JPH09244799A - Pen type input device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To increase the pattern recognition precision of the pen type input device with simple constitution since the needs for a small-sized accurate input device is increasing. SOLUTION: A coordinate arithmetic part 4 calculates the angle of rotation of a pen shaft from acceleration in each axial direction of a pen axis coordinate system and the angular velocity of rotation on each axis of the pen axis coordinate system, and calculates the movement direction and movement distance of the pen tip part on a gravitational coordinate system on the basis of the calculated angle of rotation of the pen shaft and the acceleration of the pen tip part on the pen axis coordinate system. A neural network arithmetic part 6 calculates an intermediate recognition result of an input pattern on the basis of the movement direction and movement quantity of the pen tip part. A postprocessing part 7 decides the input pattern on the basis of the intermediate recognition result of the input pattern obtained by the neural network arithmetic part 6.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pen-type input device for inputting figures and characters.

[0002]

2. Description of the Related Art Keyboards, mice, digitizers, light pens, tablets and the like are used as input devices such as computer devices. With the miniaturization of computer devices, the need for portable terminal devices has increased and the number of users has been increasing year by year. Therefore, a small input device has been required.

[0003] There is a limit in miniaturizing a keyboard in terms of a human interface, and it is not practical as an input device of a portable terminal device. Although a mouse can be downsized as a pointing device, it is not suitable for inputting figures, characters, and the like.

For this reason, a pen-type input device using a tablet and a pen is often used as an input device of a portable terminal device. In order to further reduce the size of the pen-type input device using the tablet, the size of the tablet becomes a problem. Therefore, for example, a pen type computer input device disclosed in Japanese Patent Laid-Open No. 6-67799, a data input device disclosed in Japanese Patent Laid-Open No. 7-84716, and Japanese Patent Laid-Open No. 7-84716.
A tabletless input device, such as the handwriting input device disclosed in Japanese Patent Publication No. 200127, has been developed.

A pen-type computer input device disclosed in Japanese Patent Application Laid-Open No. 6-67799 checks the moving direction and the moving amount with an acceleration sensor, and detects the moving direction and the moving amount detected by the acceleration sensor with a piezoelectric vibrating gyroscope. It compensates for the effects of computer input device rotation. further,
The data input device disclosed in Japanese Patent Laid-Open No. 7-84716 detects the direction and amount of movement of the device based on signals indicating the polarity and amplitude from vibrating gyroscopes arranged at right angles to each other. Furthermore, the handwriting input device disclosed in Japanese Patent Application Laid-Open No. 7-200127 obtains the moving direction and the moving distance of the device based on signals from two acceleration sensors.

[0006] Further, the present invention is not related to a pen-type input device, but is used for, for example, a game machine, and has a moving speed of a human head,
Japanese Patent Application Laid-Open No. 7-29424 detects the position, posture, and the like.
The position sensor disclosed in Japanese Patent Publication No. 0 is provided with an acceleration sensor that detects acceleration in the X-axis direction, the Y-axis direction, and the Z-axis direction, and a gyro that detects angular velocities around the X-axis, Y-axis, and Z-axis. Based on the detected acceleration and angular velocity, a calculation of the strap-down method is performed to detect the moving speed, position, posture, and orientation of the head.

[0007]

However, in the pen type computer input device disclosed in Japanese Patent Laid-Open No. 6-67799, the effect of rotation of the device is corrected, and when the device is accompanied by a dynamic tilt. Cannot be corrected. Since the normal writing operation involves a dynamic tilt of the device, the detection result may be inaccurate.

Further, the data input device disclosed in Japanese Patent Application Laid-Open No. 7-84716 is not suitable for inputting a figure or the like because it detects the rotational movement of the wrist and inputs the moving direction and the moving distance.

Further, in the handwriting input device disclosed in Japanese Patent Laid-Open No. 7-200127, there is no correction means for the inclination of the device, so that the detection result may be inaccurate.

The position sensor described in Japanese Patent Application Laid-Open No. 7-294240 spatially detects the moving speed, position, posture, and orientation of the head, and thus employs a complicated calculation process. Since the pen-type input device is required to be miniaturized, the moving direction and the moving distance on the writing surface must be accurately detected by simple arithmetic processing.

The present invention has been made to solve the above disadvantages, and an object of the present invention is to obtain a small pen type input device for accurately detecting a handwriting input with a simple structure.

[0012]

A pen type input device according to the present invention has three acceleration sensors, three gyros, a coordinate calculation unit, a neural network calculation unit and a post-processing unit, and three acceleration units. The sensors are each in the Xs axis direction of the pen axis coordinate system (Xs, Ys, Zs) with the pen axis as the Zs axis, and Y.
The accelerations in the s-axis direction and the Zs-axis direction are detected, the three gyros detect the rotational angular velocities around the Xs axis, the Ys axis, and the Zs axis, respectively, and the coordinate calculation unit and the initial rotation angle calculation unit and the rotation angle change. The calculation unit, the rotation angle calculation unit during writing, the coordinate conversion calculation unit, and the movement amount calculation unit are provided, and the initial rotation angle calculation unit calculates the rotation angle of the pen shaft based on the acceleration detected by the three acceleration sensors in the unwritten state. The initial value is calculated, the rotation angle change calculation unit calculates the change in rotation angle by integrating the angular velocities detected by the three gyros in the writing state, and the rotation angle calculation unit during writing is calculated by the initial rotation angle calculation unit. The rotation angle of the pen axis during writing is calculated based on the initial value of the rotation angle and the change in the rotation angle calculated by the rotation angle change calculation unit, and the coordinate conversion calculation unit calculates the rotation angle of the pen shaft during writing. The rotation angle of the pen axis and the pen axis coordinate system (Xs Ys, gravity coordinate system in which the axis extending in the direction of gravitational acceleration based on the acceleration due Zs) in Zg axis (X
g, Yg, Zg), and the movement amount calculation unit calculates the acceleration in the gravity coordinate system (Xg, Y) calculated by the coordinate conversion calculation unit.
The moving direction and the moving amount of the pen tip portion are calculated based on the acceleration of g, Zg), and accurate coordinate input is performed by correcting the influence of the change in the rotation of the pen shaft.

The neural network operation unit extracts the characteristics of the recognition target pattern on the pen tip calculated by the coordinate calculation unit based on the moving direction and the moving amount, and based on the characteristics of the extracted recognition target pattern and the result of learning in advance. The intermediate recognition result of the input pattern is calculated, and the post-processing unit determines the input pattern based on the intermediate recognition result of the input pattern of the neural network operation unit to increase the pattern recognition rate and shorten the pattern recognition time.

The storage unit stores the movement direction and the movement amount of the pen tip portion calculated by the coordinate calculation unit, and the neural network calculation unit recognizes based on the movement direction and the movement amount of the pen tip unit stored in the storage unit. The feature of the target pattern is extracted, the intermediate recognition result of the input pattern is calculated based on the extracted feature of the recognition target pattern and the result of learning in advance, the pattern recognition rate is increased, the pattern recognition time is shortened, and the neural network operation unit Lighten the load on.

Further, the preprocessing unit standardizes the features of the recognition target pattern extracted by the coordinate calculation unit and removes singular points,
Furthermore, the pattern recognition time is shortened.

Further, the number of data indicating the features of the recognition target pattern stored in the storage unit is kept constant, and the pattern recognition rate is further increased.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION The pen type input device of the present invention is for inputting characters, symbols, figures, etc. to a computer device or the like. The pen-type input device of the present invention detects accelerations in the unwritten state in the Xs-axis direction, the Ys-axis direction, and the Zs-axis direction of the pen-axis coordinate system (Xs, Ys, Zs) with the pen axis as the Zs axis, Detected pen axis coordinate system (Xs, Ys, Zs)
The initial value of the rotation angle of the pen axis is obtained from the acceleration in each axis direction of. Further, the pen-type input device detects rotational angular velocities around the Xs axis, the Ys axis, and the Zs axis of the pen axis coordinate system (Xs, Ys, Zs) during writing to detect changes in the rotational angle of the pen axis. . By this, the rotation angle of the pen axis at the time of writing is obtained, and the acceleration in the gravity coordinate system (Xg, Yg, Zg) in which the direction of gravity acceleration is the Zg axis is based on the acceleration in the pen axis coordinate system (Xs, Ys, Zs). To calculate. The pen-type input device uses the gravity coordinate system (Xg, Yg, Z
Neural network calculation is performed on the basis of the acceleration in g) to recognize the input pattern to improve the recognition accuracy of the input pattern and at the same time to shorten the recognition time.

The pen-type input device has three acceleration sensors, three gyros, a coordinate calculation section, a neural network calculation section, and a post-processing section. The three acceleration sensors are respectively in the Xs axis direction of the pen axis coordinate system (Xs, Ys, Zs),
The accelerations in the Ys axis direction and the Zs axis direction are detected. The three gyros detect rotational angular velocities around the Xs axis, the Ys axis, and the Zs axis, respectively. The calculation unit includes an initial rotation angle calculation unit, a rotation angle change calculation unit, a writing rotation angle calculation unit, a coordinate conversion calculation unit, and a movement amount calculation unit. The initial rotation angle calculation unit calculates the initial value of the rotation angle of the pen axis based on the acceleration of the pen axis coordinate system (Xs, Ys, Zs) detected by the three acceleration sensors in the unwritten state. The rotation angle change calculation unit calculates the change in the rotation angle of the pen shaft based on the angular velocities detected by the three gyros in the writing state. The rotation angle calculator during writing is based on the initial value of the rotation angle of the pen shaft calculated by the initial rotation angle calculation unit and the change in the rotation angle of the pen shaft calculated by the rotation angle change calculation unit. Find the corner. The coordinate conversion calculation unit is based on the rotation angle of the pen axis during writing detected by the rotation angle calculation unit during writing and the acceleration of the pen axis coordinate system (Xs, Ys, Zs) detected by the acceleration sensor based on the gravity coordinate system (Xg, Yg, Zg) acceleration is calculated. The movement amount calculation unit calculates the movement direction and the movement distance of the pen tip in the gravity coordinate system (Xg, Yg, Zg) based on the acceleration converted by the coordinate conversion calculation unit.

The neural network operation unit extracts the feature of the recognition target pattern based on the moving direction and the movement amount of the pen tip calculated by the coordinate operation unit, and based on the extracted feature of the recognition target pattern and the result of learning in advance. The intermediate recognition result of the input pattern is calculated. Here, the neural network operation unit is configured by a neural network that analyzes the input signal into some features based on cooperation, competitive action, and self-organizing ability and integrates them. For example, the neural network operation unit actually inputs information to be recognized in advance, samples the data, inputs these data to the input layer, prepares neurons for the number of patterns to be classified in the output layer, and prepares each pattern and the neuron. Correspond. For example, the back propagation method (back propagation method) is learned so that when a certain information is input, the output of the corresponding neuron becomes “1” and the output of the other neurons becomes “0”. The neural network operation unit performs the neural network operation with the same coupling coefficient as the coupling coefficient thus obtained. The post-processing unit selects the one having the largest neuron output based on the intermediate recognition result of the input pattern of the neural network operation unit and judges the input pattern, thereby increasing the pattern recognition rate and simultaneously shortening the pattern recognition time.

Since the output from the coordinate calculation unit is a time series, the neural network calculation unit must be of recurrent type. On the other hand, the calculation result of the coordinate calculation unit may be temporarily stored in the storage unit, and the neural network calculation unit may perform the neural network calculation using the calculation result of the coordinate calculation unit stored in the storage unit. Thereby, the structure of the neural network can be made hierarchical.

Further, the data showing the characteristics of the recognition target pattern extracted by the coordinate processing unit in the pre-processing unit is pre-processed into data suitable for the neural network calculation so as to increase the pattern recognition rate in the neural network calculation unit. May be. The preprocessing unit standardizes the data indicating the characteristics of the recognition target pattern by, for example, scaling, binarization, or a combination thereof, and removes the data when apparently abnormal data is input.

Further, generally, the moving speed of the pen-type input device is not constant, and the number of data of the calculation result of the coordinate calculating part varies according to the moving speed of the pen-type input device. If the number of data is too large, the operation speed of the neural network operation unit becomes slow, and if the number of data is too small, the pattern recognition rate may decrease. Therefore, a data processing unit is provided, and when the actual number of data in the data processing unit is small, the data is interpolated, and when the number of data is large, the data is thinned and the number of data of the calculation result of the coordinate calculation unit stored in the storage unit is kept constant. You may keep it.

[0023]

FIG. 1 is a block diagram of a pen-type input device according to an embodiment of the present invention. As shown in the figure, the pen type input device 1 includes acceleration sensors 2a, 2b, 2c, gyros 3a, 3b,
3 c, a coordinate calculation unit 4, a storage unit 5, a neural network calculation unit 6, a post-processing unit 7, and a power supply unit 8. The acceleration sensors 2a, 2b, 2c are provided in the Xs-axis direction, the Ys-axis direction, and the Zs-axis direction that are orthogonal to the Zs-axis when the pen axis 9 is the Zs-axis. , Ys axis direction and Zs axis direction acceleration are detected. The acceleration sensors 2a, 2b, 2c may be of a piezo resistance type, a piezoelectric type, or a capacitance type. Gyro 3a, 3b, 3c is X
The angular velocities around the s axis, the Ys axis, and the Zs axis are detected. In the following description, the pen shaft 9 is set to Z unless otherwise specified.
The coordinate system with the s axis is the pen axis coordinate system (Xs, Ys, Zs)
The two axes orthogonal to the pen axis 9 will be described as the Xs axis and the Ys axis. The axis extending in the direction of gravitational acceleration is Zg
A coordinate system having axes is called a gravity coordinate system (Xg, Yg, Zg), and two axes orthogonal to the Zg axis are called Xg axis and Yg axis.

As shown in FIG. 2, the coordinate calculation section 4 uses an A / D
Converters 41a to 41f, low pass filters 42a to 42
f, a high-pass filter 43, an initial rotation angle calculation unit 44, a rotation angle change calculation unit 45, a writing rotation angle calculation unit 46, a coordinate conversion calculation unit 47, and a movement amount calculation unit 48. The A / D converters 41a to 41f are the acceleration sensors 2a and 2f, respectively.
The analog signals from b, 2c and gyros 3a, 3b, 3c are converted into digital signals. Low pass filter 42
a to 42f are acceleration sensors 2a, 2b, 2c and gyros 3a, 3 generated by the frictional force between the pen tip portion 8 and the writing surface.
The high frequency components of the signals from b and 3c are cut off. The high-pass filter 43 uses, for example, 10 Hz acceleration sensors 2a, 2
The high frequency components due to the frictional force of the signals from b, 2c and gyros 3a, 3b, 3c are extracted.

The initial rotation angle calculation unit 44 detects the rotation angle of the pen shaft 9 based on the acceleration in the pen axis coordinate system (Xs, Ys, Zs) detected by the three acceleration sensors 2a, 2b, 2c in the unwritten state. The initial values θ0, φ0 and Ψ0 of are calculated. The rotation angle change calculation unit 45 includes three gyros 3a, 3b, 3 in the writing state.
Based on the rotational angular velocities P, Q and R detected by c, the changes Δθ, Δφ and ΔΨ in the rotational angle of the pen shaft 9 are calculated. The rotation angle calculation unit 46 during writing writes the initial values θ0, φ0 and Ψ0 of the rotation angle of the pen shaft 9 calculated by the initial rotation angle calculation unit 44 and the change Δθ of the rotation angle of the pen shaft 9 calculated by the rotation angle change calculation unit 45. , Δφ and ΔΨ based on the rotation angles θ, φ and Ψ of the pen shaft 9 during writing.
Ask for. Here, the during-writing rotation angle calculation unit 46 calculates the changes Δθ, Δφ, and ΔΨ of the rotation angles with Ψ0 set to zero.
The coordinate conversion calculation unit 47 detects the rotation angles θ, φ and Ψ of the pen shaft 9 during writing detected by the rotation angle calculation unit during writing and the pen axis coordinate system (Xs, Y) detected by the acceleration sensors 2a, 2b, 2c.
Gravity coordinate system (Xg, Y
g, Zg) to calculate the acceleration. Movement amount calculation unit 48
Is the gravity coordinate system (Xg, Y calculated by the coordinate conversion calculation unit 47.
The moving direction and the moving distance of the pen tip 10 on the writing surface are calculated based on the acceleration of g, Zg).

Here, the calculation of the rotation angle will be described. The conversion from the gravity coordinate system to the pen axis coordinate system can be performed by the following equation.

[0027]

[Equation 1]

This equation is converted into the pen axis coordinate system (Xs, Ys, Z
The transformation formula from s) to the gravity coordinate system (Xg, Yg, Zg) is as follows.

[0029]

[Equation 2]

The above equation is approximated by a linear approximation equation to form a conversion equation for the acceleration vector. Where Axs, Ays and Azs
Are acceleration sensors 2a, 2b, 2c in the pen axis coordinate system.
Is an acceleration vector detected by Axg, Ayg, and Azg are acceleration vectors in the gravity coordinate system.

[0031]

(Equation 3)

The acceleration vector Ax
By substituting s, Ays, Azs and rotation angles θ, φ, Ψ, acceleration vectors Axg, Ayg, Azg on the writing surface can be obtained.

On the other hand, the acceleration in the stationary state can be expressed by the following equation.

[0034]

(Equation 4)

Substituting this, the acceleration sensors 2a, 2a
The acceleration detected by b and 2c can be expressed by the following equation.

[0036]

(Equation 5)

Thus, the rotation angles θ 0, φ 0, Ψ of the pen shaft 9 in the stationary state in the gravity coordinate system (Xg, Yg, Zg).
You can ask for 0.

Since three equations can be established for the rotation angles θ0 and φ0 in the stationary state, the gravitational acceleration g can also be treated as an unknown number, and the stationary state can be obtained without defining the value of g. The absolute values of the rotation angles θ0 and φ0 of can be calculated.

When the rotational angular velocities of the axes Xs, Ys, Zs of the pen axis coordinate system (Xs, Ys, Zs) are P, Q, R,
The relationship between the rotational angular velocities P, Q, R and the rotational angle changes ΔΨ, Δθ, Δφ can be obtained by the following formula.

[0040]

(Equation 6)

The neural network operation unit 6 extracts the feature of the recognition target pattern input based on the moving direction and the movement amount of the pen tip unit 10 calculated by the coordinate operation unit 4, and learns in advance with the extracted feature of the recognition target pattern. The intermediate recognition result of the input pattern is calculated based on the result. Here, the neural network operation unit 6 cooperates as shown in FIG.
It is constructed by a recurrent neural network that analyzes the input signal into some features based on the competitive action and self-organizing ability and integrates them. For example, the neural network operation unit 6 actually inputs information to be recognized in advance, samples the data, inputs these data to the input layer 61, and prepares in the output layer 62 neurons for the number of patterns to be classified,
Each pattern is associated with a neuron. When certain information is input, the output of the corresponding neuron is "1"
Then, the back propagation method (back propagation method) or the like is used for learning so that the outputs of the other neurons become “0”. The neural network operation unit 6 performs the neural network operation with the same coupling coefficient as the coupling coefficient thus obtained. The post-processing unit 7 selects the one having the largest neuron based on the intermediate recognition result of the input pattern of the neural network operation unit 6 and determines the input pattern.

The operation of the pen type input device 1 having the above structure will be described with reference to the flowchart of FIG.

The acceleration sensors 2a, 2b and 2c detect accelerations in the Xs direction, the Ys direction and the Zs direction, respectively. The high-pass filter 43 extracts a high-frequency component from the signals from the acceleration sensors 2a, 2b, 2c and the gyros 3a, 3b, 3c input via the A / D converters 41a-41f, for example, at 10 Hz as a boundary, and extracts one of them. When the signal of 1 includes a high frequency component, the signal of the high frequency component is output to output a signal indicating whether or not writing is in progress. In this way, since the high frequency signal generated by the frictional force between the pen tip portion 10 and the writing surface is detected to determine whether or not the writing is being performed, it is possible to easily and accurately detect whether or not the writing is being performed.

When the initial rotation angle calculation unit 44 does not receive a signal indicating that writing is in progress from the high-pass filter 43, Xs
Signals from the acceleration sensor 2a for the axis, the acceleration sensor 2b for the Ys axis, and the acceleration sensor 2c for the Zs axis are input to calculate initial values θ0, φ0, and Ψ0 of the rotation angle of the pen shaft 10 (step S1).

When the rotation angle change calculation unit 45 receives a signal indicating that writing is in progress from the high-pass filter 43 (step S2), the rotation angle velocities P, Q, R detected by the three gyros 3a, 3b, 3c are detected. Based on this, changes Δθ, Δφ and ΔΨ in the rotation angle of the pen shaft 9 are calculated (step S3).

The in-writing rotation angle calculation unit 46 calculates the initial values θ0, φ0, Ψ0 of the rotation angle of the pen shaft 9 calculated by the initial rotation angle calculation unit 44 and the pen calculated by the rotation angle change calculation unit 45 as described above. Based on the changes Δθ, Δφ, ΔΨ in the rotation angle of the shaft 9, φ =
φ0 + Σ (Δφ) dt, θ = θ0 + Σ (Δθ) dt, Ψ = Ψ0
+ Σ (ΔΨ) dt is calculated and the rotation angle θ of the pen axis during writing,
φ and Ψ are obtained (step S4). Coordinate conversion calculation unit 47
Is the gravity coordinate based on the rotation angle during writing detected by the rotation angle during writing unit 46 and the acceleration Axs, Ays, Azs of the pen axis coordinate system (Xs, Ys, Zs) detected by the acceleration sensors 2a, 2b, 2c. Acceleration Axg, A by the system (Xg, Yg, Zg)
yg and Azg are calculated (step S5). The acceleration Ax based on the gravity coordinate system (Xg, Yg, Zg) based on the accelerations Axs, Ays, and Azs in the pen axis coordinate system (Xs, Ys, Zs).
To calculate g, Ayg, and Azg, the conversion formula already described is used. The movement amount calculation unit 48 calculates the movement direction and movement distance of the pen tip unit 10 based on the acceleration of the pen tip unit 10 converted by the coordinate conversion calculation unit 47 (step S6).

The neural network operation unit 6 extracts the features of the recognition target pattern based on the movement direction and the movement amount of the pen tip unit 10 calculated by the coordinate operation unit 4, and the input pattern based on the extracted features of the recognition target pattern. The intermediate recognition result of is calculated (step S7). The post-processing unit 7 inputs the input pattern intermediate recognition result of the neural network operation unit 6 and determines the input pattern (step S8). As a result, it is possible to input characters, figures, and their constituent elements, and to recognize the input pattern accurately and quickly. Here, for example, a method such as pattern matching is used for recognizing a character when a character component is input.

The pen type input device 1 operates as described above (step S
3 to S8) are repeated while the high-pass filter 43 outputs a signal indicating that writing is in progress, and a figure or the like is input (step S9). As described above, by eliminating the influence of the tilt of the pen-type input device 1 in the gravity coordinate system (Xg, Yg, Zg), it is possible to accurately input a figure or the like.

In the above embodiment, the coordinate calculation unit 4
Since the calculation result of is directly input to the neural network calculation unit 6, the neural network calculation unit 6 is composed of a recurrent type neural network, but the calculation result of the coordinate calculation unit 4 is temporarily stored in the storage unit 5 as shown in FIG. The calculation result of the coordinate calculation unit 4 stored in the storage unit 5 may be input to the neural network calculation unit 6. As a result, the neural network operation unit 6
Can be composed of a hierarchical neural network. Further, by temporarily storing the calculation result of the coordinate calculation unit 4 in the storage unit 5, it becomes possible to recognize an input pattern that requires a lot of time series data.

Further, as shown in FIG. 6, a preprocessing section 11 for standardizing the calculation result of the coordinate calculation section 4 and removing the singular point may be provided. The pre-processing unit 11 uses scaling, 2
The calculation result of the coordinate calculation unit 4 is standardized by digitization or a combination thereof, and when apparently abnormal data is input, this data is removed. In this way, the pattern recognition rate in the neural network operation unit 6 can be increased by processing the operation result of the coordinate operation unit 4 into data suitable for the neural network operation in advance.

Further, as shown in FIG. 7, a data processing unit 12 may be provided to keep the number of data of the calculation result of the coordinate calculation unit 4 stored in the storage unit 5 constant. Generally, the movement speed of the pen-type input device 1 is not constant, and the number of data of the calculation result of the coordinate calculation unit 4 changes according to the movement speed of the pen-type input device 1. When the number of data is too large, the operation speed of the neural network operation unit 6 becomes slow, and when the number of data is too small, the pattern recognition rate may decrease. Therefore, the pattern recognition rate of the neural network calculation unit 6 can be improved by interpolating the data when the actual data number is small in the data processing unit 12 and thinning out the data when the actual data number is large. Here, for the data interpolation, for example, interpolation processing by polynomial approximation is performed, and for the data thinning, a method of evenly thinning is used.

Further, if the mounting positions of the acceleration sensors 2a, 2b, 2c are apart from the pen tip portion 10, an error may occur between the actual acceleration and the detected acceleration due to the rotation, so this may occur during the coordinate conversion of the acceleration. The error may be corrected.

[0053]

As described above, the present invention calculates the acceleration in the gravity coordinate system based on the acceleration in each axis direction in the pen axis coordinate system and the rotational angular velocity around each axis, and calculates the acceleration in the calculated gravity coordinate system. Since the moving direction and the moving distance of the pen tip portion are calculated based on, the moving direction and the moving distance of the pen tip portion moving on the writing surface can be accurately detected by a small-sized device.

Furthermore, the movement direction and movement amount of the pen tip portion are input to the neural network to output the pattern intermediate recognition result, and the final recognition result is output based on the intermediate recognition result of the input pattern. The input pattern can be determined.

Further, since the moving direction and moving amount of the pen tip portion are temporarily stored, the input pattern can be promptly judged by using the hierarchical neural network.

Furthermore, since the moving direction and moving amount of the pen tip portion are standardized to remove the singular point, the pattern recognition speed of the neural network can be increased.

Further, since the number of pieces of data on the movement direction and movement amount of the pen tip portion is kept constant, the pattern recognition rate of the neural network can be increased.

[Brief description of drawings]

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

FIG. 2 is a configuration diagram of a calculation unit.

FIG. 3 is a configuration diagram of a neural network operation unit.

FIG. 4 is a flowchart showing an operation of the pen type input device.

FIG. 5 is a configuration diagram in a case where the calculation result of the coordinate calculation unit is temporarily stored in a storage unit.

FIG. 6 is a configuration diagram when a pre-processing unit is provided.

FIG. 7 is a configuration diagram when a data processing unit is provided.

[Explanation of symbols]

 1 Pen-Type Input Device 2 Acceleration Sensor 3 Gyro 4 Coordinate Calculation Unit 43 High-Pass Filter 44 Initial Rotation Angle Calculation Unit 45 Rotation Angle Change Calculation Unit 46 Writing Rotation Angle Calculation Unit 47 Coordinate Conversion Calculation Unit 48 Movement Distance Calculation Unit 5 Storage Unit 6 Neural network operation section 7 Post-processing section 9 Pen axis 10 Pen tip section 11 Pre-processing section 12 Data processing section

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Norihiko Murata 1-3-6 Nakamagome, Ota-ku, Tokyo Inside Ricoh Co., Ltd. (72) Etsuko Fujisawa 1-3-6 Nakamagome, Ota-ku, Tokyo In stock company Rico (72) Inventor Takao Inoue 1-3-6 Nakamagome, Ota-ku, Tokyo Inside stock company Rico

Claims (4)

[Claims] 1. A pen axis coordinate system having three acceleration sensors, three gyros, a coordinate operation section, a neural network operation section, and a post-processing section, each of the three acceleration sensors having a pen axis of Zs axis. The accelerations of (Xs, Ys, Zs) in the Xs axis direction, the Ys axis direction, and the Zs axis direction are detected, and the three gyros are respectively around the Xs axis, around the Ys axis, and Zs.
The rotation angular velocity around the axis is detected, and the coordinate calculation unit includes an initial rotation angle calculation unit, a rotation angle change calculation unit, a writing rotation angle calculation unit, a coordinate conversion calculation unit, and a movement amount calculation unit. In the unwritten state, the initial value of the rotation angle of the pen axis is calculated based on the acceleration detected by the three acceleration sensors, and the rotation angle change calculation unit integrates the angular velocities detected by the three gyros in the written state and rotates. The rotation angle calculation unit during writing calculates the rotation angle of the pen shaft based on the initial value of the rotation angle calculated by the initial rotation angle calculation unit and the rotation angle change calculated by the rotation angle change calculation unit. The angle is calculated, and the coordinate conversion calculation unit calculates gravity based on the rotation angle of the pen axis during writing calculated by the rotation angle calculation unit during writing and the acceleration by the pen axis coordinate system (Xs, Ys, Zs) detected by the acceleration sensor. Gravity coordinate system (Xg, Yg, g), the acceleration of the pen tip is calculated, and the movement amount calculation unit calculates the movement direction and the movement amount of the pen tip based on the acceleration of the gravity coordinate system (Xg, Yg, Zg) calculated by the coordinate conversion calculation unit. Then, the neural network operation unit extracts the feature of the recognition target pattern based on the moving direction and the movement amount of the pen tip calculated by the coordinate operation unit, and inputs based on the extracted feature of the recognition target pattern and the previously learned result. A pen-type input device characterized in that an intermediate recognition result of a pattern is calculated, and a post-processing unit judges an input pattern based on an intermediate recognition result of the input pattern by a neural network operation unit. 2. A storage unit for storing the movement direction and movement amount of the pen tip portion calculated by the coordinate calculation unit, wherein the neural network calculation unit is based on the movement direction and movement amount of the pen tip unit stored in the storage unit. 2. The pen-type input device according to claim 1, wherein the features of the recognition target pattern are extracted in step S1, and the intermediate recognition result of the input pattern is calculated based on the extracted features of the recognition target pattern and the result of learning in advance. 3. The pen-type input device according to claim 1, further comprising a preprocessing unit that normalizes the features of the recognition target pattern extracted by the coordinate calculation unit and removes singular points. 4. The pen type input device according to claim 4, wherein the number of data representing the features of the recognition target pattern stored in the storage unit is kept constant.