JP2011250915A - Posture recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a posture recognition device by which a sitting posture of a user can be recognized.SOLUTION: The posture recognition device is built in a pedometer 1 worn on the waist of a user 10. Output signals of an acceleration sensor 2 commonly used as the pedometer 1 are input to a calculation processor 5. When the calculation processor 5 recognizes by output signals of the acceleration sensor 2 that the user 10 sits, the posture recognition device recognizes a sitting posture of the user 10 on the basis of signals output from the acceleration sensor 2 in a preset period. During a long time when the user 10 is wearing the pedometer 1 on, sensing and recognition of sitting postures useful for health even when the user sits except a time of walking or running are made.

Description

  The present invention relates to a posture recognition device that is worn and used by a user and recognizes the posture of the user using an acceleration sensor, and more particularly to a posture recognition device that recognizes the posture when the user is sitting.

  2. Description of the Related Art Conventionally, an apparatus and a system that incorporate an acceleration sensor and recognize human behavior using the acceleration sensor have been proposed.

  Patent Document 1 (Japanese Patent Application Laid-Open No. Sho 62-106742) describes a device that analyzes the waveform of an acceleration sensor worn by a user and recognizes a human motion state such as standing, sitting, walking, and running. Has been.

  In Patent Document 2 (Japanese Patent Laid-Open No. 10-024026), a gyro sensor is used in addition to an acceleration sensor, and a rough outline of the user such as walking, sitting, eating, and meeting is used. An apparatus capable of recognizing behavior is described.

  Patent Document 3 (Japanese Patent Laid-Open No. 10-113343) is a system capable of recognizing various human behaviors and activities, and describes using an acceleration sensor.

  However, the techniques described in Patent Documents 1 to 3 described above cannot recognize the user's sitting posture.

Japanese Patent Laid-Open No. 62-106742 Japanese Patent Laid-Open No. 10-024026 JP-A-10-113343

  Accordingly, an object of the present invention is to provide a posture recognition device capable of recognizing the user's sitting posture, and further to provide a posture recognition device capable of notifying the user when a bad sitting posture of the user is recognized. There is to do.

In order to solve the above problems, an attitude recognition device of the present invention includes an acceleration sensor that is worn on a user's body,
A determination unit for determining whether the user is sitting based on an output signal of the acceleration sensor;
And a recognition evaluation unit that recognizes and evaluates the sitting posture of the user based on an output signal of the acceleration sensor when the determination unit determines that the user is sitting.

  According to the posture recognition device of the present invention, the determination unit determines whether or not the user is sitting based on an output signal of an acceleration sensor worn on the user's body, and the recognition evaluation unit When the determination unit determines that the user is sitting, the sitting posture of the user can be recognized based on the output signal of the acceleration sensor.

  The acceleration sensor is mounted, for example, on the user's waist, and detects the inclination of the posture around the user's waist. Further, as the acceleration sensor, a triaxial acceleration sensor is employed as an example.

Further, in the posture recognition device according to one embodiment, the recognition evaluation unit includes:
A neural network trained in advance so that the user's sitting posture can be recognized based on an output signal from the acceleration sensor and an output signal from the acceleration sensor.

  According to this embodiment, the recognition evaluation unit can perform complex and advanced recognition of the user's sitting posture using a neural network.

Further, in the posture recognition device of one embodiment, the sitting posture that the neural network learns in advance includes a plurality of different bad sitting postures that adversely affect health,
The recognition evaluation unit
When the user's sitting posture recognized by the neural network is the bad sitting posture, the information processing unit has a notifying unit for notifying the user of information representing an adverse effect of the bad sitting posture on the user's health.

  According to this embodiment, when the user's seating posture recognized by the neural network is the bad seating posture, the notification unit causes the bad seating posture to be healthy for the user. The user can be notified of information indicating the adverse effect.

Further, in the posture recognition device according to one embodiment, the recognition evaluation unit includes:
In a predetermined period, a plurality of average values for each preset time of the output signal of the acceleration sensor are obtained, and the seating posture of the user is recognized based on the plurality of average values.

  According to this embodiment, the recognition evaluation unit obtains a plurality of average values of the output signals of the acceleration sensor, thereby increasing the data used for the recognition of the sitting posture, and the complicated and advanced recognition regarding the user's sitting posture. Is possible.

Further, in the posture recognition device according to one embodiment, the recognition evaluation unit includes:
A plurality of average values for each detection time set in advance of the output signal of the acceleration sensor are obtained in a period from a time that is a predetermined time or more before the current time to the current time, and based on the plurality of average values To recognize the user's sitting posture.

  According to this embodiment, the recognition evaluation unit obtains the plurality of average values of the output signals of the acceleration sensor in a period from a time before a preset time to a current time from a current time, By increasing the data used for the recognition of the sitting posture, it becomes possible to recognize complex and advanced recognition of the user's sitting posture.

  In one embodiment, the set time is 10 seconds or 60 seconds, and the detection time is 1 second.

  According to this embodiment, the recognition evaluation unit calculates a plurality of the average values of the output signals of the acceleration sensor every second for 10 seconds from the current time or for a period from the time before 60 seconds to the current time. It has been found that the determination is preferable in terms of the characteristics of the human body in determining the seating posture of the user.

  Moreover, the pedometer of one Embodiment was provided with the said attitude | position recognition apparatus.

  This pedometer senses and recognizes a useful sitting posture related to health even when the user is seated while walking or running, during the long time the user is wearing the pedometer. be able to. Further, by using the acceleration sensor also as the acceleration sensor of the pedometer, the posture when the user is sitting can be recognized using the acceleration sensor as a component of the pedometer.

  Moreover, the active mass meter of one Embodiment was provided with the said attitude | position recognition apparatus.

  According to this activity meter, sensing and recognition of a useful sitting posture related to health even when sitting other than when walking or running, while the user is wearing the activity meter. It can be performed. Further, by using the acceleration sensor also as the acceleration sensor of the activity meter, the posture when the user is sitting can be recognized using the acceleration sensor as a component of the activity meter.

  That is, even in a situation where the user is not very active and the pedometer or activity meter is not very useful, there is a sitting posture as information to be detected that is useful for those who are interested in health. For example, good or bad posture when sitting for a long time is useful for knowing the user's current and future body distortions.

  Currently, there are many pedometers that use acceleration sensors as the main sensing means, and activity meters that allow you to see the calories burned, the size of exercise, and the amount of activity. There are many people who think that lack of exercise is a problem among these users, and it can be inferred that there is little time actually walking or running, but for example, if the work is a desk work person, Therefore, even if the activity meter is worn, the device may perform its function only for a short time walking during commuting. In such a case, the present embodiment is useful.

  According to the posture recognition device of the present invention, the determination unit determines whether or not the user is sitting based on the output signal of the acceleration sensor attached to the user's body, and when the user is sitting When the determination unit determines, the recognition evaluation unit can recognize the user's sitting posture based on the output signal of the acceleration sensor.

It is a schematic diagram which shows the front of the pedometer with which embodiment of the attitude | position recognition apparatus of this invention was integrated, the side surface, and the decomposition | disassembly side surface in order from the left. It is a decomposition | disassembly schematic diagram of the said pedometer. It is a figure which shows typically the user who mounted | wore the waist (belt) with the said pedometer. It is a schematic diagram of the neural network with which the said embodiment is provided. It is a schematic diagram which shows a mode that the user wearing the said pedometer is sitting in a good posture. It is a schematic diagram which shows a mode that the said user is sitting with an example of a bad attitude | position. It is a schematic diagram which shows a mode that the said user is sitting in the other example of a bad attitude | position. It is a figure which shows the corresponding | compatible table | surface with the output layers n1-n7 of the said neural network, and an operation state (type of sitting posture). It is a figure which shows the learning flowchart and attitude | position recognition flowchart of the said neural network.

  Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments.

  FIG. 1A shows the front, side, and exploded side of a pedometer 1 in which an embodiment of the posture recognition device of the present invention is incorporated in order from left to right. 1B shows a front panel 1A of the pedometer 1 to which a display 3 for displaying the state is attached, a main body substrate 1B, a display 3, and a rechargeable battery (secondary battery) 6 as a power source for the main body substrate 1B. It is a schematic diagram which shows the back cover 1C. FIG. 1C schematically shows the user 10 wearing the pedometer 1 on his / her waist (belt).

  As shown in FIG. 1A, the main body substrate 1B and the rechargeable battery 6 are accommodated between the front panel 1A and the back cover 1C. As shown in FIG. 1B, an acceleration sensor 2, an arithmetic processing unit 5, a wireless communication function unit 4, and a speaker 15 are mounted on the main board 1B. The display 3 and the speaker 15 constitute a notification unit. The acceleration sensor 2 is a three-axis acceleration sensor that can detect the acceleration in the triaxial directions of the X axis, the Y axis, and the Z axis within a range of ± 2G. The arithmetic processing unit 5 is composed of a microcomputer or the like, and includes a neural network 20 and a memory 5A described later. The output signal of the acceleration sensor 2 is input to the arithmetic processing unit 5. The arithmetic processing unit 5 constitutes a discrimination unit and a recognition evaluation unit. That is, the acceleration sensor 2 and the arithmetic processing unit 5 constitute the posture recognition device of the present embodiment, and the acceleration sensor 2 is shared with the acceleration sensor for measuring the number of steps of the pedometer 1.

  The wireless communication function unit 4 is for wireless communication between the mobile phone, the personal computer, etc. and the pedometer 1. It is assumed that the pedometer 1 is connected to a mobile phone by the wireless communication function unit 4 via Bluetooth.

  As shown in FIG. 1C, the user 10 uses the pedometer 1 attached to the side of the waist, generally at the height of the belt. The user 10 wears the pedometer 1 at the above-described position and lives in the daytime. While the user 10 is walking, the pedometer 1 functions as a pedometer, and displays standard display calories and the like corresponding to the number of steps of the user 10 and the number of steps.

  On the other hand, when the user 10 sits on a chair or the like, the arithmetic processing unit 5 recognizes that the user 10 is sitting based on an output signal input from the acceleration sensor 2. Here, the seating operation and the rising operation of the user 10 include output patterns of the acceleration sensor 2 corresponding to each operation. Therefore, recognition that the user 10 sat down is possible by various methods. For example, when the arithmetic processing unit 5 detects that the number of steps does not increase over a long time from the output signal of the acceleration sensor 2, the user 10 has a high probability. You can determine that you are sitting. The arithmetic processing unit 5 can also recognize movement corresponding to sitting and rising in the background of the step count by using pattern matching by the neural network 20 described later. Here, the neural network 20, which will be described later, is for determining the sitting posture, and is a system with a large time constant in which data per second input from the acceleration sensor 2 is determined retroactively 60 seconds before. However, when recognizing non-stationary movements such as a sitting movement and a rising movement, the time constant is determined by, for example, determining data every 100 ms (milliseconds) input from the acceleration sensor 2 up to 5 seconds before. Recognize the motion with a small.

  When the user 10 sits on a chair or the like and the arithmetic processing unit 5 determines that the user 10 is in a sitting state based on an output signal from the acceleration sensor 2, the posture recognition device incorporated in the pedometer 1 is Then, transition to the sitting posture detection mode. In this sitting posture detection mode, the arithmetic processing unit 5 that constitutes the recognition and evaluation unit sets the average value A for one second of the output signal output from the acceleration sensor 2 as an average value A, and the average value for one minute (hereinafter referred to as the average value A). , The average value B), the dispersion of the average value A over 1 minute, and the number of zero crossings of the average value A are used to recognize the sitting posture of the user 10.

  For example, as shown in FIG. 3A, a case where the user 10 sitting is in a good posture will be described. In this case, the average value B of the output signal of the acceleration sensor 2 is a value representing the direction of gravity as viewed from the three-axis XYZ orthogonal coordinate system of the acceleration sensor 2. In the case of the good posture shown in FIG. 3A, the Z axis of the acceleration sensor 2 of the posture recognition device is facing upward. For this reason, the acceleration sensor 2 of the posture recognition device outputs a value equivalent to 1G in the −Z direction as an output signal. Further, when the user 10 takes a good posture as shown in FIG. 3A, the dispersion of the average value A of the output signal of the acceleration sensor 2 in one minute and the number of zero crossings of the average value A tend to be small. is there.

  On the other hand, when the sitting user 10 takes the bad posture shown in FIG. 3B, the direction of gravity is greatly deviated from the −Z direction, so the average value B of the output signal of the acceleration sensor 2 is from the −Z direction. It becomes a value equivalent to 1G in the direction greatly deviated. In the case of the bad posture shown in FIG. 3B, the dispersion of the average value A of the output signal of the acceleration sensor 2 in one minute and the number of zero crossings of the average value A tend to be not so large although they are in a bad posture.

  In addition, when the sitting user 51 takes the bad posture shown in FIG. 3C, the -Z direction in the case of the good posture shown in FIG. 3A and the -Z in the case of the bad posture shown in FIG. 3B. The direction is about the middle of the direction. In the bad posture of FIG. 3C, the average value B of the output signal of the acceleration sensor 2 becomes a value equivalent to 1G in a direction shifted from the −Z direction of FIG. 3C, but from the −Z direction as compared with the bad posture of FIG. 3B. The shift is small. The bad posture shown in FIG. 3B is a posture in which the back muscles are rounded, and therefore there is a great burden on the internal organs, the skeleton, the muscles of the shoulders and the neck. Therefore, the posture of the user 10 tends not to settle down. As a result, the variance of the average value A in one minute and the zero cross of the average value A become large.

  As described above, the good posture and the bad posture can be discriminated by using both the average value B of the average value A of the output signal of the acceleration sensor 2, the variance of the average value A, and the zero cross. 3A, FIG. 3B, and FIG. 3C, there are subtle differences in the good and bad postures other than those shown in the examples as seen in the average value B, variance, and zero cross value of the output signal of the acceleration sensor 2, respectively. . For this reason, it is not a simple case classification as described in the example of FIG. 3A to FIG. 3C, and if a more advanced pattern recognition / discrimination method is used, the sitting posture can be distinguished and recognized by the above-mentioned subtle differences. The posture recognition apparatus according to this embodiment performs such more advanced determination using the neural network 20 shown in FIG.

  The output value corresponding to the user's movement (posture) output from the acceleration sensor 2 is passed to the arithmetic processing unit 5 for appropriate signal processing, and passed to the neural network 20 set in the arithmetic processing unit 5. .

  The neural network 20 has been adjusted (learned) in advance to recognize a typical example of a good posture and a bad posture from the output value of the acceleration sensor 2. When an input signal is input, the “good posture”, Returns an appropriate response corresponding to each posture such as “posture in which the abdominal muscles weaken”, “poor posture in the lower back”.

  The result (sitting posture) recognized by the neural network 20 is displayed on the display 3 as the current sitting posture state. The recognized sitting posture can also be displayed from the wireless communication function unit 4 on the screen of a cellular phone or a personal computer with which wireless connection is established.

  At the same time, by providing the mobile phone or personal computer with a database on the health, beauty benefits, disadvantages, etc. caused by the posture recognized by the neural network 20, "it is bad for the waist if you keep the current posture", " Significant information such as “causes stiff shoulders” and “may cause headaches” can be fed back to the user.

  Next, specific processing of recognition by the posture recognition device of this embodiment will be described. The time at which the operating state is recognized is t1, and the time 60 seconds before the time t1 is t0. The acceleration sensor 2 measures the acceleration of 100 samples per second and outputs an output signal representing the acceleration of 100 samples to the arithmetic processing unit 5 per second. The arithmetic processing unit 5 uses the 100 samples of acceleration data to calculate an average value (average value A) of accelerations in the X, Y, and Z directions every second.

  The arithmetic processing unit 5 calculates an average value A of accelerations in the X direction, the Y direction, and the Z direction in a period of 1 second for 60 seconds from time t0 to time t1, and these three kinds of average values A are calculated. In the memory built in the unit 5, values from 1 second before the current time (time t1) to 60 seconds before (time t0) (3 directions × 60) are held.

  The arithmetic processing unit 5 uses the log of the average value A between time t0 and time t1, and further calculates the average value (average value) of the average value A in the X direction, the Y direction, and the Z direction for 60 seconds. B) is created, and the average values are X1, Y1, and Z1, respectively. Further, based on the average value A, standard deviations in the same section (t0 to t1) are also obtained and are set as standard deviations X2, Y2, and Z2, respectively.

  Next, the arithmetic processing unit 5 searches for the maximum value and the minimum value of the average value A of the acceleration in each direction in the X, Y, and Z directions between the times t0 and t1, and calculates the difference between the maximum value and the minimum value, respectively. Let X3, Y3, Z3. The arithmetic processing unit 5 also counts the number of zero crossings of the average value A of acceleration in each direction in the X, Y, and Z directions in the same section (t0 to t1), and the zero crossing times are X4, Y4, and Z4, respectively. To do.

  Actual measurement by the acceleration sensor 2 itself is performed at a cycle of 10 ms (milliseconds), but the average values X1, Y1, Z1, standard deviations X2, Y2, Z2, and the difference between the maximum value and the minimum value X3, Y3, The calculation of Z3 and the number of zero crossings X4, Y4, and Z4 is carried out using the average value A from the time t0 60 seconds before to the current time t1 in a cycle of 1 second. For example, the average value B (X1, Y1, Z1) of acceleration in each direction of the X axis, Y axis, and Z axis is the current time (t1) in one of the X axis, Y axis, and Z axis. That is, an average value B calculated from an average value A of 60 samples from the current time (t1) to 60 seconds before (t0) is generated every second.

  In the example of this embodiment, the time from the time t0 to the time t1 is set to 60 seconds. However, the time from the time t0 to the time t1 is not limited to 60 seconds. It may be set to 5 seconds, for example, or may be set to other desired time. Further, in the example of this embodiment, the cycle for obtaining the average value A of the acceleration is a 1-second cycle. However, the cycle for obtaining the average value A of the acceleration is not limited to the 1-second cycle, and for 3 seconds, It may be set to 2 seconds, 0.5 seconds, etc., and can be set to other desired times. Further, the period for obtaining the average value A of the acceleration may be set to a certain period before a certain time from the current time.

  Then, in the arithmetic processing unit 5, the total 12 pieces of data (X1, Y1, Z1, X2, Y2, Z2, X3, Y3, Z3, X4, Y4, Z4) are converted into 12 of the neural network 20 shown in FIG. Input to the input layers i1 to i12. As a result, the operating state (posture) of the user 10 is acquired from the output layers n1 to n7. The average acceleration values X1, Y1, and Z1 are input to the input layers i1, i2, and i3, and the standard deviations X2, Y2, and Z2 are input to the input layers i4, i5, and i6. Differences X3, Y3, and Z3 between the maximum value and the minimum value are input to the input layers i7, i8, and i9, and the number of zero crossings X4, Y4, and Z4 are input to the input layers i10, i11, and i12.

  Here, it is assumed that seven types of postures are learned in advance in the neural network 20 as the operation states to be discriminated. This neural network 7 is composed of 12 input layers i1 to i12, 7 intermediate layers m1 to m7, and 7 output layers n1 to n7, and BP (back propagation). Learning with an algorithm called law. In the method using the BP method, a set of inputs (here, 12 types of values of X1 to X4, Y1 to Y4, Z1 to Z4) and an appropriate operation state (sitting posture) corresponding to the set are input. A large number of sets are prepared and used for learning. Specifically, the data corresponding to the acceleration data X1 to X4, Y1 to Y4, and Z1 to Z4 when the user sits in advance and takes a good posture or a bad posture is provided to a developer or a test staff. Measure the target and use it for learning.

  FIG. 4 is a correspondence table showing the relationship between the output layers n1 to n7 of the neural network 20 and the operation state (type of sitting posture). Twelve acceleration data X1 to X4, Y1 to Y4, and Z1 to Z4 corresponding to several types of postures (7 to 7 types of postures in the output layers n1 to n7 in FIG. 4) are acquired, and 12 of them are acquired. The neural network 7 is made to learn using the acceleration data X1 to Z4. Then, when the acceleration data X1 to Z4 are input to the input layers i1 to i12, the determination results are output to the output layers n1 to n7. Each of the output layers n1 to n7 corresponds to one sitting posture different from each other as shown in the table of FIG. 4, and is based on 12 pieces of acceleration data X1 to Z4 input to the 12 input layers i1 to i12. , Takes values from 0 to 1. For example, as a result of inputs (X1 to Z4) to the input layers i1 to i12 of the neural network 20 generated from the output of the acceleration sensor 2 by the above-described process, the value of the output layer n1 is close to 1, and the other output layers When the values of n2 to n7 are close to 0, it is determined that the posture of the user 10 that is the basis of the input (X1 to Z4) is the posture corresponding to the output layer n1, that is, the “poor posture for waist”. To do.

  Learning of the neural network 20 is performed by executing an algorithm for learning on a PC (personal computer) or the like using data measured for learning. The general flow is the same as the learning flowchart shown in FIG.

  That is, first, in step S1, acceleration data corresponding to a plurality of sitting postures to be discriminated are actually measured and acquired. The plurality of sitting postures to be determined are, for example, as shown in FIG. 4, “Poor posture on hips”, “Poor posture on right and left balance”, “Good posture”, “Poor posture on internal organs”, “Abdominal muscles” Are postures that cause weakness, postures that cause stiff shoulders, and postures that cause headaches. Then, the acceleration data output from the acceleration sensor 2 when the user 10 is in these postures is input to the arithmetic processing unit 5.

  Next, in step S2, the arithmetic processing unit 5 calculates the above average values X1, Y1, Z1, standard deviations X2, Y2, Z2, the difference between the maximum value and the minimum value X3, Y3, Z3, zero crossing from the acceleration data. Teacher data including the number of times X4, Y4, Z4, etc. is calculated.

  In step S3, the neural network 20 is trained using the teacher data. Next, it progresses to step S4 and the numerical formula for attitude | position discrimination | determination is obtained by the said learning. Next, the process proceeds to step S5, and the mathematical expression for posture determination is incorporated into the arithmetic processing unit 5.

  As a result of learning according to the learning flowchart of FIG. 5, a calculation formula equivalent to the neural network 20 that has completed learning is obtained. By incorporating this calculation formula into the arithmetic processing unit 5, a real-time posture determination result can be fed back to the user 10 using the pedometer 1 in which the posture recognition device is incorporated.

  The determination of the sitting posture by the arithmetic processing unit 5 is always performed in real time while it is determined that the user 10 is sitting (sitting) during the time when the user 10 is wearing the pedometer 1 and is active. . The general flow of the posture determination is as shown in the recognition flowchart of FIG.

  First, in step S11, the arithmetic processing unit 5 calculates an average value A of accelerations in the X direction, the Y direction, and the Z direction at a cycle of 1 second. Next, in step S12, the arithmetic processing unit 5 calculates the average of 60 pieces × 3 directions from the average value A every second from t0 60 seconds before time t1 to 1 second before time t1 when posture recognition is performed. Values X1, Y1, and Z1, standard deviations X2, Y2, and Z2, differences between maximum and minimum values, X3, Y3, and Z3, and the number of zero crossings X4, Y4, and Z4 are calculated.

  Next, in step S13, the arithmetic processing unit 5 calculates the calculated average values X1, Y1, Z1, standard deviations X2, Y2, Z2, the difference between the maximum value and the minimum value X3, Y3, Z3, and the number of zero crossings X4, Y4, Z4 is input to the input layers i1 to i12 of the neural network 20. Then, in step S14, output data relating to seven types of sitting postures as shown in the correspondence table of FIG. 4 is obtained from the output layers n1 to n7 of the neural network 20. In step S15, the sitting posture of the user 10 at time t1 when posture recognition is determined from the output data from each of the output layers n1 to n7. In step S16, the sitting posture of the user 10 at time t1 is determined. The determination result is displayed on the display 3. The determination result is stored in a log in a mobile phone or a personal computer by the wireless communication function unit 4.

  The flow from step S11 to step S16 in this recognition flowchart is performed at a cycle of 1 second. As a result, the sitting posture of the user 10 is recognized every second, and the recognition result is recognized in real time by the user 10 or an external system using the recognition result (for example, an application for a mobile phone for the purpose of maintaining health). Can provide feedback. Further, when feeding back to the user 10 that the sitting posture is bad, the speaker 15 built in the main body of the pedometer 1 in which the posture recognition device is incorporated is used to give the user 10 a warning sound. Take action to promote awareness.

  In this way, the average value, standard deviation, difference between the maximum and minimum values, and the number of zero crossings in a certain time interval of the output of the acceleration sensor 2 are calculated, and the data used to recognize the seating posture of the user 10 is virtually The plurality of parameters are simultaneously recognized as inputs to the neural network 20 having a large number of inputs. As a result, even though the hardware is simple, the situation in which the sitting posture of the user 10 is determined using a high-level algorithm from a large number of data virtually in terms of software results, and as a result, the complexity of the posture of the user 10 is complicated. Therefore, advanced recognition is possible. Whether the posture around the waist is tilted by using the neural network 20 that can process a large amount of information at the same time as the recognition algorithm and discriminating these parameters generated from the output of the acceleration sensor 2 in a certain time interval. In addition to simple recognition as described above, for example, it is possible to recognize complex contents such as “whether the internal organs have a bad posture” as described above.

  As described above, the posture recognition device according to the present embodiment is incorporated in the pedometer 1, and the posture of the user 10 sitting using the components (acceleration sensor 2, display 3, etc.) of the pedometer 1 is determined. If the estimated posture is detected, a warning or an instruction can be given to the user 10 through a mobile phone, a PC (personal computer), or the display 3 or speaker 15 of the pedometer 1.

  Therefore, according to the posture recognition device of the present embodiment, it is possible to recognize advanced posture with simple hardware without using a plurality of sensors, and it is not necessary to attach sensors to various parts of the body. The posture can be detected in a simple device such as a pedometer. Therefore, in the long time that the user 10 is wearing the pedometer 1, the user 10 can perform useful sensing and recognition of the sitting posture related to health even when sitting other than when walking or running. Can provide feedback on health. In other words, in a pedometer that is always worn and used, information useful for the user's health can be fed back during “sitting time” that is not very useful, and the significance of wearing the pedometer constantly improves.

  The posture recognition device of the above embodiment may be incorporated in the activity meter, although it is incorporated in the pedometer, and may be a single posture recognition device without being incorporated in another device. In the above embodiment, the pedometer 1 is worn on the user's waist, but the posture recognition device of the present invention is worn on the legs, arms, back, abdomen, chest, head, shoulders other than the user's waist. Etc.

DESCRIPTION OF SYMBOLS 1 Pedometer 1A Front panel 1B Main board 1C Back cover 2 Acceleration sensor 3 Display 4 Wireless communication function part 5 Arithmetic processing part 5A Memory 6 Rechargeable battery 10 User 15 Speaker 20 Neural network

Claims (8)

An acceleration sensor worn on the user's body;
A determination unit for determining whether the user is sitting based on an output signal of the acceleration sensor;
A posture comprising: a recognition evaluation unit for recognizing and evaluating the sitting posture of the user based on an output signal of the acceleration sensor when the determination unit determines that the user is sitting; Recognition device. The posture recognition device according to claim 1,
The recognition evaluation unit
A posture recognition apparatus comprising: a neural network that has been trained in advance so that the user's sitting posture can be recognized based on an output signal from the acceleration sensor and an output signal from the acceleration sensor. The posture recognition apparatus according to claim 2,
The sitting posture that the neural network learns in advance includes a plurality of different bad sitting postures that adversely affect health,
The recognition evaluation unit
When the user's sitting posture recognized by the neural network is the bad sitting posture, the information processing device has a notifying unit for notifying the user of information indicating an adverse effect of the bad sitting posture on the user's health. Characteristic posture recognition device. In the posture recognition device according to any one of claims 1 to 3,
The recognition evaluation unit
In a predetermined period, a plurality of average values for each detection time set in advance of the output signal of the acceleration sensor are obtained, and the seating posture of the user is recognized based on the plurality of average values. Posture recognition device. In the posture recognition device according to any one of claims 1 to 4,
The recognition evaluation unit
A plurality of average values for each detection time set in advance of the output signal of the acceleration sensor are obtained in a period from a time that is a predetermined time or more before the current time to the current time, and based on the plurality of average values A posture recognition device characterized by recognizing the sitting posture of the user. In the posture recognition device according to claim 5,
The posture recognition apparatus characterized in that the set time is 10 seconds or 60 seconds, and the detection time is 1 second.   A pedometer comprising the posture recognition device according to any one of claims 1 to 6.   An activity meter comprising the posture recognition device according to any one of claims 1 to 6.

Images