JP6386810B2 - Group emotion estimation device, group emotion estimation method, and group emotion estimation program - Google Patents

Description

  The present invention relates to the field of estimating an emotional state of a group from body measurement data of a plurality of persons.

  Regarding emotion estimation from a biometric sensor, a method is known that uses heartbeat data to estimate arousal level (arousal, sedation) and emotional value (pleasant, neutral, unpleasant) expressing emotions (Non-patent Document 1). ). According to this, low arousal / neutral, high arousal / pleasant, and high arousal / discomfort can be estimated to some extent by a combination of heart rate (HR) and high frequency component (HF) of heart rate variability, but there is a limit to accuracy. There is.

  In addition, a method for predicting more emotions using not only heartbeats but also brain waves and skin conductance levels (sweating) is known (Non-Patent Document 2). Although heartbeat, brain waves, and sweating can be used to estimate emotions such as surprise, anger, enjoyment, fear, sadness, and neutrality to some extent, accuracy is still limited.

Satoshi Toyosawa, Takashi Kawai, “Video shortening method using viewer's heartbeat activity and its evaluation”, Journal of the Institute of Image Information and Television Engineers, 63 (1), 2009, pp86-94 Junya Otani, Hiroaki Kudo, Tetsuya Matsumoto, Yoshinori Takeuchi, Noboru Onishi, “Emotion estimation using multiple biological signals during video viewing”, IEICE Technical Report, IMQ2012-39, 2013, pp.29-34

  Conventionally, emotion estimation using personal biometric information is possible to some extent, but accuracy is limited. In view of the problems of the prior art, an object of the present invention is to estimate a group's emotions with high accuracy by simultaneously acquiring and using the biological information of the group.

  Therefore, the present invention estimates group emotion information based on feature information extracted from biological information received from a plurality of biological sensors.

As an aspect of the collective emotion estimation apparatus of the present invention, a collective emotion estimation apparatus for estimating collective emotion information from biometric information received from a plurality of individual biometric sensors, the biometric information received from a plurality of individual biometric sensors. input means for inputting a pre-processing means for outputting the feature data of the feature pattern ID associated with the biometric information corresponding to the characteristic information of the biometric information, as the estimated condition of the emotion estimation, this particular time and the particular time An estimation condition input means for inputting a decreasing function based on time information consisting of a time interval from and a decreasing function based on position information consisting of a specific position and a distance from the specific position, and input in the estimation condition input means It was given to a single said feature pattern ID or sets of the feature pattern ID of the feature data selected according to the estimated condition The emotion corresponding to the element number of the emotion estimation vector having the maximum load sum for the plurality of individuals of the emotion estimation vector that is a probability vector indicating an emotion estimation candidate consisting of a group of emotion pattern IDs corresponding to emotion content Emotion estimation means for outputting the emotion content of the pattern ID as group emotion information, output condition input means for inputting an output condition, and output means for outputting the group emotion information that matches the input output condition The preprocessing means adds time information and position information at the time when the feature information is extracted to the feature data, and the emotion estimation means is the time input by the estimation condition input means. by decreasing function based on decreasing function or the position information based on the information, before the time information and the positional information in the pre-processing means is added Weighting the emotion estimation vector based on the feature data.

A group emotion estimation method according to the present invention is a group emotion estimation method executed by a group emotion estimation apparatus that estimates group emotion information from biological information received from a plurality of individual biometric sensors. An input step for inputting the received biometric information, a preprocessing step for outputting feature data corresponding to the feature pattern ID corresponding to the feature information of the biometric information, and a specific time as an estimation condition for emotion estimation And an estimation condition input step for inputting a decreasing function based on time information consisting of a time interval from this specific time, and a decreasing function based on position information consisting of a specific position and a distance from this specific position , and this estimation condition input single said feature pattern ID or the feature pattern of the feature data selected according to the estimated conditions input in step Emotion estimation vector that is a probability vector indicating an emotion estimation candidate consisting of a group of emotion pattern IDs corresponding to the emotion content given to the group consisting of the IDs, and that the sum of the weights of the plurality of individuals is the maximum value. The emotion estimation step for outputting the emotion content of the emotion pattern ID corresponding to the element number of the vector as group emotion information, the output condition input step for inputting an output condition, and the input condition that matches the input output condition An output step of outputting emotion information of a group, wherein the preprocessing step adds time information and position information at the time when the feature information is extracted to the feature data, and the emotion estimation step includes the step of by decreasing function based on decreasing function or the position information based on the time information input in the estimation condition input step, the front The time information and the positional information in the physical step is weighted to the emotion estimation vector based on the added the feature data.

  Note that the present invention may be in the form of a program that causes a computer to function as each unit of the apparatus or a program that causes a computer to execute the steps of the method.

  According to the present invention, the emotional state of a group can be estimated with high accuracy.

The block block diagram of the group emotion estimation apparatus in embodiment of this invention. The flowchart of the group emotion estimation process in the embodiment. The flowchart of the data input process in the embodiment. The flowchart of the pre-processing process in the embodiment. The flowchart of the emotion estimation process in the embodiment. The flowchart of the estimation condition input process in the embodiment. The flowchart of the output process in the embodiment. The flowchart of the output condition input process in the embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the present invention is not limited to these embodiments.

[Overview]
The group emotion estimation apparatus 1 of the present embodiment shown in FIG. 1 weights an emotion estimation vector according to an established distribution related to estimation of group emotion information based on feature information extracted from biological information received from a plurality of individual biological sensors. To estimate the group's emotional information.

  Technical terms described in the description of this embodiment will be described.

  “Emotion”: Human emotion. ID is assigned to emotion information such as “anger”, “sadness”, “joy” and the like.

  “Characteristic data”: data obtained by adding a feature pattern ID to feature information extracted from sensor data (biometric information) received from a biosensor attached to an individual. More information can be included.

  “Characteristic pattern ID”: ID defined according to the type of abnormal value when an abnormal value is detected in the sensor data. Even if no abnormal value is detected, the feature pattern ID can be defined for the normality itself.

  “Time”: Time information when feature information of sensor data is extracted. It is given as a time range or a specific time. When video data or audio data is viewed asynchronously, it may be expressed not by an absolute time but by a relative elapsed time from the start of the data.

  “Place”: Position information of the biosensor when the feature information of the sensor data is extracted. It is not limited to a physical position, but may be virtual position information (a group of people watching a certain program, a group of people in a specific chat room, etc.).

[Device configuration]
The group emotion estimation apparatus 1 includes a data input unit 2, a preprocessing unit 3, an estimation condition input unit 4, an emotion estimation unit 5, an output condition input unit 6, and an output unit 7, as shown in FIG.

  The pre-processing unit 3 is not necessary when the sensor data provided from the data input unit 2 has already undergone feature extraction. The estimation condition input unit 4 is also unnecessary when it is not necessary to set the estimation conditions in the emotion estimation unit 5 at any time. The output condition input unit 6 is also unnecessary if it is not necessary to set display conditions in the output unit 7 as needed.

  The functional units 2 to 7 are realized by computer hardware resources. That is, the collective emotion estimation apparatus 1 includes hardware resources relating to a computer such as at least a computing device (CPU), a storage device (memory, a hard disk device, etc.), and a communication interface. These hardware resources cooperate with software resources (OS, applications, etc.) to implement the function units 2-7. Moreover, you may make it each implement | achieve the function parts 2-7 in each computer.

  The data input unit 2 determines whether the sensor data received from the biosensors worn by a plurality of users is data whose features have been extracted. When the input data is data before feature extraction, the data is supplied to the preprocessing unit 3. On the other hand, when the input data is data whose features have been extracted, the data is provided to the emotion estimation unit 5.

  The preprocessing unit 3 extracts the feature information from the sensor data determined not to have been subjected to the feature extraction provided from the data input unit 2, and the feature data obtained by adding the feature information to the data is the emotion estimation unit 5. To serve. More specifically, feature information is extracted from sensor data before feature extraction provided from the data input unit 2, and a feature pattern ID that is an identifier of the feature information is associated with a user ID and a detection item, and the data is included in the data. The characteristic data is configured by providing the data, and the characteristic data is provided to the emotion estimation unit 5.

  The estimation condition input unit 4 inputs an estimation condition for estimating the emotion information to the emotion estimation unit 5.

  The emotion estimation unit 5 estimates the emotion information of the group including the user based on the feature information of the feature data provided from the data input unit 2 or the preprocessing unit 3, and provides this emotion information to the output unit 7. More specifically, the emotion corresponding to the feature information of the feature data provided from the data input unit 2 or the preprocessing unit 3 by referring to the emotion estimation candidate table in which the emotion estimation candidates are defined in association with the feature information. An emotion estimation vector composed of probability vectors indicating estimation candidates is acquired. Then, the emotion information of the group including the user is calculated from the load sum of the vectors. When the estimation condition is specified by the estimation condition input unit 4, feature data is selected according to the estimation condition, and emotion information is estimated for this feature data.

  The output condition input unit 6 provides a condition for filtering emotion information to be output to the output unit 7.

  The output unit 7 outputs the emotion information estimated by the emotion estimation unit 5. When the output condition is specified by the output condition input unit 6, emotion information is output based on the specified output condition.

[Explanation of emotion estimation process in this embodiment]
The emotion estimation processes S1 to S4 of the present embodiment will be described with reference to FIGS.

(Data input)
S1: The data input unit 2 receives sensor data from biometric sensors worn by a plurality of users, and supplies the sensor data to the emotion estimation unit 5 when the sensor data is feature-extracted data. On the other hand, when the sensor data is data before feature extraction, the data is provided to the preprocessing unit 3.

  A specific input process S101 to S105 of S1 will be described with reference to FIG.

  (S101) Sensor data (for example, electrocardiogram, heartbeat, brain wave, muscle potential, sweating, acceleration, etc.) received from a plurality of user biosensors is read.

  (S102) It is determined whether the sensor data has been extracted (feature data) or has not been extracted (raw data).

  (S103) When the sensor data is feature data, this data is provided to the emotion estimation unit 5 without going through the preprocessing unit 3. The feature data includes a feature pattern ID. In order to filter the results, position information, time information, detection items, user information, and the like can be added thereto.

  (S104) If the sensor data is raw data, the data is provided to the preprocessing unit 3. The raw data includes a set of detection items and their data (discrete or continuous). In order to filter the emotion estimation result, position information, time information, user information, and the like can be added thereto.

  (S105) Steps S101 to S104 are executed for all sensor data received from the plurality of biological sensors.

(Preprocessing)
S2: The preprocessing unit 3 extracts feature information from the raw data provided from the data input unit 2, and assigns the feature pattern ID, which is an identifier of the feature information, to the data in association with the user ID and the detection item. Thus, the feature data is configured. Then, this feature data is provided to the emotion estimation unit 5.

  Specific preprocessing steps S201 to S205 of S2 will be described with reference to FIG.

  (S201) The raw data provided from the data input unit 2 is read.

  (S202) Feature information that matches a predefined feature pattern is detected from the read raw data.

  The feature pattern is defined in advance according to the detection item. Typically, a pattern of data that is considered non-standard with respect to a living organism is used. As a specific example, the pattern is defined such that the heart rate exceeds the standard deviation of the person's normal value by more than the standard deviation, and the body temperature is 0.2 degrees or more higher than normal. The feature pattern can also be defined for the normal state. For example, it can be defined as the feature pattern that the normal state continues for a time exceeding a predetermined threshold. As will be described later, the feature pattern is shared by the preprocessing unit 3, the estimation condition input unit 4, and the emotion estimation unit 5.

  (S203) A feature pattern ID corresponding to the information is assigned to the raw data in which the feature information is detected.

  The feature pattern ID is given to this data when a characteristic numerical value (for example, an abnormal value) that matches a predefined feature pattern is detected in the raw data.

  The feature pattern is encoded by giving a feature pattern ID. The naming rule of the feature pattern ID is arbitrary, and can be given as a numerical value, or the type name can be added to a part of the ID (such as “heartbeat_1”).

  (S204) The raw data to which the feature pattern ID is assigned is provided to the emotion estimation unit 5 as feature data. Information accompanying sensor data acquisition, such as time information, location information, or a user ID, can be further added to the feature data.

  (S205) Steps S201 to S204 are executed for all raw data provided from the input unit 2.

(Emotion information estimation)
S3: The emotion estimation unit 5 estimates the emotion information of the group based on the feature information of the feature pattern of the feature data provided from the data input unit 2 or the preprocessing unit 3, and supplies this emotion information to the output unit 7.

  Here, when the estimation condition is specified by the estimation condition input unit 4, feature data is selected according to the estimation condition, and emotion information is estimated based on the feature data. Further, when time information or place information is given from the estimation condition input unit 4, it is possible to perform emotion estimation weighted according to the time, the time from the point, or the distance in space.

  First, specific estimation condition input processes S311 to S312 executed by the estimation condition input unit 4 will be described with reference to FIG.

  (S311) The estimation condition is read from the storage area of the collective emotion estimation apparatus 1.

  When the feature data includes information other than the feature pattern ID, an estimation condition based on the information is given. For example, if time information and location information are included in the feature data, detection can be performed under conditions such as a specific time zone and a specific area. Further, if information such as a user ID and a user attribute is included in the feature data, the user ID and the user attribute can be specified as the estimation condition.

  (S312) The read estimation condition is provided to the emotion estimation unit 5.

  Next, specific emotion estimation processes S301 to S304 executed by the emotion estimation unit 5 will be described with reference to FIG.

  (S301) It is determined whether or not an estimation condition is designated by the estimation condition input unit 4.

  (S302) When the estimation condition is designated, feature data to be used is selected according to the designated estimation condition. For example, when the feature data includes time information, position information, or a user ID corresponding to the feature data, when the time information, position information, or user ID is input from the estimation condition input unit 4 as an estimation condition, Only feature data that matches this estimation condition is selected.

  (S303) Emotion information is estimated for the selected feature data group.

  In the estimation, an emotion pattern for expressing an emotion and an emotion estimation candidate based on an emotion pattern estimated for a single sensor data for each feature pattern ID are defined in advance. The emotion pattern is obtained by patterning a human emotion state and giving an emotion ID. The emotion estimation candidate is obtained by setting an estimated emotion ID in the form of a list, a weighted list, a probability vector, or the like for given single feature data.

  For emotion estimation, for each feature pattern ID of the feature data group, a corresponding emotion estimation candidate is selected, and the selected emotion estimation candidates are combined to perform group emotion estimation based on the entire feature data group.

  Hereinafter, specific steps S303 (1) to S303 (4) for emotion estimation will be described.

  S303 (1): A feature pattern is defined in advance.

  Table 1 shows examples of feature pattern definitions. Here, the normal state is also defined. In addition, detection items of sensor data are also defined so that they can be used as estimation conditions from the estimation condition input unit 4.

  The feature pattern is shared by the preprocessing unit 3, the estimation condition input unit 4, and the emotion estimation unit 5. When the data input from the data input unit 2 is encoded by the feature pattern ID, the feature pattern defined at the encoding stage for data input needs to be shared.

  S303 (2): An emotion pattern is defined in advance.

  An example of emotion pattern definition is shown in Table 2. The emotion pattern defined corresponding to the emotion content is also shared by the emotion estimation unit 5, the output condition input unit 6, and the output unit 7.

  S303 (3): Emotion estimation candidates are defined in advance.

  The relationship between the feature pattern ID and emotion data estimated from single feature data is defined in advance as an emotion estimation candidate. Tables 3 and 4 show examples of emotion estimation candidate tables storing definitions of emotion estimation candidates estimated for the feature information (feature pattern ID) of the feature data.

Table 3 is an example of a temporary emotion estimation candidate, and shows an estimated emotion pattern ID (f ₀ , f ₁ , f ₂ ,...) When a feature pattern of single feature data is given. Emotion estimation candidates can be expressed as a list as in the “emotion estimation candidate” column, or can be expressed as a probability vector as in the “vector expression” column. This probability vector is an M-dimensional vector for the number M of emotion patterns, and is referred to as an emotion estimation vector.

  Estimating emotion information for a feature pattern can also be defined for a composite of a plurality of feature patterns. Table 4 shows emotion estimation candidates based on a combination of a plurality of feature pattern IDs. This is a list or vector (emotion estimation vector) representing emotion estimation when a plurality of feature patterns appear simultaneously in the feature data.

  S303 (4): Emotion information is estimated.

  An example of emotion information estimation is shown by a method using a weighted sum of vectors. In this example, the following processes S321 and S322 are executed in the group emotion estimation for the feature data set S.

(S321) A collective feature pattern vector q _s for the feature data set S is created by the following steps (1) to (3). The vector q _s is an N-dimensional vector for the number N of feature pattern IDs.

(1) In the initial state, all elements of the vector q _s are set to 0.

(2) For one feature data sεS, if the feature pattern ID is p _i , add 1 to the value of the i-th element of the vector q _s .

(3) The vector q _s is constructed by repeating the step (2) for all elements of the feature data set S.

(S322) The function f _s indicating the emotion estimation pattern of the set S from the emotion estimation vectors v ₀ , v ₁ ,..., V _N for each feature pattern ID and the vector q _s is obtained by the following processes (1) to (3). calculate.

  (1) For the emotion estimation vector, the matrix V is configured as in the following equation (1).

(2) By calculating the product of the matrix V and the vector q _s , the product of frequency and probability is calculated for each feature pattern ID for each feature pattern ID, and the product sum for the emotion pattern is expressed by the following equation (2) _Is obtained as a vector g _s indicated by

(3) The element number of the element having the maximum value is obtained for the element of the vector g _s . If argmax is defined as a function f _s that returns an element number having the maximum value of a given vector, the emotion estimation result is calculated as an emotion pattern represented by the following function f _s (3).

(4) Estimation of weighted group emotions When a decreasing function according to a specific time and time interval is input from the estimation condition input unit 4 or when a decreasing function according to a specific point and distance is input, The estimation can be calculated with weight for each feature data.

For example, consider a case where a specific point and a decreasing function f (x) = 1 / x for the distance x are input as estimation conditions. At this time, if the distance from the designated point is L with respect to the feature data s∈S, in the step (2), the value of the i-th element of the vector q _s is added by 1, By adding 1 / L instead of 1, collective emotion estimation can be performed in consideration of the weight due to distance.

(5) Estimating Emotion Information by Composite of Feature Pattern IDs When using emotion estimation candidates by composite of feature pattern IDs, the number of dimensions of the group feature pattern vector q _s is 2 ^s dimensional with respect to the number of feature patterns N. A vector is used, and a combination label of feature patterns may be given to each element.

  (S304) The emotion pattern obtained by the estimation in step S303 is provided to the output unit 7 as emotion information. This emotion information includes the emotion type of the estimation result. The emotion information can also include numerical values such as strength or certainty.

(Output emotion information)
S4: The output unit 7 outputs the emotion information calculated by the emotion estimation unit 5.

  Here, when the output condition is designated from the output condition input unit 6, the emotion information is filtered according to the designated output condition. As a result, it is possible to output an estimation that is limited to positive emotions, or an output that excludes emotions that are determined to be unable to ensure sufficient estimation accuracy.

  First, the output condition input processes S411 to S413 executed by the output condition input unit 6 will be described with reference to FIG.

  (S411) The output condition is read from the storage area. In specifying the output condition, the emotion pattern ID used in emotion estimation is applied.

  (S412) The read output condition is encoded. When the output condition is not expressed by the emotion pattern ID, it is encoded into the emotion pattern ID. It is also possible to specify a threshold value to be output for the emotion strength or certainty factor.

  (S413) The encoded output condition is provided to the output unit 7.

  Next, specific output processes S401 to S404 executed by the output unit 7 will be described with reference to FIG.

  (S401) Receives emotion information from the emotion estimation unit 5.

  (S402) It is determined whether or not an output condition is specified by the output condition input unit 6.

  (S403) When the output condition is designated, the output condition is provided from the output condition input unit 6 to the output unit 7.

  (S404) The emotion information received in step S401 or the emotion information read based on the output condition specified in step S403 is output. This emotion information is further output and displayed on a display unit such as a monitor.

  The emotion information output from the output unit 7 includes at least three sets of “time information”, “position information”, and “estimated group emotion information”. It should be noted that a numerical value such as emotion strength or certainty can be added to the emotion information.

[Effect of this embodiment]
As described above, according to the collective emotion estimation apparatus 1 of the present embodiment, the emotion estimation vectors are weighted and integrated according to the probability distribution related to the emotion estimation of each feature data in the grouped sensor data set. Thus, emotion information of the group is estimated. Therefore, even if the emotion estimation accuracy is not sufficiently high at the individual level, the emotional state of the group can be estimated with high accuracy without estimating the emotion at the individual level. Also, by combining a plurality of biosensor data, emotion estimation that cannot be expected with only one biosensor data can be realized with high accuracy. In particular, when a group has the same experience, such as sharing a place, emotion estimation is expected to be more accurate.

  In particular, even if the biometric information provided from the biosensor is not already extracted, the preprocessing unit 3 extracts the feature information, so that the biometric information can be provided to the emotion estimation unit 5.

  Further, when estimating the emotion information, the estimation condition input unit 4 that inputs a function for weighting the estimation of the emotion information can arbitrarily weight the estimation. For example, as the function to be weighted, a decreasing function based on a specific time and an interval between the specific times or a decreasing function based on a distance between the specific position and the specific position is provided, thereby depending on time information or position information. Emotion information can be estimated.

  Furthermore, by providing the conditions for filtering the output of emotion information to the output unit 7 from the output condition input unit 6, the emotion information output from the output unit 7 is arbitrarily defined, for example, time information or position information. It can be arbitrarily defined according to

[Other Embodiments of the Present Invention]
The present invention can be realized by configuring a program that causes a computer to function as part or all of the functional units 2 to 7 constituting the collective emotion estimation apparatus 1 and causing the computer to execute the program. Alternatively, it can be realized by configuring a program that causes a computer to execute a part or all of the above-described steps S1 to S4 executed by the apparatus 1, and causing the computer to execute the program. The program can be provided by being stored in a known recording medium (for example, a hard disk, a flexible disk, a CD-ROM, etc.) that can be read by the computer. Alternatively, the program can be provided via the network via the Internet or e-mail.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

DESCRIPTION OF SYMBOLS 1 ... Collective emotion estimation apparatus 2 ... Data input part (input means)
3. Pre-processing unit (pre-processing means)
4 ... Estimation condition input section (estimation condition input means)
5 ... Emotion estimation unit (Emotion estimation means)
6 ... Output condition input section (output condition input means)
7. Output unit (output means)

Claims (7)

A group emotion estimation device for estimating group emotion information from biological information received from a plurality of individual biological sensors,
Input means for inputting biometric information received from a plurality of individual biometric sensors;
Preprocessing means for outputting feature data in which feature pattern IDs corresponding to feature information of the biological information are associated with the biological information;
As estimation conditions for emotion estimation, an estimation function that inputs a decreasing function based on time information including a specific time and a time interval from the specific time, and a decreasing function based on position information including a specific position and a distance from the specific position Condition input means;
Emotion pattern ID corresponding to emotion content given to a single feature pattern ID of the feature data selected according to the estimation condition input by the estimation condition input means or a set of a plurality of feature pattern IDs The emotion content of the emotion pattern ID corresponding to the element number of the emotion estimation vector having the maximum load sum for the plurality of individuals of the emotion estimation vector, which is a probability vector indicating an emotion estimation candidate consisting of Emotion estimation means to output as group emotion information,
An output condition input means for inputting an output condition;
Output means for outputting emotion information of the group that matches the input output condition,
The preprocessing means adds time information and position information at the time when the feature information is extracted to the feature data,
The emotion estimation unit, by decreasing function based on decreasing function or the position information based on the time information input in the estimated condition input means, the characteristic data to which the time information and the positional information in the pre-processing means is added A collective emotion estimation apparatus characterized in that the emotion estimation vector is weighted based on the above. The group emotion estimation apparatus according to claim 1, wherein the time information is an absolute time or a relative elapsed time from the start of a specific event. The group emotion estimation apparatus according to claim 1, wherein the position information is based on physical position information or virtual position information. A group emotion estimation method executed by a group emotion estimation apparatus that estimates group emotion information from biological information received from a plurality of individual biological sensors,
An input step for inputting biometric information received from a plurality of individual biometric sensors;
A preprocessing step of outputting feature data in which a feature pattern ID corresponding to feature information of the biometric information is associated with the biometric information;
As estimation conditions for emotion estimation, an estimation function that inputs a decreasing function based on time information including a specific time and a time interval from the specific time, and a decreasing function based on position information including a specific position and a distance from the specific position Condition input step,
Emotion pattern ID corresponding to emotion content given to a single feature pattern ID or a set of a plurality of feature pattern IDs of the feature data selected according to the estimation condition input in the estimation condition input step The emotion content of the emotion pattern ID corresponding to the element number of the emotion estimation vector having the maximum load sum for the plurality of individuals of the emotion estimation vector, which is a probability vector indicating an emotion estimation candidate consisting of An emotion estimation step to output as group emotion information;
An output condition input step for inputting an output condition;
An output step of outputting emotion information of the group that matches the input output condition;
The preprocessing step adds time information and position information at the time when the feature information is extracted to the feature data,
The emotion estimation step, by diminishing function based on decreasing function or the position information based on the time information input in the estimated condition input step, the feature data to which the time information and location information is added in the pretreatment step A group emotion estimation method characterized in that the emotion estimation vector is weighted based on the above. 5. The group emotion estimation method according to claim 4, wherein the time information is an absolute time or a relative elapsed time from the start of a specific event. 5. The group emotion estimation method according to claim 4, wherein the position information is based on physical position information or virtual position information.   A group emotion estimation program for causing a computer to function as each means constituting the group emotion estimation apparatus according to any one of claims 1 to 3.

Images