JP2017142736A - Gesture recognition device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to recognize a gesture of a hand with high accuracy even when a wrist band is re-worn to a wrist or a direction of the wrist is changed.SOLUTION: In a learning mode, a gesture recognition device is configured to: detect an average value and maximum value among three pieces of pressure sensors 31, 32 and 33 of detection values to be obtained by the three sensors when performing an input operation of a gesture of a recognition object; and store a time average of the time-wise detection as an amount of characteristic. In a usage mode, the gesture recognition device is configured to: firstly perform the input operation of a reference gesture; detect an average value and maximum value among the pressure sensors 31, 32 and 33 at a peak time of the detection values of the pressure sensors; obtain a difference between corresponding amounts of characteristic and these values as a correction value; subsequently detect an average value and maximum value among the pressure sensors 31, 32 and 33 at a peak time of the detection values of the pressure sensors when the input operation of an arbitrary gesture is performed; correct the average value and maximum value by the correction value; and recognize the gesture on the basis of post-corrected values.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a gesture recognition apparatus, method, and program for recognizing hand gestures.

  In general, a dedicated remote controller, a portable terminal such as a smartphone, and the like are used as means for operating personal devices such as a wearable device, an air conditioner, and a television receiver. However, in situations where you can only take a few moments away from the main activities, such as jogging, cooking, or walking in a crowded city, take out the remote controller or mobile device and open the operation panel. In many cases, it is difficult to perform a button operation or a touch operation while visually recognizing.

  Therefore, recently, many attempts have been made to operate wearable devices and personal devices by attaching wristband devices to the wrist and recognizing hand gestures. When this type of device is used, it is possible to operate the operation target without temporarily interrupting the main action, which is very convenient.

  Various techniques have been proposed for recognizing hand gestures, such as using computer vision and measuring surface myoelectric potential. In recent years, wrist shape changes caused by finger movements have been proposed. A new method has been proposed for detecting the above using a plurality of pressure sensors (see Non-Patent Document 1, for example).

  Since most of the muscles that move the fingers are in the forearm, the tendon that moves the forearm muscles and the fingers moves by the movement of the fingers, and as a result, the shape of the wrist changes. This change is detected as pressure dispersion received by the wristband using a plurality of pressure sensors provided in the wristband. This method is very promising because the sensor itself is low in power consumption and inexpensive, and can be easily attached compared to a myoelectric sensor that requires a conductive paste to improve accuracy.

Dementyev, A. and Paradiso, J. A .: WristFlex: Low-Power Gesture Input with Wrist-Worn Pressure Sensors, In Proceeding of User Interface Software and Tchnology 2014, pp.161-166, ACM (2014).

  However, the method of detecting the wrist shape change as the pressure dispersion received by the wristband using multiple pressure sensors equipped on the wristband is when the wristband is remounted on the wrist or the direction of the arm is changed. In some cases, the problem that the recognition performance is greatly reduced has been reported. This is because the wearing position and the tightening condition of the band change by re-mounting the wrist band, or the wrist shape changes when the direction of the arm is changed.

  The present invention has been made paying attention to the above circumstances, and its purpose is to recognize a hand gesture with high accuracy even when the wristband is re-attached to the wrist or the direction of the wrist is changed. To provide a gesture recognition apparatus, method, and program.

  In order to solve the above-described problem, a first aspect of the present invention includes means for receiving time-series data indicating detected pressure values from a plurality of pressure sensors arranged in a wristband, and provides a reference gesture. A peak value is detected from time-series data received from each of the plurality of pressure sensors during a first period during which an input operation is performed, and an average value and a maximum value between the time-series data at the time of detection of the peak value Is detected, and a correction value is generated based on at least one of the detected average value and maximum value. Then, in a second period in which an arbitrary gesture input operation is performed following the first period, at least one of an average value and a maximum value between the time series data respectively received from the plurality of pressure sensors is obtained. Detecting, correcting at least one of the detected average value and maximum value based on the correction value, and recognizing the arbitrary gesture based on at least one of the corrected average value and maximum value. It is a thing.

  The second aspect of the present invention further includes a storage unit that stores feature amounts corresponding to each of a plurality of gestures to be recognized. Then, the correction is performed based on at least one of an average value and a maximum value between the time-series data at the time of detection of the peak value and a feature amount corresponding to the reference gesture stored in the storage unit. A value is generated.

  The third aspect of the present invention further includes learning means for registering the feature amount of the gesture. Then, by this learning means, for each of the plurality of gestures to be recognized, at least one of an average value and a maximum value between the time series data received from the plurality of pressure sensors during the input operation period is detected. , Calculating an average in the input operation period for at least one of the detected average value and maximum value, and storing the calculated average in the storage means as a feature amount corresponding to the corresponding gesture It is.

  According to a fourth aspect of the present invention, there is provided means for generating operation command information corresponding to an arbitrary gesture recognized by the means for recognizing the gesture, and sending the generated operation command information to a device to be operated. Further, it is provided.

  According to the first aspect of the present invention, an average value or maximum value between time-series data obtained by a plurality of pressure sensors is used, and subsequently, based on the average value or maximum value obtained by the reference gesture. The average value or maximum value of the input arbitrary gesture is corrected, and the arbitrary gesture is recognized based on the corrected value. For this reason, even when the wristband is remounted on the wrist or the direction of the arm is changed, it is possible to accurately recognize the hand gesture, and thereby it is possible to provide a gesture recognition device that can be used anytime and anywhere. .

  According to the second aspect of the present invention, the feature amount corresponding to each of the plurality of gestures to be recognized is stored in advance, and the average value between the time series data obtained during the input operation period of the reference gesture and A correction value is generated based on at least one of the maximum values and the feature amount corresponding to the stored reference gesture. Accordingly, it is possible to generate an accurate correction value.

  According to the third aspect of the present invention, the feature quantity for each gesture is generated and stored in the learning mode. For this reason, the feature amount for each gesture can be accurately set for each user, and thus the accuracy of correction and the recognition accuracy of the gesture can be increased.

  According to the fourth aspect of the present invention, the operation command information is generated according to the gesture recognition result, and the generated operation command information is sent to the operation target device. For this reason, it becomes possible to directly control the operation target device from the gesture recognition device.

  That is, according to the present invention, it is possible to provide a gesture recognition device, method, and program capable of accurately recognizing a hand gesture even when the wristband is remounted on the wrist or the direction of the wrist is changed.

Sectional drawing which shows the state which mounted | wore the wrist with the gesture recognition apparatus which concerns on one Embodiment of this invention. The block diagram which shows the function structure of the gesture recognition apparatus which concerns on one Embodiment of this invention. The flowchart which shows the recognition processing procedure and processing content by the gesture recognition apparatus shown in FIG. The figure which shows an example of the kind of gesture by hand. The figure which shows an example of the feature-value of the sensor data corresponding to the gesture shown in FIG. The figure which shows an example of the change of the sensor detection voltage when performing gesture by hand. The figure which showed the change of the detection voltage value of three sensors according to the kind of gesture when a gesture recognition apparatus is mounted | worn with a wrist on three different conditions. The figure which showed the peak value in the time-sequential direction of the detection voltage average value between three sensors when the three types of conditions shown in FIG. 7 were used as parameters, according to the type of gesture. The figure which showed the peak value in the time-sequential direction of the detection voltage maximum value between three sensors when the three types of conditions shown in FIG. 7 were used as a parameter according to the kind of gesture. The figure which showed the change of the detected voltage value of three sensors according to the kind of gesture when a gesture recognition apparatus is mounted | worn again to the wrist 3 times. The figure which showed the peak value in the time-sequential direction of the detection voltage average value between three sensors when the reattachment of 3 times shown in FIG. 10 was made into the parameter, according to the kind of gesture. The figure which showed the peak value in the time-sequential direction of the detection voltage maximum value between three sensors when the reattachment of 3 times shown in FIG. 10 was made into the parameter, according to the kind of gesture. The recognition accuracy rate according to the conventional method for recognizing a gesture using the detected voltages of the three sensors as they are and the peak value in the time series direction of the average value / maximum value of the three sensors as a feature value. The figure which shows an example of the comparison result with the recognition correct rate by the method of one Embodiment which correct | amends and recognize a gesture.

Embodiments according to the present invention will be described below with reference to the drawings.
[One Embodiment]
(Constitution)
FIG. 1 is a cross-sectional view of a wrist when a gesture recognition device according to an embodiment of the present invention is attached to the wrist.
The gesture recognition device 1 is a device in which the device main body 10 is attached to a wristband 20 and is detachably attached to a user's wrist by the wristband 20. A sensor array 30 is disposed on a portion of the wristband 20 that is opposite to the mounting position of the apparatus main body 10. The sensor array 30 has three pressure sensors 31, 32, 33 arranged in the circumferential direction of the wristband. Each of the pressure sensors 31, 32, and 33 is, for example, a kind of variable resistor, and constitutes a resistance voltage dividing circuit. And the pressure added with the motion of the some tendon 41 which exists in the user's wrist vicinity is detected, and the detection signal which consists of an analog value is output in time series. FIG. 42 shows the ulna and 43 shows the radius.

  FIG. 2 is a block diagram illustrating a functional configuration of the apparatus main body 10. The apparatus body 10 is composed of a wearable terminal having a clock function, a music playback function, a wireless communication interface for web access and mail transmission / reception, for example, and operation keys are arranged on the upper surface. Part or all of the operation keys are used as the teacher data input unit 14. The teacher data input unit 14 is used to designate and input the type of gesture to be recognized in the learning mode. For example, numbers from “0” to “8” are assigned to nine operation keys. Yes.

  The apparatus main body 10 includes a control unit 11, an A / D conversion unit 12, and a feature amount storage unit 13 in addition to the teacher data input unit 14 as functions for realizing the present embodiment. The A / D converter 12 samples the detection signals output from the three pressure sensors 31, 32, and 33 of the sensor array 30 at predetermined sampling periods (for example, 500 Hz) and converts them into digital values, Outputs time-series data representing the pressure detection value. The feature amount storage unit 13 includes a NAND flash memory, and is used to store the feature amount of the sensor detection value for each gesture set in the learning mode.

  The control unit 11 includes a central control unit (CPU), and includes a control unit 111, a maximum value extraction unit 112, an average value calculation unit 113, a peak detection unit 114, and an operation command generation unit 115 as functions thereof. Have. These functional units are realized by causing the CPU to execute a program stored in a program memory (not shown).

  The control unit 111 comprehensively controls the operation in the control unit 11. The maximum value extraction unit 112 extracts the maximum value by comparing the digital values of the detection signals output from the three pressure sensors 31, 32, and 33 for each sampling period. The average value calculation unit 113 calculates the average value of the digital values of the detection signals between the three pressure sensors 31, 32, and 33 for each sampling period. The peak detection unit 114 detects a peak value on a time series in a predetermined period for each of the maximum value extracted by the maximum value extraction unit 112 and the average value calculated by the average value calculation unit 113.

  The operation command generation unit 115 includes a memory table in which an ID of a gesture to be recognized and an operation command corresponding to the gesture are stored in association with each other. When the gesture is recognized by the control unit 111 in the use mode, the operation command generation unit 115 also stores the ID. The corresponding operation command is read from the memory table and output.

(Operation)
Next, the operation of the gesture recognition device 1 configured as described above will be described. FIG. 3 is a flowchart showing a processing procedure and processing contents by the control unit 11.

(1) Learning Mode Prior to the use mode, it is necessary for the gesture recognition device 1 to perform machine learning on a gesture to be recognized, and this learning operation is performed as follows. There are various methods for machine learning, such as the nearest neighbor method and support vector machine, and any method can be used for learning, but the simplest nearest neighbor method is used here for ease of explanation. An example will be described.

  The gesture learning is performed by wearing the wristband 20 on the wrist and performing the gesture operation for a predetermined time in a state where the key corresponding to the gesture to be recognized is pressed in the teacher data input unit 14. For example, assuming that there are nine types of gestures from “0” to “8” shown in FIG. 4 and the “1” gesture among them is to be learned, the teacher data input unit 14 uses “1”. The gesture operation indicated by “1” in FIG. 4 is performed for one second while the “” key is pressed.

  When the learning mode is set and the control unit 111 detects a key pressing operation by the teacher data input unit 14 in step S10, the control unit 111 proceeds to step S11 and subsequently executes the feature amount registration processing in the learning mode as follows. To do.

  That is, digital values (hereinafter referred to as detection voltages) of detection signals output from the three pressure sensors 31, 32, and 33 are fetched from the A / D conversion unit 12 at each sampling timing, and the average value calculation unit 113 and the maximum value are calculated. Each value is given to the value extraction unit 112. The average value calculation unit 113 calculates the average value of the detection voltages of the three pressure sensors 31, 32, 33 with the same sampling timing, and returns the calculated average value to the control unit 111. The maximum value extraction unit 112 extracts the largest one of the detected voltages of the three pressure sensors 31, 32, 33 among the same sampling timing, and returns the extracted maximum value to the control unit 111. .

  The control unit 111 calculates the average value and the maximum value between the sensors obtained by the average value calculation unit 113 and the maximum value extraction unit 112 during the period when the “1” key of the teacher data input unit 14 is pressed. The data is stored in the buffer memory in 111. As a result, if the “1” key is pressed for 1 second, for example, since the sampling rate is 500 Hz, the buffer memory stores 500 average values and maximum values between the sensors.

  When the pressing of the “1” key is released, the control unit 111 calculates the average value between the sensors for one second (500 pieces) stored in the buffer memory and the average of the maximum values, that is, the average in the time direction. calculate. Then, the calculated average value between the sensors and the average of the maximum values in the time direction are stored in the feature amount storage unit 13 as the feature amount corresponding to the gesture “1”.

  When the “2” key is pressed for 1 second in the teacher data input unit 14 and the corresponding operation of the gesture “2” shown in FIG. 4 is performed, between the 500 sensors calculated during this period. The average of the average value and the average of the 500 maximum values are stored in the feature amount storage unit 13 as the feature amount corresponding to the gesture “2”.

  Similarly, with respect to other gestures, an average of 500 average values between sensors and an average of 500 maximum values between sensors calculated for one second are used as feature amounts corresponding to the respective gestures. 13 is stored. FIG. 5 shows an example of the feature amount of each gesture stored in the feature amount storage unit 13 as described above.

(2) Usage Mode When the feature quantity corresponding to the gesture to be recognized is stored in the feature quantity storage unit 13, the gesture recognition process can be performed thereafter.

(2-1) Calculation of Correction Value Using Reference Gesture In the use mode, the user first performs a reference gesture input operation. The control unit 111 takes in the digital value (detection voltage) of the detection signal output from the pressure sensors 31, 32, 33 by the input operation of the reference gesture from the A / D conversion unit 12. At this time, since no key operation is performed on the teacher data input unit 14, the control unit 111 proceeds from step S10 to step S12, and the detected voltages of the pressure sensors 31, 32, and 33 taken in are detected by the peak detection unit 114. To pass.

  The peak detector 114 detects any one peak value of the detection voltages of the pressure sensors 31, 32, and 33. The detection voltages of the three pressure sensors 31, 32, 33 are input to the average value calculation unit 113 or the maximum value extraction unit 112 to obtain an average value or maximum value between the sensors, The peak value may be detected based on the maximum value.

  For example, assume that a sensor detection voltage as shown in FIG. 6 is obtained by the input operation of the reference gesture. FIG. 6 shows an average value among the three pressure sensors 31, 32, 33. The peak detector 114 sequentially stores the average value between sensors obtained by the input operation of the reference gesture in a buffer memory (not shown) in time series. Then, the stored average values between sensors are sequentially compared between adjacent sampling timings, and the peak value of the average value between the sensors is detected. Peak detection is a process of detecting a phenomenon in which the average value between sensors increases continuously for a certain period of time and then decreases continuously for a certain period of time. Here, if it increases continuously for 0.2 seconds or more and then decreases continuously for 0.15 seconds or more, the average value between sensors at the timing when the increase starts to decrease and the maximum value between sensors at the same timing. Are detected as a peak average value and a peak maximum value, respectively, and returned to the control unit 111.

  When the control unit 111 receives the peak average value and the peak maximum value from the peak detection unit 114, the control unit 111 reads the average value and the maximum value indicating the feature amount corresponding to the reference gesture from the feature amount storage unit 13. In step S13, a difference between the peak average value and the average value indicating the feature amount is obtained. Similarly, the difference between the peak maximum value and the maximum value indicating the feature amount is obtained. Then, these difference values are stored in the buffer memory in the control unit 111.

  For example, it is assumed that the reference gesture is “4”, the peak average value detected by the peak detection unit 114 is 1.8, and the peak maximum value is 2.5. If the average value and the maximum value indicating the feature value of the gesture “4” stored in the feature value storage unit 13 are the average value 1.65 and the maximum value 2.2 as shown in FIG. The unit 111 calculates the difference between the peak detection value and the average value as 0.15 and the difference between the peak maximum value and the maximum value as 0.3, and stores the difference in the buffer memory.

(2-2) Recognition of Gesture to be Recognized After the input operation of the reference gesture is completed, the user once returns the hand to a relaxed state and then continuously inputs a desired gesture to be recognized for a predetermined time or more. Operate. The control unit 111 takes in the digital value (detection voltage) of the detection signal output from the pressure sensors 31, 32, 33 by the input operation of the reference gesture from the A / D conversion unit 12. Then, the average value calculation unit 113 and the maximum value extraction unit 112 detect the average value and the maximum value between the detected voltages of the three pressure sensors 31, 32, and 33.

  Subsequently, in step S13, the control unit 111 calculates the average value and the maximum value between the sensors detected by the input operation of the gesture to be recognized based on the difference value calculated and stored in the input operation of the reference gesture. to correct. For example, if the average value between sensors and the maximum value detected by the input operation of the gesture to be recognized are 1.6 and 1.9, respectively, and the difference value is subtracted from these values, The average value between sensors is 1.45, and the maximum value is 1.6.

  Next, in step S <b> 14, the control unit 111 compares the corrected average value and maximum value group with the average value and maximum value group of each gesture stored in the feature amount storage unit 13. Then, one ID of the gesture having the feature quantity closest to the corrected average value and maximum value is selected from the feature quantity storage unit 13. For example, the Euclidean distance between the corrected average value and maximum value and the average value and maximum value of each gesture stored in the feature amount storage unit 13 is calculated. As a result, in the example of the feature amount storage data illustrated in FIG. 5, the distance from the feature amount corresponding to the gesture “2” is the shortest, and the gesture in which the user performs the input operation is the gesture “2”. It is recognized.

  Finally, in step S15, the operation command generator 15 reads the operation command corresponding to the recognized gesture “2” from the memory table, and outputs this operation command. For example, if the function to be operated is a music playback function of a wearable terminal having the function of the gesture recognition device of the present embodiment, for example, the operation command is output to the music playback function and, for example, a “play” operation is performed. Make it. If the operation target is an air conditioner or a television receiver, the operation command is transmitted to the air conditioner or the television receiver by radio or infrared, for example, to perform a “cooling” operation or a “channel switching” operation. .

(Effect of embodiment)
As described above in detail, in the embodiment, in the learning mode, when the gesture to be recognized is input, the average value and the maximum value of the detection values obtained by the three pressure sensors 31, 32, and 33 are detected. Then, the time average is stored in the feature amount storage unit 13 as a feature amount corresponding to the gesture. On the other hand, in the use mode, first, a reference gesture is input, and the average value between sensors and the maximum value at the peak of the detection values obtained by the pressure sensors 31, 32, 33 at this time are detected. A difference value between the maximum value and the corresponding feature value is obtained and used as a correction value. Subsequently, the average value and the maximum value between the sensors at the peak of the detection values of the pressure sensors 31, 32, and 33 when an arbitrary gesture is input are detected and corrected by the correction value. A gesture having a feature amount close to the later average value and maximum value between sensors is selected from the feature amount storage unit 13.
Therefore, even when the wristband is re-mounted on the wrist or the direction of the arm is changed, it becomes possible to accurately recognize hand gestures, which makes it possible to use a gesture recognition device that can be used regardless of time or place. Can be provided.

  In addition, a learning mode is provided, and the feature value of the sensor detection value corresponding to each gesture to be recognized in this learning mode is calculated and stored in the storage unit. Corresponding optimum feature amounts can be registered, whereby optimum correction can be performed for each user.

(Evaluation results)
7 (a) to 7 (c), each of one subject wears a wristband 20 having three pressure sensors 31, 32, and 33 on his left wrist, and his arms are directed downward (Down) and forward (Forward). The detection values obtained by the three pressure sensors 31, 32, 33 when the gesture shown in FIG. 4 is performed once in a state of being arranged rightward (Right) are recorded. FIG. 8 shows the result of calculating the average value of the detected values of the three pressure sensors 31, 32, 33 for each arm direction, and FIG. 9 shows the result of the three pressure sensors 31, 32, 33. The results of calculating the maximum value between sensors for each arm direction are shown.

  As shown in FIG. 7, it can be seen that when the detection values by the same gesture when the directions of the arms are different are compared, there is a large fluctuation. For example, when the gesture “5” is input, the values of the pressure sensors 31 to 33 are close to 2.1, 2.6, and 2.4 when the arm is in the downward direction (Down). When the direction is (Forward), it is around 2.2, 2.0, 2.5, and when the arm is in the right direction (Right), it is around 1.6, 2.3, 1.4. Everything is different. On the other hand, as shown in FIG. 8, the graph showing the average value between the sensors for each arm direction has a parallel relationship. Similarly, the result of the maximum value between sensors shown in FIG. 9 is also substantially coincident or close to parallel.

  On the other hand, FIGS. 10A, 10 </ b> B, and 10 </ b> C show detection values of individual sensors for each type of gesture when wristband replacement is performed, and arm directions are all downward ( Down). FIG. 11 shows the result of calculating the average value between the detected values of the three pressure sensors 31 to 33 every time the wristband 20 is replaced, and FIG. 12 lists the maximum value between the sensors of the three pressure sensors 31 to 33. The results calculated for each band change are shown.

  As shown in FIG. 10, it can be seen that the values of the same gesture for each replacement of the wristband 20 are greatly changed. For example, the detected value of the pressure sensor of the gesture “6” is around 2.7, 2.8, and 2.9 at the first time, but becomes 2.3, 1.6, and 2.7 at the second time. The third time is 2.6, 2.5, and 2.9. That is, as in the case of FIGS. 7A, 7B, and 7C, the magnitudes of the detected values of the pressure sensors 31 to 33 are switched. On the other hand, in the graph of the average value between sensors of FIG. 11, it has a substantially parallel relationship. The graph of the maximum value between the sensors in FIG.

  From the above results, each time average value of the inter-sensor average value and the inter-sensor maximum value is stored as a feature amount, and the reference gesture is input before inputting the desired gesture, and the pressure sensors 31 to 33 at this time are operated. A correction value is calculated based on the average value and maximum value of the detected values between sensors, and then the sensor is mounted if the average value and maximum value between sensors when the desired gesture is input is corrected with the correction value. Even if it is changed again or the direction of the arm is changed, the recognition accuracy of the gesture can be maintained.

  FIG. 13 shows a method of using the detection values acquired from three pressure sensors as feature values as they are as before (three-sensor independent detection method) and the recognition method according to the present embodiment, and actually attaching a wristband. It is the figure which compared the recognition correct answer rate of gesture by the average of five test subjects by redoing or changing the direction of an arm.

  As is apparent from FIG. 13, in the conventional method, when the wristband is remounted or the direction of the arm is changed, the accuracy rate is reduced to 50%. In the “method to do”, the correct answer rate was always 80% or higher under any conditions. From this result, it is possible to obtain a high recognition rate according to the method of the present embodiment in practical use situations in which the wristband is remounted or the gesture direction is input by changing the direction of the arm. I have great expectations.

  Note that the accuracy rate in the conventional method under the condition that “the wristband is not removed and the arms are in the same direction” is as high as 90.7%. It is not a number that reflects actual use.

[Other Embodiments]
In the embodiment, the recognition is performed using both the average value between sensors and the maximum value between sensors. However, the recognition is performed based on one of the average value between sensors and the maximum value between sensors. It may be.

  In the embodiment, the operation command is generated and output based on the gesture recognition result. However, the information indicating the gesture recognition result is output as it is, and the device or the electric device having the function to be operated is Information representing the recognition result may be received, and an operation command may be generated to control the function.

  The number of pressure sensors may be two or four or more in addition to three, and the position of the pressure sensor in the wristband can be set as appropriate within a range in which tendon movement can be detected. In addition, the configuration of the apparatus, the recognition processing procedure, the processing content, and the like can be variously modified and implemented without departing from the gist of the present invention.

  In short, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

  DESCRIPTION OF SYMBOLS 1 ... Gesture recognition apparatus, 10 ... Main body, 20 ... Wristband, 30 ... Sensor array, 31, 32, 33 ... Pressure sensor, 41 ... Tendon, 42 ... Ulna, 43 ... Rib, 11 ... Processing unit, 12 ... A / D conversion unit, 13 ... feature quantity storage unit, 14 ... teacher data input unit, 111 ... control unit, 112 ... maximum value extraction unit, 113 ... average value calculation unit, 114 ... peak value detection unit, 115 ... operation command generation Department.

Claims (8)

Means for receiving time-series data indicating detected pressure values from a plurality of pressure sensors arranged in a wristband;
During a first period in which an input operation of a reference gesture is performed, a peak value is detected from time series data received from each of the plurality of pressure sensors, and the time series data at the time of detection of the peak value is detected. Means for detecting at least one of an average value and a maximum value of and generating a correction value based on at least one of the detected average value and maximum value;
Detecting at least one of an average value and a maximum value between the time series data respectively received from the plurality of pressure sensors in a second period in which an arbitrary gesture input operation is performed following the first period. Means for correcting at least one of the detected average value and maximum value based on the correction value;
A gesture recognition device comprising: means for recognizing the arbitrary gesture based on at least one of the corrected average value and maximum value. A storage unit storing feature amounts corresponding to each of the plurality of gestures to be recognized;
The means for generating the correction value includes at least one of an average value and a maximum value between the time-series data at the time of detection of the peak value, and a feature amount corresponding to the reference gesture stored in the storage unit. The gesture recognition apparatus according to claim 1, wherein the correction value is generated based on:   For each of the plurality of gestures to be recognized, at least one of the average value and the maximum value between the time series data received from the plurality of pressure sensors during the input operation period is detected, and the detected average value And learning means for calculating an average during the input operation period for at least one of the maximum value and storing the calculated average as a feature amount corresponding to the corresponding gesture in the storage unit. The gesture recognition device according to claim 2.   The apparatus further comprises means for generating operation command information corresponding to an arbitrary gesture recognized by the means for recognizing the gesture, and sending the generated operation command information to a target device. The gesture recognition device according to any one of Items 1 to 3. A gesture recognition method executed by a device for recognizing a gesture represented by a hand movement using a plurality of pressure sensors arranged in a wristband,
A process of receiving time-series data indicating detected pressure values from the plurality of pressure sensors in a first period in which a reference gesture input operation is performed;
A peak value is detected from the received time series data, and at least one of an average value and a maximum value between the time series data at the time of detection of the peak value is detected, and at least one of the detected average value and maximum value is detected. The process of generating correction values based on one side,
A process of receiving time-series data indicating detected pressure values from the plurality of pressure sensors in a second period in which an input operation of an arbitrary gesture is performed following the first period;
Detecting at least one of an average value and a maximum value between a plurality of time-series data received in the second period, and correcting at least one of the detected average value and maximum value based on the correction value; Process,
And a step of recognizing the arbitrary gesture based on at least one of the corrected average value and the maximum value. Further comprising storing a feature amount corresponding to each of the plurality of gestures to be recognized in the storage unit,
The process of generating the correction value includes at least one of an average value and a maximum value between the time-series data at the time of detection of the peak value, and a feature amount corresponding to the reference gesture stored in the storage unit. The gesture recognition method according to claim 5, wherein the correction value is generated based on:   The process of storing the feature amount in the storage unit includes, for each of the plurality of gestures to be recognized, at least an average value and a maximum value between time series data respectively received from the plurality of pressure sensors during the input operation period. One is detected, an average in the input operation period is calculated for at least one of the detected average value and maximum value, and the calculated average is stored in the storage unit as a feature amount corresponding to the corresponding gesture The gesture recognition method according to claim 6, wherein:   The program which makes the processor with which the said gesture recognition apparatus performs the process by the means with which the gesture recognition apparatus in any one of Claims 1 thru | or 4 is provided is performed.