 2, the human sensor 100 is installed in the living room L and the dining room D. As shown in FIG.

On the other hand, the human recognition sensor 200 includes a sensor that acquires personal information such as images, voices, ToF, and the like to identify an individual, a sensor that measures the acceleration of opening and closing a door, and the like. Since these sensors detect information that includes the privacy of residents, it is necessary to consider privacy when installing them. For example, it is preferable to install it in a place with relatively high commonality, such as an entrance or a hallway. In FIG. 2, the human recognition sensor 200 is installed at the entrance E or the like.

Although FIG. 2 illustrates the robot cleaner 500 as a representative home appliance, any home appliance that handles image information, such as a robot cleaner, can be used as the human recognition sensor 200 . In addition, when sensors and operation information of refrigerators, washing machines, air conditioners, etc. are used as sensor detection information, these home appliances can be utilized as the human sensor 100 . Behavior estimation device 300 and remote management unit 400 are not shown in the figure.

FIG. 2 is an example showing an outline of a method for installing the human sensor 100 and the human recognition sensor 200 in the living space 2000 according to the present invention. It is not limited to FIG.

Procedures for detecting and recognizing behavior information of an individual resident to be detected in the behavior recognition system 1000 configured as described above will be described with reference to FIGS. 3 to 7. FIG.

First, various information used in the action recognition system 1000 according to the present invention will be described. A plurality of human sensors 100 and a plurality of human recognition sensors 200 are installed in a living space 2000 to be detected, and basic information of residents, houses, sensors, etc. are registered in the storage unit 350 of the behavior estimation device 300 in advance. Information to be registered includes, for example, resident information, room information, sensor installation information, residence information, spatial movement information, and the like. An example of such information will be described with reference to FIG.

As shown in FIG. 3, the resident information is constructed from the name of the resident, an individual ID for distinguishing between residents, and feature data relating to the feature amount of the resident for specifying an individual by the human recognition sensor 200 . The feature amount for identifying an individual is, for example, height, physique, posture, walking style, dominant hand, image, voice, and other information that can be detected by various sensors. Room information is constructed from a room name and a room ID. The sensor installation information is information for managing installation information on installation of the human sensor 100 and the human recognition sensor 200, and includes the sensor ID, the room ID where the sensor is installed, and the sensor type (0: human recognition sensor, 1: non-human recognition sensor).

The residence information is a database of information about spaces in which a detection target resident is or is absent on a daily basis. (Weekend), but the classification is not limited to this embodiment), time period (for example, in this embodiment, 0: morning (4:00 to 10:00), 1: noon (10:00) 16:00 to 16:00), 2: Night (16:00 to 22:00), 3: Midnight (22:00 to 4:00), but the division method is not limited to this embodiment), space utilization status (0 : Present/1: Absent/-: Unknown, but case classification is not limited to this embodiment).

Spatial movement information (movement information) is information indicating a list of ON and OFF sensor ID strings of a plurality of sensors that react when a person moves from one room to another. The moving route can be estimated by comparing the detection information of the sensor with

The basic information as described above is registered in the storage unit 350 of the action estimation device 300, and the action recognition system 1000 is operated. FIG. 4 is a flow diagram showing the processing performed by the action recognition system 1000 after receiving a detection signal from the human sensor 100 or the human recognition sensor 200 until it is stored in the storage unit 350 as action detection information or individual detection information. .

When the action recognition system 1000 is activated, first, the process proceeds to operation step S100. After moving to operation step S100, the action recognition system 1000 determines whether any of the human sensor 100 and the human recognition sensor 200 has detected the action of the resident. If either the human sensor 100 or the human recognition sensor 200 detects the action of the resident and the sensor responds, the operation step shifts to S110, and shifts to the process of sorting/identifying the sensor information.

If the sensor does not respond, the operation step proceeds to S200, and it is determined whether or not the operating time (T) from the start of detection of the response of the sensor is equal to or longer than a predetermined elapsed time (Tan). Note that the predetermined elapsed time (Tan) can be set as, for example, 6 hours. If the operating time (T)≧elapsed time (Tan) in operation step S200, the operation step is moved to S210 in FIG. 6 to proceed to the process of estimating a moving person. This processing will be described later. If the operating time (T)<elapsed time (Tan), the operation step returns to S100 after adding the elapsed time (T+ΔT).

After moving to operation step S110, the action recognition system 1000 selects sensor information and moves the operation step to a process of accumulating detection data according to the sensor information. After reading the sensor installation information in operation step S110, the action recognition system 1000 proceeds to operation step S120.

In operation step S120, the action recognition system 1000 compares the detection sensor ID obtained from the detection information received from the human sensor 100 or the human recognition sensor 200 with the sensor installation information, and the sensor that transmitted the detection information is the human recognition sensor. 200 or not (human sensor 100). In operation step S120, if the sensor transmitting the detection information is not the human recognition sensor 200, that is, if it is determined to be the human sensor 100, the action recognition system 1000 shifts the operation step to S121, The detection information is created, the operation step is shifted to S<b>160 , and the action detection information is stored in the storage unit 350 of the action estimation device 300 .

Here, the behavior detection information is a data string composed of room IDs and detection times as shown in FIG. That is, the behavior detection information does not have information that can identify a person, but has information about the place and time when the sensor reacts. When it is determined in operation step S120 that the sensor transmitting the detection information is the human recognition sensor 200, the action recognition system 1000 shifts the operation step to S130.

The action recognition system 1000, which has shifted the process to operation step S130, reads the resident information from the storage unit 350 of the action estimation device 300, and then shifts the operation step to S140.

After moving to operation step S140, the action recognition system 1000 compares the detection information received from the human recognition sensor 200 with the individual feature data registered in the resident information. If there is no personal feature data that matches the detection information, the action recognition system shifts the operation step to S121 and stores it in the storage unit 350 as action detection information. If there is personal feature data that matches the detection information, the action recognition system 1000 moves the operation step to S150.

The action recognition system 1000 that has moved the action step to S150 determines that the detection information is personal action information, and creates individual detection information and action detection information. As shown in FIG. 5, the individual detection information is configured as a data string including a room ID, individual ID, and detection time. In other words, the individual detection information includes information related to the specified individual in addition to the action detection information.

After creating the individual detection information and the action detection information in operation step S150, or creating the action detection information in operation step S121, the action recognition system 1000 shifts the operation step to S160. Then, after storing the behavior detection information and the individual detection information in the storage unit 350 of the behavior estimation device 300, the operation step is shifted to S200 again. moves to S210, and moves to a moving person estimation process, which is a process of estimating the movement information shown in FIG. 6 and identifying an individual held by the movement information.

In operation step S210, the action recognition system 1000 reads the spatial movement information from the storage unit 350, and then compares the action detection information divided by the analysis time width w with the spatial movement information to create inter-room movement information. . Here, inter-room movement information is information relating to a room (a room in which a sensor that has reacted is installed) to which a person has moved during a certain period of time included in the behavior detection information. Specifically, taking the behavior detection information in FIG. 5 as an example, a person moves from room R1 to room R2 between time t1 and time t2. Referring to this and the spatial movement information in FIG. 3, it can be seen that the movement from room R1 to room R2 has two routes of S1→S2 and S1→S6→S5→S2 as reaction sensor arrays. That is, inter-room movement information is information regarding these two routes. This inter-room movement information is calculated along with the detection time included in the behavior detection information.

The created inter-room movement information is combined to create movement route candidate information, and the operation step is moved to S220. Note that, as shown in FIG. 7, the moving route candidate information is information connecting the information on moving between rooms occurring within the elapsed time (Tan), and each moving route candidate information includes the ID of the room in which the movement was made and the time when the movement was completed. This is information arranged in chronological order by a set of .

After moving to operation step S220, the action recognition system reads the residence information from the storage unit 350, and uses presence/absence/unknown information in each room ID for each individual ID included in the individual detection information and the residence information, A weighting factor (w) for the room ID and the time indicated by each moving route candidate information is determined in operation steps S235 to S265.

Here, the weighting factor (w) is an index representing the probability of whether the moving route candidate is the moving route candidate followed by a person identified by the personal ID, and is calculated by the following flow.

After moving to operation step S230, the action recognition system 1000 determines whether or not individual detection information including a room ID and a detection time matching the room ID and inter-room movement time shown in the movement route candidate information exists. do. Here, the moving time between rooms refers to the time t1-t2 for moving from R5 to R2 in #1 of the moving route candidate information in FIG. 7, for example. This is compared with the individual detection information, and it is determined whether or not there is detection information indicating that the person was in room R2 during time t1-t2. This processing is performed for all route candidates.

If it is determined in S230 that there is individual detection information that matches the moving route candidate information, the action recognition system 1000 proceeds to operation step S235, sets the weighting factor to +1, and proceeds to operation step S270. If no matching individual detection information exists in S230, the action recognition system proceeds to operation step S240.

When the action recognition system 1000 proceeds to operation step S240, information that matches the room ID and the inter-room travel time indicated in the movement route candidate information exists, and if there is a room ID that matches "in the room" in the residence information. determine whether or not to If there is a room ID that matches the presence in the room in the residence information DB, the action recognition system 1000 moves the operation step to S245, sets the weighting factor to +1, and moves the operation step to S270. For example, in #1 of the moving route candidate information in FIG. 7, room R2 at time t2 is listed as a candidate. Referring to the residence information in FIG. 3, for example, if this time t2 is Day: WD and TimeZone: 2, the person identified by personal ID: P1 corresponds. is weighted by +1. If there is no room ID that matches the presence in the room in the residence information in operation step S240, the action recognition system proceeds to operation step S250.

After moving to operation step S250, the action recognition system 1000 determines whether there is a room ID whose information matching the room ID and the inter-room moving time shown in the moving route candidate information matches "absent" in the residence information DB. determine whether or not If there is a room ID that matches the absence in the residence information DB, the action recognition system 1000 moves the operation step to S255, sets the weighting factor to -1, and moves the operation step to S270. For example, in the residential information of FIG. 3, if there is sensor detection information for room ID: R1 in Day: WD, TimeZone: 1, the time when the person specified by personal ID: P1 is "absent" This person is a belt, and therefore the possibility of this person as a route candidate is low, so a weight of -1 is given. If there is no room ID that matches the absence in the residence information DB, the action recognition system 1000 shifts the operation step to S265, sets the weighting factor to 0, and shifts the operation step to S270.

After proceeding to operation step S270, the action recognition system 1000 calculates a room occupancy score using the room occupancy time and the weighting factor in each room ID of each moving route candidate, and moves the operation step to S280. Note that the room occupancy score S _i is obtained by multiplying the weighting factor w (RID _k ) for each room by the time (T _k+1 −T _k ) spent in the room, to the measurement end time.

The action recognition system 1000, which has shifted the operation step to S280, determines whether or not calculation has been performed for all movement route candidates, and when the calculation of all room presence scores has been completed, shifts the operation step to S290, If calculation of all room occupancy scores has not been completed, the operation step proceeds to S220.

The action recognition system 1000, which has shifted the operation step to S290, determines the moving route candidate with the highest room presence score among the results of the weighting calculation for each moving route candidate as the moving route of the individual for the individual ID to be calculated. Then, the operation step moves to S300. The action recognition system 1000 having shifted the operation step to S300 determines whether or not the moving routes of all the residents of the living environment to be detected have been determined, and when the moving routes of all the residents have been determined, shifts the operating step to S310. do.

If the moving routes of all residents have not been determined, the operation step proceeds to S220. The action recognition system 1000, which has shifted the operation step to S310, saves the movement information of the individual to be detected in the storage unit 350 of the action estimation device 300, resets the operating time (T) to 0, and then shifts the operation step to S100. do.

According to the embodiments of the present invention described above, the following effects are obtained.
(1) A behavior recognition system according to an embodiment of the present invention includes a group of sensors including a plurality of human sensors that detect humans, and a plurality of human recognition sensors that detect humans and identify the humans. and a behavior estimation device for recognizing the behavior of the person identified by the human recognition sensor based on the detection information received from the sensor group, wherein the behavior estimation device is located in an area where at least one of the sensors is installed. a storage unit that stores at least movement information related to a route taken by a person from a sensor group to another area where at least one other sensor group is installed; a time series analysis of information detected by the sensor group; Based on the movement information, moving route candidate information along which the person to be recognized can move is created, and based on the moving route candidate information and information detected by the plurality of human recognition sensors, a plurality of human recognition sensors. and a computing unit that recognizes the behavior of the identified person.

With the above configuration, an estimation unit that estimates the behavior of each resident living in the target living environment by connecting a plurality of sensors and motion sensors provided in the living environment with a behavior estimating device provided in the living environment. makes it possible to determine the behavior of residents. Furthermore, in order to create human behavior information using human sensors and detection information from home appliances, it is possible to reduce the number of sensors that recognize people. of action recognition system can be provided.

(2) The behavior estimation device further has an external communication unit that transmits information about human behavior outside the area where the sensor group is installed via the network. Thereby, when it is determined that there is an abnormality in the movement information of the resident, it is possible to report the abnormality of the resident to an external monitoring device.

(3) The storage unit further stores person information related to feature amounts for identifying a person to be detected, and the person recognition sensor compares the detection content with the person information to identify the person do. This makes it possible to use information that can be detected by various sensors such as height, physique, posture, walking style, handedness, image and voice, so that an individual can be identified from various angles, and identification accuracy is improved.

(4) The storage unit further stores the behavior record of the person to be detected in the area where the sensor group is installed. Human behavior is recognized based on the candidate information and, in addition, the behavior record. As a result, even when a person cannot be identified based on the feature amount related to the person, the behavior of the person can be recognized based on the behavior records stored in advance (for example, which room is whose room, etc.). becomes possible.

(5) The computing unit refers to the information detected by the plurality of human recognition sensors to recognize the behavior of the person before referring to the behavior record. As a result, since the processing to be performed is prioritized, it is expected that the load of arithmetic processing will be reduced and the processing speed will be increased.

(6) When a human action is recognized by referring to information detected by a plurality of human recognition sensors, the computing unit assigns a positive weight, does not refer to the action record, and detects by a plurality of human recognition sensors. When the person's action cannot be recognized by referring to the obtained information, the person's action is recognized by referring to the action record without weighting. As a result, human behavior can be quantitatively recognized based on the detected information, and the recognition accuracy is improved.

(7) The calculation unit refers to the action record and assigns a positive weight when the human action is recognized, and assigns a negative weight when the human action is not recognized. This makes it possible to perform behavior recognition quantitatively in the same manner as in (6), and by performing negative weighting, it becomes possible to eliminate information due to erroneous sensor detection.

It should be noted that the present invention is not limited to the above embodiments, and various modifications are possible. For example, the above embodiments have been described in detail in order to facilitate understanding of the present invention, and the present invention is not necessarily limited to aspects having all the described configurations. Also, part of the configuration of one embodiment can be replaced with the configuration of another embodiment. It is also possible to add the configuration of another embodiment to the configuration of one embodiment. Moreover, it is possible to delete a part of the configuration of each embodiment, or to add or replace another configuration.

100 human sensor, 200 human recognition sensor, 300 action estimation device, 320 external communication unit, 340 calculation unit, 350 storage unit

Claims (7)

An action recognition system for recognizing human action,
a sensor group including a plurality of human sensors that detect a person and a plurality of human recognition sensors that detect and specify the person;
a behavior estimation device that recognizes the behavior of the person identified by the person recognition sensor based on the detection information received from the sensor group;
The action estimation device is
a storage unit that stores at least movement information related to a route along which a person moves from an area in which at least one of the sensor groups is installed to another area in which at least one other of the sensor groups is installed;
time-series analysis of the information detected by the sensor group, and based on the result of the analysis and the movement information, moving route candidate information along which the person to be recognized can move is created, and the moving route candidate information and , and a calculation unit that recognizes the behavior of the person identified by the plurality of human recognition sensors based on the information detected by the plurality of human recognition sensors,
An action recognition system characterized by: The action recognition system according to claim 1,
The behavior estimation device further has an external communication unit that transmits information about the behavior of the person outside the area where the sensor group is installed via a network.
An action recognition system characterized by: The action recognition system according to claim 1,
The storage unit further stores person information related to a feature amount for specifying a person to be detected,
The person recognition sensor identifies the person by comparing the detection content with the person information.
An action recognition system characterized by: The action recognition system according to claim 1,
The storage unit further stores an action record of the person to be detected within the area where the sensor group is installed,
The computing unit recognizes the behavior of the person based on the behavior record in addition to the information detected by the plurality of person recognition sensors and the moving route candidate information.
An action recognition system characterized by: The action recognition system according to claim 4,
The computing unit refers to information detected by the plurality of human recognition sensors to recognize the behavior of the person before referencing the behavior record.
An action recognition system characterized by: The action recognition system according to claim 5,
The computing unit refers to the information detected by the plurality of human recognition sensors and, when the behavior of the person can be recognized, performs a positive weighting, does not refer to the behavior record, and does not refer to the behavior record. If the action of the person cannot be recognized by referring to the information detected by, the action record of the person is recognized without weighting by referring to the action record.
An action recognition system characterized by: The action recognition system according to claim 6,
The calculation unit refers to the action record and performs positive weighting when the person's action can be recognized, and negative weighting when the person's action cannot be recognized.
An action recognition system characterized by:

Images