JP2003222514A - Sensor configuration determination device, sensor configuration determination method, sensor configuration determination program and recording medium for program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To determine the kind of a sensor for recognizing a wearable action and the mounting position on a human body, while taking automatically a mounting load or a device cost into consideration. <P>SOLUTION: Model accuracy, comfortability to a mounting person, and a cost are evaluated respectively on reference to a sensor information database 101 wherein each data of the kinds, costs, parameters mounting spots of various sensors, and ID are stored, a learning action database 103 wherein learning action data are recorded, and an evaluation action database 105 wherein action data for model evaluation are recorded, and overall evaluation is performed by a model overall evaluation means 109, to thereby determine the kind of the optimum sensor for recognizing the wearable action and the mounting position on the human body. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sensor configuration determination which can automatically determine the type of sensor and the mounting position while taking into consideration the burden of mounting and the cost of the device based on the information of the sensor and the human body. The present invention relates to an apparatus, a sensor configuration determining method, a sensor configuration determining program, and a recording medium of the program.

[0002]

2. Description of the Related Art At present, in a limited space, it has become possible to grasp human motions from images taken by cameras and sensor information of a motion capture system. On the other hand, in the case where the space is not limited, in recent years, due to the miniaturization and weight reduction of the acceleration sensor and the like, and the development of the recognition algorithm, it is becoming possible to wear the sensor on the human body and judge the action.

Motion recognition without such space limitation (=
As an example of wearable motion recognition), Japanese Patent Laid-Open No. 11-4
As described in Japanese Patent No. 4220, "walking" from an acceleration sensor to control a cardiac pacemaker in detail.
Those that recognize behavior such as "run" and "stair climb",
Kobayashi, K .: Measurement method à la carte (14) Measurement of human body and its application Measurement and evaluation of working posture / motion, Journal of Japan Society of Agricultural Machinery, Vol.60, No.4, pp90-94, 1998. There are devices that evaluate the posture of farm workers.

[0004]

However, when developing a wearable motion recognition method and a device for the wearable motion recognition method that many people actually use, it is not only to increase the motion recognition rate but also to You must consider the burden and costs. Regarding the determination of the type of sensor and the mounting position of the human body to solve these problems,
There was no choice but to repeat trial and error based on the expert knowledge of the target motion.

The present invention has been made to solve the above-mentioned problems, and an object thereof is to automatically consider the burden of mounting and the cost of the device based on the information of the sensor and the human body. It is to realize a mechanism that determines the type and mounting position.

[0006]

A sensor configuration determining device according to the present invention includes a "sensor information database" in which data of various types of sensors, prices, parameters, and mounting locations is stored together with an ID, and the sensor information database. "Sensor combination information determining means" for determining sensor combination candidates to be used, "learning operation database" in which operation data is recorded for learning, and sensor combination determined by the sensor combination information determining means. "Model creating means" for creating a model by referring to the learning operation database by inputting information, and "evaluation operation database" in which operation data is recorded for model evaluation,
“Model accuracy evaluation means” for evaluating the accuracy of model learning with reference to the evaluation operation database for the model created by the model creation means, and sensor combination information determined by the sensor combination information determination means. "Comfort evaluation means" for evaluating the comfort of the user while referring to the sensor information database as an input;
"Cost evaluation means" for evaluating the cost of the sensor to be used while referring to the sensor information database by inputting the sensor combination information determined by the sensor combination information determination means, the model accuracy evaluation means, the comfort evaluation Means for performing a comprehensive evaluation of the model using the results of the cost evaluation means and the cost evaluation means, and a sensor when the evaluation results of the model comprehensive evaluation means do not satisfy the prescribed conditions. The present invention is characterized by including "sensor combination narrowing means" for narrowing down the combination candidates and reducing the sensor combination candidates to be evaluated next.

Further, in the sensor configuration determining method of the present invention, the "sensor information database" in which the types, prices, parameters, and mounting location data of various sensors are stored together with the ID, and the operation data are recorded for learning. It is characterized in that a "learning operation database" and an "evaluation operation database" in which operation data is recorded for model evaluation are prepared in advance, and the following processes are performed using these. That is, referring to the sensor information database, a “sensor combination information determining process” for determining a sensor combination candidate to be used, and the learning operation database using the sensor combination information determined in the sensor combination information determining process as an input. A "model creation process" for creating a model with reference, and a "model accuracy evaluation process" for evaluating the accuracy of model learning with reference to the evaluation operation database for the model created in the model creation process, A "comfort evaluation process" for evaluating the comfort of the user while inputting the sensor combination information determined in the sensor combination information determination process and referring to the sensor information database, and the sensor combination information determination process. The sensor information database described above is used as input. , The cost evaluation process for evaluating the cost of the sensor to be used, and the model accuracy evaluation process, the comfort evaluation process, and the cost evaluation process are used to perform a comprehensive evaluation of the model. "Comprehensive evaluation process" and the evaluation result of the model comprehensive evaluation process does not satisfy the specified conditions,
It has a "sensor combination narrowing down process" for narrowing down the sensor combination candidates and then reducing the sensor combination candidates to be evaluated.

The present invention also relates to a program for causing a computer to execute the above processing, and a recording medium of the program.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, reference numeral 101 is a sensor information database in which data of various types of sensors, prices, parameters, and mounting locations are stored together with IDs. For example, the sensor type may be an acceleration sensor, a magnetic sensor, or the like, the parameters may be the X-axis direction, the Y-axis direction, or the like, and the mounting location may be the right elbow, the head, or the like. Of course, it is not limited to these.

Reference numeral 102 is a sensor information database 101.
Is a sensor combination information determining means for determining a sensor combination candidate to be used, and the sensor information database 1
The combination of IDs recorded in 01 is repeatedly output. By adding, deleting, or replacing one ID with respect to the combination output last time, a combination to be evaluated next is created and output. Only in the initial state, a combination example prepared separately is read and output as it is.

Reference numeral 103 is a learning motion database in which motion data is recorded for learning. Specifically, the sensor data when the target motion (eg, tennis swing) is measured by some sensing means is recorded.

Reference numeral 104 is a sensor combination information determining means 10
With the sensor combination information determined in 2 as input,
The model creating means creates a model while referring to the learning operation database 103. The model created here is an operation model obtained from the learning operation database 103 corresponding to the combination of sensors, and is represented by a column of information such as an operation name and sensor data corresponding to the operation. The obtained data may be processed according to the model accuracy evaluation method to be performed next. For example, a recognition algorithm such as DP matching may correspond to this, and the coefficient information may be output.

Reference numeral 105 denotes an evaluation operation database in which operation data is recorded for model evaluation. In particular,
Similar to the learning motion database 103, sensor data when a target motion (eg, tennis swing) is measured by some sensing means is recorded.

Reference numeral 106 is a model accuracy evaluation means for evaluating the accuracy of model learning for the model created by the model creation means 104 while referring to the evaluation operation database 105. That is, it is evaluated how much the model created by the model creating means 104 matches the motion data recorded in the evaluation motion database 105.

Reference numeral 107 denotes a sensor combination information determining means 10
With the sensor combination information determined in 2 as input,
It is a comfort evaluation means for evaluating the comfort of the user while referring to the sensor information database 101. For example, the total number of attachment points is the evaluation standard. Alternatively, if even sensor weight data is recorded in the sensor information database 101, the total sensor weight can also be used as an evaluation standard.

Reference numeral 108 is a sensor combination information determining means 10
With the sensor combination information determined in 2 as input,
It is a cost evaluation means for evaluating the cost of the sensor while referring to the sensor information database 101. Evaluation is performed based on the price data of each sensor recorded in the sensor information database 101.

Reference numeral 109 is a model comprehensive evaluation means for performing a comprehensive evaluation of the model using the evaluation results of the model accuracy evaluation means 106, the comfort evaluation means 107, and the cost evaluation means 108. For example, evaluation is performed using an evaluation function or the like in which a weighting factor is added to the scalar value output by these three evaluation means.

Reference numeral 110 narrows down the sensor combination candidates when the evaluation result of the model comprehensive evaluation means 109 does not satisfy a prescribed condition (for example, threshold value processing for the evaluation function), and then evaluates it. It is a sensor combination narrowing-down means for reducing the sensor combination candidates.

FIG. 2 is a flow chart showing an embodiment of the present invention. FIG. 3 shows the sensor information database 10
1, an example of the learning operation database 103,
FIG. 5 is an example of the evaluation operation database 105, and FIG.
An example of model accuracy evaluation is shown. Hereinafter, the processing mode of the configuration shown in FIG. 1 will be described with reference to FIGS. 2 to 6.

The sensor combination information determining means 102 refers to the sensor information database 101 and determines the sensor combination candidate to be used. In the initial state, the initial input data is read and output as it is, but thereafter, the combination of IDs recorded in the sensor information database 101 is repeatedly output. One I for the previously output combination
By adding, deleting, and replacing D, a combination to be evaluated next is created and output (step 201).

In the sensor information database 101, for example, as shown in FIG. 3, an ID for identifying each sensor item, a sensor type, a price, a parameter, a mounting position on the human body, Data such as weight is stored. In the initial state, the sensor combination information determination means 102 outputs a predetermined combination of sensors such as (a, d, h), but after the second time,
Add one ID like (a, d, h, m) or delete one ID like (a, d).
Create a combination such as exchanging one ID such as (b, d, h).

An arbitrary algorithm can be used to determine the order in which the combination is created. Depending on the case, you may make it randomly. However, in order to make the search efficient, combinations that are estimated not to be evaluated in advance by the sensor combination narrowing-down means 110 described later will be excluded. If the number of sensor IDs is small, make all combinations and
It is also possible to evaluate each model sequentially.

The model creating means 104 receives the sensor combination information determined by the sensor combination information determining means 102 as an input and creates a model while referring to the learning operation database 103 (step 202).

In the learning operation database 103, as shown in FIG. 4A, for example, actual measurement data 1 to M by the sensor are stored for each operation 1 to L. here,
Motion is explained as an example of a tennis swing. Motion 1
Is the forehand stroke, the movement 2 is the backhand stroke, and so on. Further, each measurement data is composed of time-series data of sensor data actually measured by all the sensors for the corresponding operation, as measurement data 1 shown in FIG. 4 (B). In this example, the measurement data is M
Collected in pairs. This is because various movements can be considered even for tennis forehand strokes, for example.

The model accuracy evaluation means 106 evaluates the accuracy of model learning for the model created by the model creation means 104 with reference to the evaluation operation database 105. That is, it is evaluated how much the model created by the model creating means 104 matches the motion data recorded in the evaluation motion database 105, and the evaluation result is output as a scalar value or the like (step 20).
3).

In the evaluation operation database 105, for example, as shown in FIG. 5A, actual measurement data 1 to N for evaluation by sensors are stored for each operation 1 to L. Since the measurement data of the same format is originally created for learning and evaluation, a basic structure of the evaluation operation database 105 is the learning operation database 10
There is no difference in 3. The table of the evaluation measurement data also has the same data structure, for example, the evaluation measurement data 1 shown in FIG. However, the number of measured data (N) is not always the same, and the number of rows and values in the measured data table are different. The number of operations L and the number of fields (columns) of the measurement data table are the same for learning and evaluation.

An example of accuracy evaluation performed by the model accuracy evaluation means 106 will be described with reference to FIG. For example, it is assumed that the sensor combination determined by the sensor combination information determination means 102 is b and x. Evaluation operation database 1
The sensor data of b and the sensor data of x are extracted from one evaluation measurement data of 05, and the operation is made unknown, and DP matching with all the measurement data in the learning operation database 103 is performed, and the similarity is obtained. Calculate the degree. Of course, the measurement data referred to in the learning operation database 103 is also b (acceleration, X axis, neck) and x (angular velocity, Z).
Measurement data for sensor combinations of shaft and lower leg.

As is well known, DP matching is a method using dynamic programming and is a method of calculating how similar two pieces of time series data are. The time axis of the time-series data is expanded / contracted non-linearly, the two most time-series data are transformed so that they are similar, and finally the similarity is calculated.

In the example of the matching result shown in FIG. 6, the similarity between the evaluation measurement data i and the measurement data 1 of the operation 1 in the learning operation database 103 is 32, and the same operation 1
, And the similarity with the measurement data 3 of action 3 is 52. Note that FIG.
In order to make it easier to understand, we have reduced the number of motion models,
An example of L = 3 and M = 3 is shown. The smaller the similarity value, the more similar the two data are. Therefore, the evaluation measurement data i that the operation is unknown is
Is judged to be the most similar to the movement 2, the measurement data 2 (similarity = 11), and the model accuracy evaluation means 106 recognizes that the movement of the evaluation measurement data i is the movement 2.

The recognition of the operation by such DP matching is performed for all the evaluation measurement data in the evaluation operation database 105, and the accuracy is evaluated by the ratio of the correct operation and the incorrect operation.

That is, the model accuracy evaluation means 106 is
The above-mentioned DP matching is performed on each measurement data for evaluation in the evaluation operation database 105, assuming that the operation is unknown, and the operation is specified. Then, the accuracy is obtained by checking whether the answer is correct. For example, if there are 100 pieces of evaluation measurement data in the evaluation movement database 105 and 80 pieces of those pieces of data can correctly recognize the movement, the accuracy is 80%.

The comfort evaluating means 107 receives the sensor combination information determined by the sensor combination information determining means 102 as an input and evaluates the comfort of the user while referring to the sensor information database 101. For example, the total number of attachment points is the evaluation standard. Alternatively, as in the example of FIG. 3, if even sensor weight data is recorded in the sensor information database 101, the total sensor weight is also used as an evaluation criterion. The evaluation result is output as a scalar value or the like (step 204).

The cost evaluation unit 108 receives the sensor combination information determined by the sensor combination information determination unit 102 as an input and evaluates the cost of the sensor while referring to the sensor information database 101. Evaluation is performed based on the price data of each sensor recorded in the sensor information database 101. The evaluation result is output as a scalar value or the like (step 205).

The model comprehensive evaluation means 109 uses the evaluation results of the model accuracy evaluation means 106, the comfort evaluation means 107, and the cost evaluation means 108 to perform a comprehensive evaluation of the model. For example, evaluation is performed using an evaluation function or the like in which a weighting factor is added to the scalar value output by these three evaluation means (step 206). Model comprehensive evaluation means 1
It is determined whether the evaluation result of 09 satisfies a specified condition (for example, threshold processing for the evaluation function),
If satisfied, the sensor combination is output as a promising candidate (step 207). Alternatively, the model evaluation may be repeated a predetermined number of times with respect to various sensor combinations, and finally the sensor combination having the best evaluation value or the sensor combinations having the highest k of evaluation values (k is a specified value) may be output. .

The sensor combination narrowing means 110 narrows down the sensor combination candidates when the evaluation result of the model comprehensive evaluation means 109 does not satisfy the specified conditions, and reduces the sensor combination candidates to be evaluated next. (Step 208).

For example, the sensor combination S1 (a,
d, h), S2 (a, d, i), S3 (a, d, i,
m), S4 (a, i, m), and S5 (b, i, m) have been evaluated. Among these combinations, we exclude those for which the evaluation value of cost or comfort exceeds the upper limit. If S3 and S5 exceed the upper limit, exclude them. Next, among the remaining combinations, n pieces (n is a specified value) having the highest precision are left and the rest are excluded. For example, if the prescribed value n is n = 2 and the accuracy of the combination of S2 and S4 is good, the sensor combination narrowing means 110 uses this information as the sensor combination information determining means 10.
Tell 2.

The sensor combination information determining means 102 determines S2
The combination of the sensors of (a, d, i) or S4 (a, i, m) is used as a seed, and one ID is further added, deleted, replaced, or the like to create a combination to be evaluated next. Of course, if the created combination has already been evaluated, another combination is created. Thus, when the number of sensor IDs is large, a practically effective combination of sensors can be automatically determined without performing a brute force combination evaluation.

The above processing can be realized by a computer and a software program, and the program can be stored in an appropriate recording medium such as a computer-readable portable medium memory, semiconductor memory, hard disk or the like. . Further, the program can be downloaded from another computer via a communication line, and can be installed and executed on the downloaded computer.

[0040]

As described above, according to the present invention, it is possible to realize wearable motion recognition that reduces the burden on the wearer and reduces the cost of the device. This makes it possible to significantly reduce the huge amount of trial and error time that was required in the past, without requiring specialized knowledge of the target motion.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of the present invention.

FIG. 2 is a flowchart showing an embodiment of the present invention.

FIG. 3 is a diagram showing an example of a sensor information database.

FIG. 4 is a diagram showing an example of a learning operation database.

FIG. 5 is a diagram showing an example of an evaluation operation database.

FIG. 6 is a diagram showing an example of model accuracy evaluation.

[Explanation of symbols]

101 sensor information database 102 sensor combination information determining means 103 Learning operation database 104 model creation means 105 Operation database for evaluation 106 model accuracy evaluation means 107 Comfort evaluation means 108 Cost evaluation means 109 Model comprehensive evaluation means 110 Sensor combination narrowing means

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Satoshi Ishibashi             2-3-1, Otemachi, Chiyoda-ku, Tokyo             Inside Telegraph and Telephone Corporation F term (reference) 2F069 AA04 AA06 BB40 DD30 GG01                       GG04 GG41 GG58 GG65 HH30

Claims (5)

[Claims] Claim: What is claimed is: 1. A sensor configuration determining device for determining a type of a sensor to be attached to a human and a mounting position for recognizing a human motion, the type, price, parameter,
A sensor information database that holds data of the mounting location; a sensor combination information determination unit that refers to the sensor information database to determine a sensor combination candidate to be used; and a learning operation database in which operation data is recorded for learning. A model creating means for creating a model with reference to the learning operation database by inputting the sensor combination information determined by the sensor combination information determining means, and an evaluation operation database in which operation data is recorded for model evaluation. The model accuracy evaluation means for evaluating the accuracy of model learning with reference to the evaluation operation database and the sensor combination information determined by the sensor combination information determination means are input to the model created by the model creation means. As referring to the sensor information database A comfort evaluation means for evaluating the comfort of the user, and a cost evaluation means for evaluating the cost of the sensor to be used while referring to the sensor information database with the sensor combination information determined by the sensor combination information determination means as an input , Model accuracy evaluation means, comfort evaluation means, and cost evaluation means, and a model comprehensive evaluation means for performing a comprehensive evaluation of the model. Various sensor combinations determined by the sensor combination information determination means. A sensor configuration determining device, characterized in that a candidate of a sensor to be mounted on a human and a candidate of a mounting position are determined based on an evaluation result of the model comprehensive evaluation means for the candidate. 2. The sensor configuration determining device according to claim 1, wherein the sensor combination information determining means determines the sensor combination determined when the evaluation result of the model comprehensive evaluating means does not satisfy a specified condition. A sensor configuration determining device comprising a sensor combination narrowing-down unit that narrows down candidates and then reduces sensor combination candidates to be evaluated next. 3. A sensor configuration determining method for determining a type of a sensor to be attached to a human and a mounting position for recognizing a motion of the human.
With reference to the sensor information database storing the data of the mounting location, the sensor combination information determining process of determining the sensor combination candidate to be used, and the sensor combination information determined in the sensor combination information determining process are input.
A model creation process in which a model is created while referring to a learning motion database in which motion data for learning is recorded,
For the model created in the model creation process, a model accuracy evaluation process for evaluating the accuracy of model learning with reference to an evaluation operation database in which operation data for model evaluation is recorded, and a sensor combination information determination process Using the determined sensor combination information as an input, the comfort evaluation process of evaluating the comfort of the user while referring to the sensor information database, and the sensor combination information determined in the sensor combination information determination process as an input,
The model is comprehensively evaluated by referring to the sensor information database and the results of the cost evaluation process for evaluating the cost of the sensor to be used, the model accuracy evaluation process, the comfort evaluation process, and the cost evaluation process. A sensor configuration characterized by including a model comprehensive evaluation process, and outputting sensor combination information having a good evaluation result by the model comprehensive evaluation means by repeating the process from the sensor combination information determination process to the model comprehensive evaluation process. How to decide. 4. A sensor configuration determining program for determining a type of a sensor to be attached to a human and a mounting position for recognizing a human motion, the data of various sensor types, prices, parameters and mounting locations. By referring to the stored sensor information database, the sensor combination information determining process for determining the sensor combination candidate to be used, and the sensor combination information determined in the sensor combination information determining process are input, and operation data for learning is input. A model creation process for creating a model with reference to the recorded learning motion database, and an evaluation motion database with motion data recorded for model evaluation with respect to the model created by the model creation process. Determined in the model accuracy evaluation process for evaluating the accuracy of model learning and the sensor combination information determination process The sensor combination information is input to the sensor information database, the comfort evaluation process for evaluating the user's comfort, and the sensor combination information determined in the sensor combination information determination process are input to the sensor. A cost evaluation process for evaluating the cost of the sensor to be used while referring to the information database, the model accuracy evaluation process, the comfort evaluation process,
Using the results of each of the cost evaluation processes, the model comprehensive evaluation process for performing a comprehensive evaluation of the model, and the process from the sensor combination information determination process to the model comprehensive evaluation process are repeated. A sensor configuration determination program for causing a computer to execute a process of outputting good sensor combination information. 5. A recording medium for recording a sensor configuration determining program for determining a type of a sensor to be attached to a human and a mounting position for recognizing a human motion, the type, price and parameters of various sensors. , Learning with reference to a sensor information database storing the data of the mounting location, sensor combination information determination processing for determining sensor combination candidates to be used, and sensor combination information determined in the sensor combination information determination processing Model creating process for creating a model by referring to a learning motion database in which motion data is recorded for evaluation, and an evaluation motion in which motion data is recorded for model evaluation with respect to the model created in the model creating process. Model accuracy evaluation processing for evaluating the accuracy of model learning with reference to a database, and the sensor combination information determination The sensor combination information determined in the fixed process is input, the comfort evaluation process for evaluating the comfort of the user and the sensor combination information determined in the sensor combination information determination process are input while referring to the sensor information database. As a result, by referring to the sensor information database, the cost evaluation process for evaluating the cost of the sensor to be used, and the results of the model accuracy evaluation process, the comfort evaluation process, and the cost evaluation process are used to comprehensively evaluate the model. The computer is caused to execute a model comprehensive evaluation process for performing the above, and a process for outputting sensor combination information having a good evaluation result by the model comprehensive evaluation process by repeating the process from the sensor combination information determination process to the model comprehensive evaluation process. A sensor structure characterized by recording a program for Decision for the program of the recording medium.