WO2020071233A1 - Information processing device, information processng method, and program - Google Patents

Abstract

In order to guide the collection of suitable learning data by an operator, provided is an information processing device (20) equipped with a guide unit (220) for guiding the collection of learning data used in learning by a classifier, wherein the guide unit dynamically guides the collection of data by an operator so as to ensure diversity of sensing data associated with a recognition subject. Also provided is an information processing method that involves a processor (871) guiding the collection of learning data used in learning by a classifier, and the guiding further includes dynamically guiding the collection of data by an operator so as to ensure diversity of sensing data associated with a recognition subject.

Description

Information processing apparatus, information processing method, and program

The present disclosure relates to an information processing device, an information processing method, and a program.

In recent years, devices that perform various types of recognition processing based on collected sensing data have become widespread. Also, a technique for efficiently generating a recognizer for realizing the above-described recognition processing has been reported. For example, Patent Literature 1 discloses a technique for efficiently collecting learning data effective for improving the generalization performance of a recognizer.

JP, 2018-60268, A

However, the technique described in Patent Literature 1 is a technique for selecting, from accidentally acquired sensing data, data suitable as learning data, and does not guide the operator to collect data appropriately.

According to the present disclosure, there is provided a guiding unit that guides collection of learning data used for learning of a recognizer, and the guiding unit is configured to collect data by an operator so that diversity of sensing data related to a recognition target is secured. An information processing device for dynamically guiding the information is provided.

Further, according to the present disclosure, the processor includes: guiding the collection of learning data used for learning of the recognizer, including, the guiding is to ensure the diversity of sensing data related to the recognition target, An information processing method is provided, further comprising dynamically guiding data collection by an operator.

According to the present disclosure, the computer includes a guiding unit that guides the collection of learning data used for learning of the recognizer, and the guiding unit ensures the diversity of sensing data related to the recognition target, A program for functioning as an information processing device for dynamically guiding data collection by an operator is provided.

1 is a block diagram illustrating a configuration example of an information processing system according to an embodiment of the present disclosure. It is a figure showing an example of the class of hand sign identification concerning the embodiment. It is a figure showing an example of appearance of an embedded device concerning the embodiment. FIG. 3 is a block diagram illustrating a functional configuration example of the embedded device according to the embodiment. FIG. 3 is a block diagram illustrating a functional configuration example of the information processing apparatus according to the embodiment. FIG. 3 is a block diagram illustrating a configuration example when the information processing apparatus according to the embodiment has a data collection function and a user interface display function. It is a flow chart which shows a flow of processing of an information processing system concerning the embodiment. It is an example of a data collection flow according to the embodiment. It is a figure showing an example of a restriction which a guidance part concerning the embodiment imposes on an operator. It is a figure for explaining the diversity of the learning data concerning the embodiment. It is a figure showing an example of a heat map concerning the frequency of a hand concerning the embodiment. It is a figure which shows an example of the guidance of the learning data collection of the pattern which a shortage is assumed by the guidance part which concerns on the embodiment. It is an example of a graph showing transition of recognition accuracy according to the embodiment. It is a figure showing an example of combination of an embedded device, a sensor, and a recognition task concerning the embodiment. It is a figure showing an example when the embedded device concerning the embodiment is realized as a wristwatch type wearable device. It is a figure showing an example of guidance using the glass equipment and tactile sense presentation equipment concerning the embodiment. FIG. 3 is a block diagram in a case where the embedded device according to the embodiment includes a reward providing unit. It is a figure for explaining diversion to the practical recognizer development concerning the embodiment. 1 is a diagram illustrating an example of a hardware configuration of an information processing device according to an embodiment of the present disclosure. FIG. 7 is a diagram for explaining the effect of learning data on the generalization performance of a recognizer.

Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant description is omitted.

The description will be made in the following order.
1. Embodiment 1.1. Overview 1.2. System configuration example 1.3. Example of functional configuration of embedded device 10 1.4. Example of functional configuration of information processing device 20 1.5. Details of processing 1.6. Modified example 2. 2. Hardware configuration example Conclusion

<1. Embodiment>
<< 1.1. Overview >>
First, an outline of an embodiment of the present disclosure will be described. As described above, in recent years, devices that perform various recognition processes based on collected sensing data have become widespread in the fields of IoT (Internet of things) and robotics.

Here, when developing a recognizer using supervised machine learning technology such as deep learning, the development cycle consisting of design of the recognizer, collection of learning data, learning of the recognizer, and evaluation of the recognizer is repeated. General.

に も Also, even if a novice in the field of machine learning tries to acquire the development technology of the recognizer, it is important that each individual experiences the development cycle as described above.

In recent years, beginners have experienced recognizer design, learning, and evaluation using, for example, free or paid machine learning frameworks and learning datasets, and mastered recognizer development techniques. It is possible.

However, in the above-described machine learning framework, although a tutorial on the design, learning, and evaluation of a recognizer is provided, learning data used for learning is, for example, an existing data set that is freely released on the Internet. The diversion remains. On the other hand, learning data collection is an important phase that greatly affects the generalization performance of a recognizer.

FIG. 20 is a diagram for explaining the effect of the learning data on the generalization performance of the recognizer. For example, the upper part of FIG. 20 shows an example in which the learning data has not been exhaustively collected in the sample space Ω. In such a case, the generated recognizer has high recognition accuracy for the data used for learning, but has low recognition accuracy for unknown data, that is, low so-called generalization performance.

On the other hand, as shown in the lower part of FIG. 20, when the learning data has been exhaustively collected in the sample space Ω, the generalization performance of the generated recognizer increases, and it is possible to correctly recognize more data. It becomes possible.

Therefore, in collecting learning data in the development of a recognizer, it is important to collect data on various assumed situations comprehensively. However, beginners who learn recognizer development techniques using a machine learning framework that diverts existing data sets as described above will learn a learning data collection method suitable for improving the generalization performance of recognizers. It is difficult.

Further, for example, as disclosed in Patent Literature 1, a technique for efficiently collecting learning data effective for improving the generalization performance of a recognizer has been proposed. The described technology is a technology for selecting data suitable for learning data from sensing data acquired accidentally, and does not allow a user to learn a suitable learning data collection method.

技術 The technical idea according to the present disclosure has been conceived by focusing on the above points, and it is possible to guide the operator to appropriately collect learning data. To this end, the information processing device 20 according to an embodiment of the present disclosure includes a guiding unit 220 that guides collection of learning data used for learning of a recognizer. In addition, the guiding unit 220 according to an embodiment of the present disclosure dynamically guides an operator to collect data so that diversity of sensing data related to a recognition target is reported.

Hereinafter, the configuration of the information processing apparatus 20 that realizes the above-described features and the effects of the above-described features will be described in detail.

<< 1.2. System configuration example >>
First, a configuration example of an information processing system according to an embodiment of the present disclosure will be described. FIG. 1 is a block diagram illustrating a configuration example of an information processing system according to the present embodiment. Note that FIG. 1 illustrates an example in which the technical concept according to the present disclosure is applied to a development tutorial for a recognizer that performs handsign identification.

ハ ン ド Here, the above-mentioned hand sign identification refers to a recognition task of classifying a human hand shown in an image into an optimal class according to its shape. FIG. 2 is a diagram illustrating an example of a class of handsign identification according to the present embodiment. As shown in FIG. 2, the hand sign identification class according to the present embodiment includes, for example, a hand shape used for rock-paper-scissors such as Rock (rock), Scissors (choki), and Paper (ぱ), and a half-heart. Various hand shapes may be included.

By using the information processing system according to the present embodiment, the operator who is a novice in the development of the recognizer can use the information processing system according to the present embodiment to improve the generalization performance of recognition for performing the above-described hand sign identification. It is possible to obtain data collection methods.

As shown in FIG. 1, the information processing system according to the present embodiment may include, for example, an embedded device 10, an information processing device 20, and a display device 30.

(Embedded device 10)
The embedded device 10 according to the present embodiment is a device that collects sensing data related to a recognition target and presents the collected sensing data (hereinafter, also referred to as learning data) to an operator in real time. For this purpose, the embedded device 10 according to the present embodiment includes the output unit 110 and the sensor 120.

In the case of the example shown in FIG. 1, the embedded device 10 captures a handsign by a beginner of the recognizer development by the sensor 120 which is a camera device, and displays an image related to the captured handsign by the output unit 110 in real time. Good.

FIG. 3 is a diagram illustrating an example of the appearance of the embedded device 10 according to the present embodiment. An operator who is a novice, for example, inserts his hand under the sensor 120 which is a camera device fixed downward, and executes data collection related to various hand signs while confirming a captured image by the output unit 110. I do.

(Information processing device 20)
The information processing device 20 according to the present embodiment allows an operator who is a novice in the development of a recognizer to experience the phases of design of the recognizer, collection of learning data, learning of the recognizer, and evaluation of the recognizer. Provide a framework. For this reason, the information processing apparatus 20 according to the present embodiment includes a guidance UI1 for performing guidance in each of the above phases and a development UI2 for an operator who is a novice of recognizer development to design a recognizer. It is provided to the user through the display device 30.

The information processing apparatus 20 according to the present embodiment has a function of dynamically guiding data collection by an operator so that the diversity of sensing data relating to a recognition target is ensured.

機能 The details of the functions of the information processing apparatus 20 according to the present embodiment will be separately described later.

(Display device 30)
The display device 30 according to the present embodiment displays the above-mentioned guidance UI1 and development UI2 based on the control by the information processing device 20. The display device 30 according to the present embodiment may be, for example, a liquid crystal display (LCD) device, an organic light emitting diode (OLED) device, or a projector.

(Network 40)
The network 40 has a function of connecting the components included in the information processing system. The network 40 may include a public network such as the Internet, a telephone network, or a satellite communication network, various LANs (Local Area Network) including Ethernet (registered trademark), and a WAN (Wide Area Network). Further, the network 40 may include a dedicated line network such as an IP-VPN (Internet Protocol-Virtual Private Network). The network 40 may include a wireless communication network such as Wi-Fi (registered trademark) and Bluetooth (registered trademark).

The configuration example of the information processing system according to the present embodiment has been described above. The configuration described above with reference to FIG. 1 is merely an example, and the configuration of the information processing system according to the present embodiment is not limited to the example. The configuration of the information processing system according to the present embodiment can be flexibly modified according to specifications and operations.

<< 1.3. Functional configuration example of embedded device 10 >>
Next, a functional configuration example of the embedded device 10 according to the present embodiment will be described. FIG. 4 is a block diagram illustrating a functional configuration example of the embedded device 10 according to the present embodiment. As illustrated in FIG. 4, the embedded device 10 according to the present embodiment may include an output unit 110, a sensor 120, a recognition execution unit 130, a driving unit 140, a learning data collection unit 150, and a communication unit 160.

(Output unit 110)
The output unit 110 according to the embodiment presents the sensing data collected by the sensor 120 to the user in real time. The output unit 110 according to the embodiment may include, for example, a display device, a speaker, a vibration device, and the like.

(Sensor 120)
The sensor 120 according to the present embodiment collects sensing data related to a recognition target. The sensor 120 according to the present embodiment may include a device according to a recognition task and a recognition target. The sensor 120 according to the present embodiment may include, for example, a camera device, a microphone, an IMU (Inertial Measurement Unit), a pulse sensor, an electrocardiograph, and the like.

(Recognition execution unit 130)
The recognition execution unit 130 performs recognition of the recognition target using a learning device generated based on the sensing data collected by the sensor 120. The recognition execution unit 130 according to the present embodiment can recognize the handsign performed by the operator in real time, for example, using a recognizer generated based on the image related to the handsign collected by the sensor 120.

(Drive unit 140)
The drive unit 140 according to the present embodiment controls the driving of the embedded device 10 itself and the sensor 120 based on the control by the information processing device 20. The drive unit 140 may change the position or orientation of the embedded device 10 or the sensor 120 based on the control of the information processing device 20, for example.

(Learning data collection unit 150)
The learning data collection unit 150 according to the present embodiment causes the sensor 120 to execute data collection based on control by the information processing device 20 and transmits the collected sensing data to the information processing device 20 via the communication unit 160. . The learning data collection unit 150 may cause the sensor 120 to execute data collection via the driving unit 140.

(Communication unit 160)
The communication unit 160 according to the present embodiment performs information communication with the information processing device 20 via the network 40. For example, the communication unit 160 transmits the sensing data collected by the sensor 120 to the information processing device 20. Further, for example, the communication unit 160 receives a control signal and a recognizer generated by the information processing device 20.

As described above, the functional configuration example of the embedded device 10 according to the present embodiment has been described. The configuration described above with reference to FIG. 4 is merely an example, and the functional configuration of the embedded device 10 according to the present embodiment is not limited to the example. The functional configuration of the embedded device 10 according to the present embodiment can be flexibly modified according to specifications and operation.

<< 1.4. Functional configuration example of information processing device 20 >>
Next, an example of a functional configuration of the information processing device 20 according to the present embodiment will be described. FIG. 5 is a block diagram illustrating a functional configuration example of the information processing apparatus 20 according to the present embodiment. As shown in FIG. 5, the information processing device 20 according to the present embodiment includes an output control unit 210, a guiding unit 220, a learning / evaluation unit 230, a learning data set storage unit 240, an evaluation data set storage unit 250, and a recognizer storage. And a communication unit 270.

(Output control unit 210)
The output control unit 210 according to the present embodiment controls a user interface (guidance UI1) related to data collection guidance based on control by the guidance unit 220.

The output control unit 210 according to the present embodiment controls a user interface (development UI2) for an operator who is a novice in developing a recognizer to design a recognizer.

The output control unit 210 according to the present embodiment may control the display of the guidance UI1 and the development UI2 by the display device 30 via the communication unit 270 and the network 40.

(Guiding unit 220)
The guiding unit 220 according to the present embodiment has a function of guiding the learning / evaluating unit 230 to collect learning data used for learning the recognizer. Specifically, one of the features is that the guiding unit 220 according to the present embodiment dynamically guides the data collection by the operator so that the diversity of the sensing data related to the recognition target is secured.

At this time, the guidance unit 220 according to the present embodiment may impose a restriction on the collection of learning data to the operator, and perform guidance so that the diversity of the sensing data is secured within the range of the restriction.

The guidance unit 220 according to the present embodiment may include a labeling unit 222 that performs labeling on collected sensing data, and a data analysis unit 224 that performs determination regarding the above-mentioned diversity restriction. The details of the function of the guiding unit 220 according to the present embodiment will be separately described later.

(Learning / evaluation unit 230)
The learning / evaluation unit 230 according to the present embodiment functions as a learning unit that performs learning related to the recognizer using the sensing data collected by the embedded device 10 and an evaluation unit that evaluates the generated recognizer.

(Learning data set storage unit 240)
The learning data set storage unit 240 according to the present embodiment stores sensing data collected by the embedded device 10 as learning data related to recognizer learning by the learning / evaluation unit 230.

(Evaluation data set storage unit 250)
The evaluation data set storage unit 250 storage unit according to the present embodiment stores an evaluation data set used for the recognizer evaluation by the learning / evaluation unit 230.

(Recognizer storage unit 260)
The recognizer storage unit 260 according to the present embodiment stores information on the recognizer generated by the learning / evaluation unit 230 through learning. The information includes, for example, a network configuration and various parameters.

(Communication unit 270)
The communication unit 270 according to the present embodiment performs information communication with the embedded device 10 and the display device 30 via the network 40. For example, the communication unit 270 receives the sensing data collected by the embedded device 10. Further, for example, the communication unit 270 transmits to the embedded device 10 a control signal generated by the guidance unit 220 and information on recognition generated by learning by the learning / evaluation unit 230.

The example of the functional configuration of the information processing device 20 according to the embodiment has been described above. The configuration described above with reference to FIG. 5 is merely an example, and the functional configuration of the information processing device 20 according to the present embodiment is not limited to the example.

For example, the information processing device 20 according to the present embodiment may further include the functions of the embedded device 10 and the display device 30. FIG. 6 is a block diagram illustrating a configuration example when the information processing apparatus 20 according to the present embodiment has a data collection function and a user interface display function. As illustrated in FIG. 6, the information processing system according to the present embodiment does not necessarily need to include the embedded device 10, the information processing device 20, and the display device 30.

On the other hand, the functions of the information processing device 20 according to the present embodiment may be realized by two or more devices. As described above, the functional configuration of the information processing device 20 according to the present embodiment can be flexibly modified according to specifications and operations.

<< 1.5. Details of processing >>
Next, a processing flow of the information processing system according to the present embodiment will be described in detail. In the following, a description will be mainly given of a case where the information processing system according to the present embodiment is applied to a development tutorial for a recognizer that performs handsign identification.

FIG. 7 is a flowchart showing a flow of processing of the information processing system according to the present embodiment. Referring to FIG. 7, first, the output control unit 210 of the information processing device 20 displays the guidance UI1 on the display device 30 under the control of the guidance unit 220, and instructs the design of the recognizer (S1101).

At this time, under the control of the guidance unit 220, the output control unit 210 uses the instructor avatar AV1 as shown in FIG. 1 or a moving image of the actual instructor to perform the recognition task in the tutorial and the recognition task in the tutorial. Provide guidance on the recognizer to solve.

An operator who is a novice in developing a recognizer may actually design a recognizer through the development UI 2 displayed on the display device 30 according to the above-described guidance.

Next, under the control of the guidance unit 220, the output control unit 210 provides guidance UI1 as to what learning data is required to generalize the recognizer designed by the novice operator through the development UI2. Is notified to the operator via an instruction to instruct the operator to actually collect data (S1102).

In the case of the above-described handsign identification, the output control unit 210, in the guidance UI1, under the control of the guidance unit 220, "take 300 images each in the order of Rock, Scissors, Paper, and Half-heart". The data collection flow as shown in FIG. 8 may be displayed.

{At this time, the guidance UI1 describes that the output control unit 210 imposes restrictions on collection of sensing data under the control of the guidance unit 220. Here, the above restriction is intended to simplify the recognition task in the tutorial of the recognizer development and to reduce the amount of learning data required for generalization of the recognizer.

FIG. 9 is a diagram illustrating an example of restrictions imposed on the operator by the guiding unit 220 according to the present embodiment. For example, the guidance unit 220 according to the present embodiment may impose a constraint on the operator that the background of the image is entirely white and the ambient light is constant. According to the above restriction, it is possible to avoid a complicated recognition task due to a different background pattern or color depending on an image, and it is possible to effectively reduce a necessary learning data amount. Note that the guidance unit 220 can determine whether or not the above-mentioned restriction is complied with, for example, by performing a density histogram analysis or a determination using an illuminance sensor.

誘導 In addition, for example, the guiding unit 220 may impose a restriction on the operator that the hand to be photographed is one person, that is, one hand of the operator, and the surface to the camera is limited to one side. As described above, the restriction according to the present embodiment may limit the recognition target to a part or behavior of the operator's body. Under the control of the guidance unit 220, the output control unit 210 may realize the description related to the restriction using, for example, the voice SO1 shown in FIG. According to the above restriction, it is possible to avoid a complicated recognition task due to a difference in hand shape, color, and photographing surface depending on an image, and to effectively reduce a necessary learning data amount. The guidance unit 220 can determine whether or not the above-described restriction is adhered to, for example, by using joint detection and either fingerprint detection or authentication in combination.

For example, the guide unit 220 may impose on the operator that the camera device included in the sensor 120 is fixed downward, as a constraint. According to the above constraint, it is possible to prevent the recognition task from being complicated due to the difference in the posture of the camera device depending on the image, and to effectively reduce the required amount of learning data. Note that the guidance unit 220 can determine whether or not the above-described restriction is adhered to, for example, using an acceleration signal or the like of the camera device.

制約 Above, the restrictions according to the present embodiment have been described with reference to specific examples. As described above, the guiding unit 220 according to the present embodiment imposes a restriction on collection of learning data to an operator who is a novice in developing a recognizer, and diversity of sensing data is secured within the range of the restriction. Thus, data collection by the operator can be dynamically guided.

According to the above function of the guiding unit 220 according to the present embodiment, it is possible to effectively simplify the recognition task in the tutorial of the recognizer development, and to greatly reduce the learning data required for generalizing the recognizer. Further, according to the above-described function of the guiding unit 220 according to the present embodiment, an operator who is a novice in the development of a recognizer can intuitively grasp the importance of a constraint in learning data collection. .

Next, the guidance unit 220 instructs the embedded device 10 to acquire learning data via the communication unit 270 (S1103). At this time, the learning data collection unit 150 of the embedded device 10 starts collection of the learning data by the sensor 120 based on the above instruction received via the communication unit 160.

If the recognition task is handsign identification, the operator inserts his / her hand in front of the camera device provided in the sensor 120 according to the above-described restrictions from step S1103, and displays it on the display device provided in the output unit 110. Experience the collection of learning data while checking the images that are displayed.

(4) The learning data collection unit 150 transmits the learning data to the information processing device 20 via the communication unit 160 as soon as learning data, that is, sensing data relating to the recognition target is obtained. The guidance unit 220 of the information processing device 20 causes the learning data set storage unit 240 to store the received learning data.

Next, the labeling unit 222 of the guidance unit 220 performs labeling on the learning data stored in step S1103 (S1104). At this time, the labeling unit 222 labels classes according to each phase on the assumption that data collection is performed according to a predetermined flow shown in FIG. 8 and presented to the operator in step S1102. You may. The labeling unit 222 may perform labeling using, for example, a generalized recognizer prepared in advance. The labeling unit 222 may use labeling according to each phase in combination with labeling using a generalized recognizer to improve the accuracy of labeling.

Next, the data analysis unit 224 of the guidance unit 220 analyzes the learning data stored in step S1103, and verifies compliance with the constraint imposed by the learning data in step S1102 (S1105).

For example, when the recognition task is handsign identification, the data analysis unit 224 can determine whether or not the constraint as shown in FIG. 9 is observed, using the verification method shown in FIG. .

{Circle around (1)} The data analysis unit 224 verifies whether sufficient diversity for generalization of the recognizer is secured within the range of the constraint imposed in step S1102 (S1106). For example, when the recognition task is hand sign identification, the data analysis unit 224 recognizes the finger or wrist joint in the image by detecting the joint, and can realize various shootings within the degrees of freedom allowed by each class. You can check if it is.

FIG. 10 is a diagram for explaining the diversity of learning data according to the present embodiment. FIG. 10 illustrates four images related to the class “Scissors”, but the finger joint positions are different from each other.

Specifically, in the two images on the left, the tip of the thumb is hidden by the ring finger and the little finger, whereas in the two images on the right, the tip of the thumb is located before the ring finger and the little finger. . In the first and third images from the left, the index finger and the middle finger are in close contact with each other, whereas in the second and fourth images from the left, the index finger and the middle finger are far apart. ing.

On the other hand, the four images shown in FIG. 10 can be said to be appropriate learning data under the condition of “an image in which Scissors is photographed by the right hand of one individual from the palm side”. For this reason, it is important to collect all the images close to the state of each joint in order to secure sufficient diversity for generalization of the recognizer.

For this reason, the data analysis unit 224 according to the present embodiment calculates the collection rate of the already captured joint pattern after defining an assumed joint pattern as shown in FIG. Rate may be used as one of the measures of diversity.

手法 In addition, a method of measuring diversity from the position of the hand as well as the state of the joint is also effective. For this reason, the data analysis unit 224 according to the present embodiment generates, for example, a heat map related to the presence frequency of the hand in the angle of view of the camera device of the sensor 120 based on the position of the hand detected from each learning data. Alternatively, the heat map may be used as one of the measures for measuring diversity.

FIG. 11 is a diagram showing an example of a heat map according to the hand presence frequency according to the present embodiment. The data analysis unit according to the present embodiment can generate a heat map as shown in FIG. 11 using, for example, a generalized hand recognizer. At this time, the data analysis unit 224 verifies whether or not the detected positions of the hands are widely distributed in the heat map.

{For example, in the case of the example shown in FIG. 11, it can be seen that no hand is detected near the region R1. In this case, the data analysis unit 224 may determine that the collected learning data cannot secure sufficient diversity for generalization of the recognizer.

Next, the learning / evaluating unit 230 according to the present embodiment performs learning of the recognizer using the learning data stored in step S1103 and the labeling result in step S1104 (S1107).

The learning / evaluation unit 230 according to the present embodiment may perform deep learning using a stochastic gradient descent method, for example. Further, the learning / evaluation unit 230 according to the present embodiment may be provided as software realized by various machine learning frameworks.

Next, the learning / evaluation unit 230 according to the present embodiment evaluates the recognizer learned in step 1107 (S1108). At this time, the learning / evaluation unit 230 according to the present embodiment may perform an open test using the evaluation data set stored in the evaluation data set storage unit 250 in advance. On the other hand, the learning / evaluation unit 230 may perform cross-validation using a collected or stored learning data set.

Next, the guiding unit 220 according to the present embodiment determines whether the development flow relating to the recognizer satisfies a predetermined termination condition (S1109). At this time, the guiding unit 220 performs the above-described determination on the condition that the amount of the collected learning data and the recognition accuracy of the recognizer obtained in the evaluation in step S1108 are equal to or higher than a predetermined threshold, for example. You may. In addition, the guidance unit 220 may set one of the end conditions such as observing the constraint imposed in step S1102, ensuring that the diversity of the collected learning data is ensured, and the like.

Here, when the development flow relating to the recognizer satisfies a predetermined termination condition (S1109: YES), the guiding unit 220 is connected to either the output unit 110 of the embedded device 10 or the display device 30, or both. The operator is instructed to end the data collection, and the recognizer trained in step S1108 is stored in the recognizer storage unit 260 (S1110).

On the other hand, when the development flow relating to the recognizer does not satisfy the predetermined termination condition (S1109: NO), the guiding unit 220 causes the operator to continue collecting learning data.

Here, for example, when the guidance unit 220 determines that the constraint is not observed in the verification of the observance of the constraint in step S1105, a warning that the constraint is not observed is issued by a novice of the recognizer development. A certain operator is notified (S1111). At this time, the guidance unit 220 may cause the output unit 110 of the embedded device 10 and / or the display device 30 to notify the above warning.

In addition, in the diversity verification in step S1106, when the guidance unit 220 determines that sufficient diversity is not secured for the generalization of the recognizer, learning data of a pattern that is assumed to be in shortage is collected. Data collection by the operator is dynamically guided (S1112).

For example, in the case of the example illustrated in FIG. 10, the guiding unit 220 may perform guidance such that images of the second and fourth joint patterns from the left, which have not yet been captured, are captured. In addition, for example, in the case of the example illustrated in FIG. 11, the guidance unit 220 may perform guidance so that an image in which the hand position is near the region R1 is captured.

FIG. 12 is a diagram illustrating an example of guidance by the guidance unit 220 according to the present embodiment for learning data collection of a pattern that is assumed to be in short supply. For example, with respect to the state of the joint, the guidance unit 220 performs learning by making the output unit 110 of the embedded device 10 and the speaker included in the display device 30 output gymnastics music and announcements in common to each class. Guidance relating to data collection may be performed.

The guidance unit 220 may visualize the collected learning data and display it on the output unit 110 or the display device 30 so as to visually guide the variety of the learning data.

For example, when the recognition target is a body sign including a hand sign or a gesture, the guidance unit 220 causes the output unit 110 or the display device 30 to display a guidance object for guiding the collection of learning data effective for securing diversity. Alternatively, the operator may be instructed to perform a predetermined operation on the guidance object.

For example, when the diversity regarding the state of the joint is not ensured for the class “Rock”, the guiding unit 220 causes the output unit 110 or the display device 30 to display the lid of the bottle as the guiding object, and the operator May be instructed to open and close the lid of the bottle.

Further, for example, in the case where the diversity relating to the state of the joint is not ensured for the class “Scissors”, the guiding unit 220 causes the output unit 110 and the display device 30 to display a thread as a guiding object, and allows the operator to It may be instructed to perform an operation of cutting the thread with a finger.

Further, for example, when the diversity related to the position of the hand is not ensured, the guiding unit 220 causes the output unit 110 and the display device 30 to display a virtual enemy in the game as a guiding object, and displays the virtual enemy at the position of the virtual enemy. It is also possible to instruct repulsion by joining hands and count the number of repulsions.

Further, for example, in the case where the diversity relating to the position of the hand is not ensured for the class “Paper”, the guiding unit 220 causes the output unit 110 and the display device 30 to display dirt, and displays the dirt with the palm. Instructions may be given to wipe off.

According to the above control by the guidance unit 220 according to the present embodiment, the state of the joints and the positions of the hands of the operator naturally diversify by the operation corresponding to the instruction, and the learning data can be effectively collected. Can be guided.

Further, the guidance unit 220 according to the present embodiment controls the learning data of the pattern whose shortage is predicted to be executed by the instructor avatar AV1 or the moving image in the guidance UI1 in addition to the above control. May be.

The guidance unit 220 according to the present embodiment can also drive the embedded device 10 and the sensor 120 to further increase the diversity of the learning data (S1113). The guidance unit 220 gives an instruction to the driving unit 140 to drive, for example, a motor, and to rotate or move the embedded device 10 or the sensor 120, so that the recognition target is in the same state. Therefore, the diversity of the learning data can be increased.

The guidance unit 220 according to the present embodiment may change the setting of the sensor 120 to further increase the diversity of the learning data (S1114). For example, when the sensor 120 includes a camera device, the guidance unit 220 changes the settings of the shutter speed, the aperture value, the ISO sensitivity, the exposure compensation, and the like in an exhaustive manner so that the recognition target is in the same state. However, the diversity of the learning data can be increased.

The guidance unit 220 according to the present embodiment controls presentation of the recognition accuracy of the recognizer that has performed learning using the learning data collected up to the present time to the operator (S1115). The guidance unit 220 may display the recognition accuracy of the recognizer calculated by the learning / evaluation unit 230 on the output unit 110 or the display device 30.

For example, when the recognition task is handsign identification, the guiding unit 220 according to the present embodiment may display a scalar value such as the accuracy or the F value at the present time, or a transition of the scalar value.

FIG. 13 is an example of a graph showing a change in recognition accuracy according to the present embodiment. In the case of the example illustrated in FIG. 13, the guiding unit 220 displays the transition of the accuracy of the recognizer in a time series. At this time, the guidance unit 220 according to the present embodiment may display the learning data collected at each time point in association with the recognition accuracy. In the example shown in FIG. 13, in the second Scissors imaging phase, the captured image is displayed in association with the accuracy. The image includes the thread-shaped guidance objects IOa and IOb described above, and indicates that the image has been captured under visual guidance by the guidance unit 220.

As described above, according to the guiding unit 220 according to the present embodiment, it is possible to visualize and present the collected learning data together with the recognition accuracy of the recognizer. It is possible to intuitively grasp what learning data has contributed to the improvement of accuracy, and to learn the importance and the point of learning data collection.

Note that the graph shown in FIG. 13 is merely an example, and the guiding unit 220 may display the recognition accuracy of the recognizer in various modes. The guidance unit 220 may cause the output unit 110 and the display device 30 to display a confusion matrix and the like, for example.

The guidance unit 220 according to the present embodiment can display the recognition result of the recognition target by the recognizer that has performed learning based on the collected learning data on the output unit 110 in real time (S1116).

At this time, the recognition execution unit 130 of the embedded device 10 acquires the recognizer learned from the information processing device 20 via the communication unit 160, and uses the recognizer to perform the learning on the learning data currently acquired by the sensor 120. Perform recognition processing. The guidance unit 220 according to the present embodiment may acquire the result of the recognition processing by the recognition execution unit 130 and cause the output unit 110 to display the result.

According to the above control by the guidance unit 220 according to the present embodiment, the operator can intuitively experience that the recognition result is grasped in real time and the accuracy is improved with the collection of the learning data. Become.

As described above, the processing flow of the information processing system according to the present embodiment has been described in detail. The information processing system according to the present embodiment may repeatedly execute the processing of steps S1103 to S1109 and S1111 to S1116 until the termination condition is satisfied.

According to the above processing by the information processing system according to the present embodiment, it is possible to effectively guide the collection of suitable learning data by the operator, and to provide the operator who is a novice in the development of the recognizer with recognition. It becomes possible to make learning the point of learning data collection in instrument development.

Note that the flowchart described with reference to FIG. 7 is merely an example, and the steps of the information processing system described above do not necessarily need to be processed in chronological order in the order described in the flowchart. For example, each step related to the processing of the information processing system may be processed in an order different from the order described in the flowchart, or may be processed in parallel.

For example, in the above description, an example has been described in which the guidance unit 220 according to the present embodiment executes learning data collection guidance using a guidance object or the like when determining that diversity of learning data is not ensured. However, the guidance unit 220 according to the present embodiment may perform the above-described guidance from the start of learning data collection.

In the above description, the operator according to the present embodiment is a novice of the recognizer development, and has been described as an example in which the novice learns the procedure of learning data collection. It is not limited to the example. The operator according to the present embodiment may be, for example, a cooperator for learning data collection that is not engaged in recognizer development (not aimed at acquiring knowledge related to recognizer development).

In such a case, it is not always necessary to inform the cooperator of the importance of learning data collection, and there may be situations in which it is desirable to prioritize the collection efficiency of learning data itself.

Therefore, in the above situation, the guidance unit 220 may omit part of the guidance relating to the importance of learning data collection and the like, and may perform guidance using guidance objects or the like from the beginning.

According to the above control by the guiding unit 220 according to the present embodiment, even when the operator does not have the knowledge regarding the recognizer development and is a cooperator who does not aim to acquire the knowledge, The behavior of the cooperator can be naturally induced, and learning data suitable for developing a recognizer can be efficiently collected.

<< 1.6. Modifications >>
Next, a modified example according to an embodiment of the present disclosure will be described. In the above description, the case where the recognition task is handsign identification is described as a main example, but the application range of the technical idea according to the present disclosure is not limited to such an example. The technical concept according to the present disclosure is applicable to various recognition tasks.

FIG. 14 is a diagram illustrating a combination example of the embedded device 10, the sensor 120, and the recognition task according to the present embodiment.

示 す As shown in FIG. 14, for example, the embedded device 10 according to the present embodiment may be a wearable device such as a wristwatch. In this case, the sensor 120 of the embedded device 10 may include, for example, an IMU. According to the above configuration, in addition to the above-described hand sign identification, a recognizer for gesture identification, operator behavior identification, and the like can be generated by learning.

For example, the embedded device 10 according to the present embodiment may be a smart speaker. In this case, the sensor 120 of the embedded device 10 may include, for example, a microphone. According to the above configuration, it is possible to generate a recognizer related to specific word detection, speaker identification, speech-to-text, or the like by learning.

For example, the embedded device 10 according to the present embodiment may be various wearable devices for health care. In this case, the sensor 120 of the embedded device 10 may include, for example, an IMU, a pulse sensor, an electrocardiograph, and the like. According to the configuration described above, it is also possible to generate an estimator for estimating a heart rate or estimating consumed energy by learning. As described above, the technical idea of the present disclosure is not limited to recognition such as hand sign identification, but can be applied to regression.

In addition, in the present embodiment, the guidance method may be appropriately modified according to the target recognition task. For example, as shown in FIG. 15, a case where the embedded device 10 is a wristwatch-type wearable device worn on the operator's arm UA, and a gesture of turning the screen to the operator's face is detected by the IMU included in the sensor 120 is illustrated. Suppose.

In this case, as shown in FIG. 16, the output unit 110 of the embedded device 10 may be configured by a glass device 110a that is compatible with virtual reality or augmented reality, and a tactile presentation device 110b that is worn by the operator U1.

At this time, the guiding unit 220 according to the present embodiment first causes the glass device 110a to display the instructor avatar AV1 in step S1112 shown in FIG. In addition, the guiding unit 220 is configured so that a variety of learning data can be acquired for both a gesture of turning the screen of the embedded device 10 that is a wristwatch-type wearable device to the face and an operation that is not similar to the gesture. Let AV1 exemplify a specific hand movement. In the case of the example illustrated in FIG. 16, the guiding unit 220 exemplifies how to move a hand with the sound SO2.

Further, at this time, the guidance unit 220 performs control such as pseudo-pulling the hand of the operator U1 by causing the tactile presentation device 110b to present traction so that diversity is generated in the learning data. Good. According to the above control by the guidance unit 220 according to the present embodiment, it is possible to intuitively teach the operator U1 what kind of hand movement brings diversity to the learning data through tactile presentation or the like. Becomes

Next, an example will be described in which the technical idea according to the present disclosure is applied to the development of a practical recognizer. The operation of the guidance unit 220 according to the present embodiment can be extended in various ways, and the processing content of each step shown in FIG. 7 may be redefined. By utilizing the extensibility as described above, it is possible to divert the recognizer generated in the tutorial to the development of a practical recognizer thereafter.

In this case, the embedded device 10 according to the present embodiment may further include a reward providing unit 170, for example, as illustrated in FIG. The following is a specific example of the conversion to practical recognition development after the tutorial.

First, the operator completes the tutorial according to the flowchart shown in FIG. Next, the operator once again develops a practical recognizer according to the above flowchart. At this time, the processing in steps S1101, S1103, and S1112 is redefined as follows.

First, the process in step S1101 will be described. In step S1101, the guiding unit 220 searches the recognizer storage unit 260, and displays a list of recognizers generated in the tutorial on the development UI2. Here, when the operator selects any one of the recognizers, the learning / evaluation unit 230 uses the parameters such as the weight of the recognizer and enables the transfer learning.

Next, the processing in step S1103 will be described. In the above description using FIG. 7, the recognizer development is for the purpose of a tutorial, and, for example, learning data is collected after imposing restrictions such as limiting the recognition target to one hand of the operator. Was mentioned. On the other hand, in the present example, unlike the above-described tutorial, learning data may be collected from various environments and various data providers.

For example, suppose a case where the accuracy of the handsign classifier generated in the tutorial is to be further improved. In this case, for example, as shown in FIG. 18, the embedded device 10 is realized as a projector and installed in a public space such as an urban area or an event venue. Further, for example, a data provider is recruited in digital signage, and learning data is collected.

In the above case, in step S1112, the guide unit 220 guides the data provider U2 by controlling the embedded device 10 as a projector in order to collect predetermined learning data. At this time, the guiding unit 220 measures position information, environmental information such as temperature and humidity, etc., based on various information collected by the sensor 120, and determines the sex, age, and And the like. The guidance unit 220 according to the present embodiment can perform optimal guidance according to the context at the time of data collection and the data provider based on the results of the measurement and recognition as described above.

For example, in the situation illustrated in FIG. 16, when it is recognized that the installation location of the embedded device 10 is an event venue and that the data provider U2 is a male, the guiding unit 220 is configured to perform Is made to project the female avatar AV2 on the screen.

Next, the guiding unit 220 controls the female talent avatar AV2 to perform the Half-heart hand sign shown in FIG. 2, and instructs the data provider U2 to match the female talent avatar AV2 with the hand of the female talent avatar AV2. -Prompt to perform a heart hand sign. At this time, the embedded device 10 can photograph the data provider U2's Half-heart handsign using the camera device provided in the sensor 120 and collect the learning data D1.

According to the above control by the guiding unit 220 according to the present embodiment, learning data by various data providers can be collected in various environments, and the generalization performance of a recognizer generated in a tutorial or the like is further improved. It becomes possible.

The guidance unit 220 may control the reward providing unit 170 and provide a reward to the data provider U2 so that the data provider U2 cooperates more actively in learning data collection.

For example, as in the example shown in FIG. 16, when collecting an image related to a Half-heart hand sign as learning data, the guiding unit 220 controls the reward providing unit 170 and sets a heart symbol together with the avatar AV2 of the female talent. The formed two-shot photograph may be provided to the data provider U2 as a reward RW.

The provision to the data provider according to the present embodiment is not limited to the image as described above, food such as confectionery, physical banknotes such as gift certificates, digital currency such as virtual currency, music and video, etc. It may be multimedia content.

As described above, the modified examples according to an embodiment of the present disclosure have been described with reference to specific examples. As described above, the information processing system according to an embodiment of the present disclosure can be flexibly deformed according to specifications and operations.

<2. Example of hardware configuration>
Next, a hardware configuration example of the information processing device 20 according to an embodiment of the present disclosure will be described. FIG. 19 is a block diagram illustrating a hardware configuration example of the information processing device 20 according to an embodiment of the present disclosure. Referring to FIG. 19, the information processing device 20 includes, for example, a processor 871, a ROM 872, a RAM 873, a host bus 874, a bridge 875, an external bus 876, an interface 877, an input device 878, and an output device 879. , A storage 880, a drive 881, a connection port 882, and a communication device 883. Note that the hardware configuration shown here is an example, and some of the components may be omitted. Further, components other than the components shown here may be further included.

(Processor 871)
The processor 871 functions as, for example, an arithmetic processing device or a control device, and controls the entire operation of each component or a part thereof based on various programs recorded in the ROM 872, the RAM 873, the storage 880, or the removable recording medium 901. .

(ROM 872, RAM 873)
The ROM 872 is a means for storing a program read by the processor 871, data used for calculation, and the like. The RAM 873 temporarily or permanently stores, for example, a program read by the processor 871 and various parameters that appropriately change when the program is executed.

(Host bus 874, bridge 875, external bus 876, interface 877)
The processor 871, the ROM 872, and the RAM 873 are mutually connected, for example, via a host bus 874 capable of high-speed data transmission. On the other hand, the host bus 874 is connected to, for example, an external bus 876 having a relatively low data transmission speed via a bridge 875. The external bus 876 is connected to various components via an interface 877.

(Input device 878)
As the input device 878, for example, a mouse, a keyboard, a touch panel, a button, a switch, a lever, and the like are used. Further, as the input device 878, a remote controller (hereinafter, remote controller) capable of transmitting a control signal using infrared rays or other radio waves may be used. Further, the input device 878 includes a voice input device such as a microphone.

(Output device 879)
The output device 879 transmits acquired information to the user, such as a display device such as a CRT (Cathode Ray Tube), an LCD or an organic EL, an audio output device such as a speaker or a headphone, a printer, a mobile phone, or a facsimile. It is a device that can visually or audibly notify the user. In addition, the output device 879 according to the present disclosure includes various vibration devices capable of outputting a tactile stimulus.

(Storage 880)
The storage 880 is a device for storing various data. As the storage 880, for example, a magnetic storage device such as a hard disk drive (HDD), a semiconductor storage device, an optical storage device, a magneto-optical storage device, or the like is used.

(Drive 881)
The drive 881 is a device that reads information recorded on a removable recording medium 901 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, or writes information on the removable recording medium 901.

(Removable recording medium 901)
The removable recording medium 901 is, for example, a DVD medium, a Blu-ray (registered trademark) medium, an HD DVD medium, various semiconductor storage media, and the like. Of course, the removable recording medium 901 may be, for example, an IC card on which a non-contact type IC chip is mounted, or an electronic device.

(Connection port 882)
The connection port 882 is, for example, a port for connecting an external connection device 902 such as a USB (Universal Serial Bus) port, an IEEE1394 port, a SCSI (Small Computer System Interface), an RS-232C port, or an optical audio terminal. is there.

(External connection device 902)
The external connection device 902 is, for example, a printer, a portable music player, a digital camera, a digital video camera, or an IC recorder.

(Communication device 883)
The communication device 883 is a communication device for connecting to a network. For example, a communication card for a wired or wireless LAN, Bluetooth (registered trademark), or WUSB (Wireless USB), a router for optical communication, and an ADSL (Asymmetric Digital) Subscriber Line) or a modem for various communications.

<3. Summary>
As described above, the information processing device 20 according to an embodiment of the present disclosure includes the guiding unit 220 that guides collection of learning data used for learning of a recognizer. In addition, the guiding unit 220 according to an embodiment of the present disclosure dynamically guides an operator to collect data so that diversity of sensing data related to a recognition target is reported. According to such a configuration, it is possible to guide the operator to collect suitable learning data.

Although the preferred embodiments of the present disclosure have been described above in detail with reference to the accompanying drawings, the technical scope of the present disclosure is not limited to such examples. It is apparent that a person having ordinary knowledge in the technical field of the present disclosure can come up with various changes or modifications within the scope of the technical idea described in the claims. Is naturally understood to belong to the technical scope of the present disclosure.

効果 In addition, the effects described in this specification are merely illustrative or exemplary and not restrictive. That is, the technology according to the present disclosure can exhibit other effects that are obvious to those skilled in the art from the description in the present specification, in addition to or instead of the above effects.

In addition, a program for causing hardware such as a CPU, a ROM, and a RAM built in the computer to exhibit the same function as the configuration of the information processing device 20 can be created. Possible non-transitory recording media can also be provided.

Note that the following configuration also belongs to the technical scope of the present disclosure.
(1)
A guiding unit for guiding the collection of learning data used for learning the recognizer,
With
The guiding unit dynamically guides data collection by the operator so that diversity of sensing data related to the recognition target is ensured,
Information processing device.
(2)
The guidance unit imposes a constraint on the collection of the learning data to the operator, and dynamically performs the data collection by the operator so as to ensure the diversity of the sensing data within the range of the constraint. The information processing device according to (1), which guides the user.
(3)
The guiding unit determines whether or not the diversity sufficient for generalization of the recognizer is secured. If the diversity is not secured, the sensing data of a pattern that is assumed to be in shortage is collected. The information processing apparatus according to (2), wherein the data collection is guided by the operator so as to be performed.
(4)
The guide unit determines whether or not the collected sensing data complies with the constraint.If the constraint is not complied with, a warning to the effect that the constraint is not complied with is issued to the operator. The information processing device according to (2) or (3), wherein the information is notified.
(5)
The operator is a first-time scholar involved in learning the recognizer,
The guiding unit dynamically guides the data collection by the beginner in the tutorial related to the collection of the learning data.
The information processing device according to any one of (2) to (4).
(6)
The information processing apparatus according to any one of (2) to (5), wherein the restriction includes a restriction that limits the recognition target to a part or behavior of the operator's body.
(7)
The guidance unit according to any one of (1) to (6), wherein the guidance unit controls presentation to the operator relating to recognition accuracy of the recognizer that has performed learning using the collected sensing data. Information processing device.
(8)
The information processing device according to (7), wherein the guidance unit causes a transition of recognition accuracy of the recognizer to be presented.
(9)
The information processing device according to (8), wherein the guidance unit causes the collected sensing data to be presented in chronological order together with the recognition accuracy of the recognizer.
(10)
The information processing according to any one of (1) to (9), wherein the guidance unit displays, in real time, a recognition result of the recognition target by the recognizer that has performed learning using the collected sensing data. apparatus.
(11)
The guidance unit according to any one of (1) to (10), wherein the guidance unit visualizes the collected sensing data, displays the sensing data in real time to the operator, and performs visual guidance so as to ensure the diversity. An information processing device according to any one of the above.
(12)
The recognition target includes a body sign or a gesture,
The guidance unit displays a guidance object that guides the collection of the sensing data effective for securing the diversity, and instructs the operator to perform a predetermined operation on the guidance object.
The information processing device according to (11).
(13)
The information processing apparatus according to any one of (1) to (12), wherein the guidance unit controls provision of a reward to the operator.
(14)
An output control unit that controls a user interface according to the guidance of the data collection based on the control by the guidance unit.
Further comprising,
The information processing apparatus according to any one of (1) to (13).
(15)
The information processing device according to (14), wherein the output control unit further controls a user interface for the operator to design the recognizer.
(16)
A learning unit that performs learning related to the recognizer using the collected sensing data,
Further comprising,
The information processing device according to any one of (1) to (15).
(17)
The processor guiding the collection of learning data used for learning the recognizer;
Including
The guiding is to dynamically guide data collection by the operator so that diversity of sensing data related to the recognition target is ensured,
An information processing method, further comprising:
(18)
Computer
A guiding unit for guiding the collection of learning data used for learning the recognizer,
With
The guiding unit dynamically guides data collection by the operator so that diversity of sensing data related to the recognition target is ensured,
A program for functioning as an information processing device.

Reference Signs List 10 embedded device 110 output unit 120 sensor 130 recognition execution unit 140 drive unit 150 learning data collection unit 170 reward providing unit 20 information processing device 210 output control unit 220 guidance unit 222 labeling unit 224 data analysis unit 230 learning / evaluation unit 240 learning data Set storage unit 250 Evaluation data set storage unit 260 Recognition device storage unit 30 Display device

Claims (18)

A guiding unit for guiding the collection of learning data used for learning the recognizer,
With
The guiding unit dynamically guides data collection by the operator so that diversity of sensing data related to the recognition target is ensured,
Information processing device. The guidance unit imposes a constraint on the collection of the learning data to the operator, and dynamically performs the data collection by the operator so as to ensure the diversity of the sensing data within the range of the constraint. The information processing apparatus according to claim 1, which guides. The guiding unit determines whether or not the diversity sufficient for generalization of the recognizer is secured. If the diversity is not secured, the sensing data of a pattern that is assumed to be in shortage is collected. The information processing device according to claim 2, wherein the data collection is guided by the operator so as to be performed. The guide unit determines whether or not the collected sensing data complies with the constraint.If the constraint is not complied with, a warning to the effect that the constraint is not complied with is issued to the operator. The information processing apparatus according to claim 2, wherein the information processing apparatus is notified. The operator is a first-time scholar involved in learning the recognizer,
The guiding unit dynamically guides the data collection by the beginner in the tutorial related to the collection of the learning data.
The information processing device according to claim 2. The information processing apparatus according to claim 2, wherein the constraint includes a constraint that limits the recognition target to a part or behavior of the operator's body. 2. The information processing device according to claim 1, wherein the guidance unit controls presentation to the operator regarding recognition accuracy of the recognizer that has performed learning using the collected sensing data. 3. The information processing apparatus according to claim 7, wherein the guidance unit causes a transition of the recognition accuracy of the recognizer to be presented. The information processing apparatus according to claim 8, wherein the guidance unit causes the collected sensing data to be presented in a time series together with the recognition accuracy of the recognizer. 2. The information processing apparatus according to claim 1, wherein the guidance unit displays, in real time, a recognition result of the recognition target by the recognizer that has performed learning using the collected sensing data. The information processing apparatus according to claim 1, wherein the guidance unit visualizes the collected sensing data and displays the collected sensing data to the operator in real time, and performs visual guidance so as to ensure the diversity. . The recognition target includes a body sign or a gesture,
The guidance unit displays a guidance object that guides collection of the sensing data effective for securing the diversity, and instructs the operator to perform a predetermined operation on the guidance object.
The information processing device according to claim 11. The information processing apparatus according to claim 1, wherein the guidance unit controls provision of a reward to the operator. An output control unit that controls a user interface according to the guidance of the data collection based on the control by the guidance unit.
Further comprising,
The information processing device according to claim 1. The information processing apparatus according to claim 14, wherein the output control unit further controls a user interface for the operator to design the recognizer. A learning unit that performs learning related to the recognizer using the collected sensing data,
Further comprising,
The information processing device according to claim 1. The processor guiding the collection of learning data used for learning the recognizer;
Including
The guiding is to dynamically guide data collection by the operator so that diversity of sensing data related to the recognition target is ensured,
An information processing method, further comprising: Computer
A guiding unit for guiding the collection of learning data used for learning the recognizer,
With
The guiding unit dynamically guides data collection by the operator so that diversity of sensing data related to the recognition target is ensured,
A program for functioning as an information processing device.

Images