JP2010118798A - Noise eliminating device, noise eliminating method, and operation correcting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an operation correcting device capable of eliminating a noise signal that temporally changes. <P>SOLUTION: A meal support apparatus is provided with: a force sensor for outputting a main signal corresponding to operation to a manipulator and an input signal IN including a tremor signal; an inputting means 15 in which the input signal IN outputted from the force sensor is inputted; and a tremor frequency estimating means 42 for determining an estimated frequency f<SB>t</SB>(i) corresponding to a frequency at each time of the tremor signal that temporally changes by updating a tap coefficient for minimizing an output of the input signal IN inputted to a notch filter to which an all-pass filter APF is applied as a tremor elimination system 18. Then, by a cutoff filter determined on the basis of the estimated frequency f<SB>t</SB>(i) determined by the tremor frequency estimating means 42, the tremor eliminating means 44 eliminates the tremor signal from the input signal IN to generate an output signal OUT. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a noise removal device, a noise removal method, and an operation correction device, and more particularly to a noise removal device, a noise removal method, and an operation correction device that remove an noise signal from an input signal and output an output signal. is there.

  For example, in medical and welfare settings, research is being actively conducted on assist devices that reduce the burden on caregivers and self-supporting robots that are intended to support the care recipient's independence (see, for example, Patent Document 1). In particular, since meals are an indispensable task every day, both caregivers and care recipients have a heavy burden, and various devices that support meals have been researched and developed.

By the way, a meal support device (motion correction device) that operates a device in accordance with a user's motion has been proposed for patients who have tremors (hereinafter referred to as tremor) in their hands and body. In such a meal support apparatus, the tremor signal (noise signal) included in the haptic signal of the hand and the signal intended by the user (main signal) are accurately separated, and the influence of tremor is removed. It is necessary to operate the apparatus with good operability. From this point of view, many studies on tremor suppression have been conducted. For example, Mr. Nawate tried to correct tremors during mouse operation using the moving average method (Non-Patent Document 1), and J. Gonzalez et al. Developed a digital filtering algorithm using an optimal equalizer in the same field. (Non-Patent Document 2). Furthermore, CN Riviere et al. Used a nonlinear adaptive denoising algorithm to suppress physiological tremor in microsurgery (Non-Patent Document 3), and S. Pledie et al. Had a feedback loop for tremor suppression in impedance control. A man-machine model was created (Non-Patent Document 4).
JP-A-11-253504 Masahiko Nawate, A Painting Tool with Blurring Compensation for People Having Involuntary Hand Motion, IEICE HIP, Vol. 103, pp.59-64, 2004 Gonzalez.JG, Filtering involuntary motion of people with tremor disability using optimal equalization, Proc.IEEE Int. Conf.On Systems, Man and Cybernetics, Vol.3, pp.2402-2407, 1995 Cameron N. Riviere, Toward Active Tremor Canceling in Handheld Microsurgical Instruments, IEEE Trans. On Robotics and Automation, Vol. 15, No. 5, October, pp.793-800, 2003 Pledgie. S, Tremor suppression through impedance control, IEEE.Trans on Rehabilitation and Engineering, Vol.8, issue.l, pp.53-59, 2000

  However, these methods (1) have a large output delay with respect to input and have operability problems. (2) Measure tremor frequency in advance to suppress only tremor at that frequency. There was a difficulty that could not respond to tremors that were subject to individual differences and time variation.

  Therefore, in view of the above problems inherent in the prior art, the present invention has been proposed to suitably solve these problems, and estimates the frequency of a noise signal such as tremor in real time and is time-variant. It is an object of the present invention to provide a noise signal removal device, a noise signal removal method, and an operation correction device that can remove a noise signal.

In order to solve the above-mentioned problem and to achieve the intended purpose suitably, the noise removing device according to claim 1 is:
Input means that is detected by the motion detection unit and receives an input signal including a main signal and a noise signal;
Noise frequency estimation that determines the estimated frequency corresponding to the frequency at each time of the time-varying noise signal by updating the tap coefficient that minimizes the output of the input signal input to the notch filter to which the all-pass filter is applied Means,
Noise removing means for removing the noise signal from the input signal to generate an output signal by a cutoff filter determined based on the estimated frequency determined by the noise frequency estimating means;
Output means for outputting the output signal generated by the noise removing means.
According to the first aspect of the present invention, the estimated frequency corresponding to the frequency of the noise signal is determined following the time-varying noise signal, so that the time-variant noise signal can be removed from the input signal.

In the noise removal apparatus according to claim 2, the noise frequency estimation means includes a main signal removal unit that removes a frequency band regarded as the main signal from the input signal, and the main signal removal unit removes the frequency band. The input signal is input to the notch filter.
According to the invention of claim 2, since the frequency band regarded as the main signal is removed from the input signal, it is possible to suppress the main signal from being removed as a noise signal.

In the noise signal removing device according to claim 3,
The noise frequency estimation unit includes a provisional frequency calculation unit that calculates a provisional frequency that is a notch frequency from the tap coefficient, and a determination unit that determines the estimation frequency based on the provisional frequency calculated by the provisional frequency calculation unit. ,
The determination unit
The deviation monitoring amount based on the difference between the provisional frequency at an arbitrary time and the estimated frequency determined immediately before the arbitrary time is greater than a preset first reference value, and the variance of the provisional frequency within a predetermined time is preset. If it is greater than the second reference value, the estimated frequency determined immediately before the arbitrary time is determined as the estimated frequency at the arbitrary time,
When the deviation monitoring amount is smaller than the first reference value or when the variance is smaller than the second reference value, the provisional frequency calculated at the arbitrary time is determined as the estimated frequency at the arbitrary time.
According to the invention of claim 3, when the deviation monitoring amount is larger than the first reference value and the variance is larger than the second reference value, the estimated frequency determined immediately before is set as the estimated frequency at the time. Even when the main signal related to the periodic operation is included in the input signal, it is possible to suppress the main signal from being removed as a noise signal. Further, when the deviation monitoring amount is smaller than the first reference value or when the variance is smaller than the second reference value, the provisional frequency at the time is determined as the estimated frequency, so that the time-varying noise signal is followed. Thus, the estimated frequency can be obtained.

In order to solve the above-mentioned problem and to achieve the intended purpose suitably, a noise signal removal method according to claim 4 is:
An input signal including a main signal and a noise signal detected by the motion detection unit is input to the input means,
The input signal input to the input means is input to a notch filter to which an all-pass filter is applied, and the tap coefficient that minimizes the output is updated to correspond to the frequency at each time of the time-varying noise signal. Determine the estimated frequency,
A cutoff filter determined based on the estimated frequency removes a noise signal from the input signal to generate an output signal,
An output means outputs the generated output signal.
According to the fourth aspect of the present invention, since the estimated frequency corresponding to the frequency of the noise signal is determined following the time-varying noise signal, the time-variant noise signal can be removed from the input signal.

In the noise signal removal method according to the fifth aspect, an input signal from which a frequency band regarded as a main signal is removed is input to the notch filter.
According to the invention of claim 5, since the frequency band considered as the main signal is removed from the input signal, it is possible to suppress the main signal from being removed as a noise signal.

In the noise signal removal method according to claim 6,
Calculate provisional frequency which is notch frequency from tap coefficient,
The deviation monitoring amount based on the difference between the provisional frequency at an arbitrary time and the estimated frequency determined immediately before the arbitrary time is greater than a preset first reference value, and the variance of the provisional frequency within a predetermined time is preset. If it is greater than the second reference value, the estimated frequency determined immediately before the arbitrary time is determined as the estimated frequency at the arbitrary time,
When the deviation monitoring amount is smaller than the first reference value or when the variance is smaller than the second reference value, the provisional frequency calculated at the arbitrary time is determined as the estimated frequency at the arbitrary time.
According to the sixth aspect of the invention, when the deviation monitoring amount is larger than the first reference value and the variance is larger than the second reference value, the estimated frequency determined immediately before is determined as the estimated frequency at the time. Even when the main signal related to the periodic operation is included in the input signal, the main signal is not removed as a noise signal. Further, when the deviation monitoring amount is smaller than the first reference value or when the variance is smaller than the second reference value, the provisional frequency at the time is determined as the estimated frequency, so that the time-varying noise signal is followed. To estimate the estimated frequency.

In order to solve the above-described problem and achieve the desired purpose suitably, the motion correction device according to claim 7 is:
An operation detection unit that outputs an input signal including a main signal and a noise signal according to the operation of the operated means;
An input means for inputting an input signal output from the motion detector;
Noise frequency estimation that determines the estimated frequency corresponding to the frequency of each time-varying noise signal by updating the tap coefficient that minimizes the output of the input signal input to the notch filter to which the all-pass filter is applied Means,
Noise removing means for removing the noise signal from the input signal to generate an output signal by a cutoff filter determined based on the estimated frequency determined by the noise frequency estimating means;
Output means for outputting the output signal generated by the noise removing means;
Drive means for actuating the operation correction target based on the output signal output from the output means.
According to the seventh aspect of the present invention, the frequency of the noise signal can be estimated following the time-varying noise signal, so that the output signal can be output by removing the time-variant noise signal from the input signal. Therefore, the operation to be corrected can be corrected and operated by the output signal obtained by removing the noise signal from the input signal to the operated means.

In the motion correction device according to claim 8, the noise frequency estimation means includes a main signal removing unit that removes a frequency band regarded as the main signal from the input signal, and the main signal removing unit removes the frequency band. An input signal is input to the notch filter.
According to the invention of claim 8, since the frequency band regarded as the main signal is removed from the input signal, it is possible to suppress the removal of the main signal as a noise signal.

In the motion correction device according to claim 9, the noise frequency estimation means includes a provisional frequency calculation means for calculating a provisional frequency that is a notch frequency from the tap coefficient, and the estimation based on the provisional frequency calculated by the provisional frequency calculation means. A determination unit for determining the frequency,
The determination unit
The deviation monitoring amount based on the difference between the provisional frequency at an arbitrary time and the estimated frequency determined immediately before the arbitrary time is greater than a preset first reference value, and the variance of the provisional frequency within a predetermined time is preset. If it is greater than the second reference value, the estimated frequency determined immediately before the arbitrary time is determined as the estimated frequency at the arbitrary time,
When the deviation monitoring amount is smaller than the first reference value or when the variance is smaller than the second reference value, the provisional frequency calculated at the arbitrary time is determined as the estimated frequency at the arbitrary time.
According to the ninth aspect of the present invention, when the deviation monitoring amount is larger than the first reference value and the variance is larger than the second reference value, the estimated frequency determined immediately before is determined as the estimated frequency at the time. Even when the main signal related to the periodic operation is included in the input signal, the main signal is not removed as a noise signal. Further, when the deviation monitoring amount is smaller than the first reference value or when the variance is smaller than the second reference value, the provisional frequency at the time is determined as the estimated frequency, so that the time-varying noise signal is followed. To estimate the estimated frequency.

  According to the present invention, a time-variant noise signal can be estimated in a follow-up manner, and the noise signal can be suitably removed from the input signal.

  Next, the noise removal apparatus, the noise removal method, and the operation correction apparatus according to the present invention will be described below with reference to the accompanying drawings by giving preferred embodiments. In the embodiment, an example in which the motion correction device according to the present invention is adopted as a meal support device that supports a user's meal will be described. Moreover, the noise removal apparatus of this invention is employ | adopted for the tremor removal system with which the meal assistance apparatus which concerns on an Example is provided, and the noise removal method of this invention is performed by the tremor removal system. Further, in the embodiment, a case will be described in which a Parkinson's disease patient who causes tremor in the hand or body uses the meal support apparatus as a user. In the meal support device according to the embodiment, the manipulator (spoon) is configured to operate in response to the user's operation on the manipulator (spoon). That is, the manipulator and spoon according to the embodiment are operated means and are also motion correction targets.

  As shown in FIG. 1, the meal support apparatus 10 according to the embodiment includes a six-axis manipulator 12, a force sensor (motion detection unit) 13 disposed on an upper end portion 38 of the manipulator 12, and the manipulator 12, a mounting portion 16 that is detachably mounted on the handle 14 a of the spoon 14, a driving means (not shown) for driving the manipulator 12, and a noise signal (hereinafter referred to as vibration) due to tremor. A tremor removal system 18 for removing a war signal). Then, the user can move the spoon 14 in a desired direction and speed after the tremor has been removed by operating the manipulator 12 by gripping the handle portion 14a and the attachment portion 16 of the spoon 14. It has become. Note that the root of the manipulator 12 is fixed by a fixing unit 22 to a table 40 used by the user for eating. Further, the tremor removal system 18 corresponds to, for example, a computer arithmetic unit that controls the meal support apparatus 10 in an integrated manner.

  The manipulator 12 includes a first joint part 24, a first arm part 26, a second joint part 28, a second arm part 30, a third joint part 32, a third arm part 34, and a fourth arm in order from the fixed part 22 side. A joint portion 36 and a tip portion 38 are provided, and first to sixth motors (not shown) as drive means are provided at each rotation portion. The output means 20 of the tremor removal system 18 sends an output signal to the first to sixth motors to control the operation, so that the first to third arm portions 26, 30, 34 are rotated. The movement of the manipulator 12 with a high degree of freedom is realized.

  A first joint portion 24 that can be rotated around an axis perpendicular to the table 40 is attached to the fixed portion 22, and the first joint portion 24 is horizontal to the upper surface of the table 40. A first arm portion 26 capable of rotating around the axis is attached. The distal end of the first arm portion 26 is connected to the second arm portion 30 via the second joint portion 28, and the second arm portion 30 is centered on an axis that is horizontal to the upper surface of the table 40. Rotation is possible. At the tip of the second arm portion 30, a third joint portion 32 that can be rotated around an axis that is horizontal with respect to the upper surface of the table 40 is attached. A third arm portion 34 that can be rotated around an axis along the longitudinal direction of the second arm portion 30 is attached to the end portion 32 a of the third joint portion 32. That is, the third arm portion 34 can be rotated in the vertical direction with respect to the table 40 and can be rotated (spinned) about the axis along the longitudinal direction of the third arm portion 34. . Further, the fourth joint portion 36 connects the third arm portion 34 and the distal end portion 38 so that the distal end portion 38 can rotate around the axis line along the longitudinal direction of the third arm portion 34. Composed.

  The mounting portion 16 attached to the distal end portion 38 has a cylindrical shape that can be easily gripped by the user. A known gripping mechanism such as a drill chuck is provided at the tip of the mounting portion 16, and the handle portion 14 a of the spoon 14 can be easily attached to and detached from the mounting portion 16 through the gripping mechanism. In addition, the spoon 14 with which the said attachment part 16 is mounted | worn employ | adopts the general thing used for a meal.

  The force sensor 13 detects a user's operation (force direction, acceleration, etc.) with respect to the manipulator 12 (spoon 14) as an input signal IN. It is electrically connected to input means 15 (described later). The user's motion detected by the force sensor 13 is input to the tremor removal system 18 as an input signal IN. The input signal IN includes a main signal due to an operation intended by the user and a tremor signal due to tremor not intended by the user. Then, the tremor elimination system 18 outputs an output signal OUT obtained by removing the tremor signal from the input signal IN, thereby causing the manipulator 12 (spoon 14) to perform a movement desired by the user excluding the influence of tremor. To get.

As shown in FIG. 2, the tremor elimination system 18 according to the embodiment includes an input unit 15 to which an input signal IN is input from the force sensor 13 and a tremor at each time of a tremor signal that changes with time. A tremor frequency estimating means (noise frequency estimating means) 42 for determining an estimated frequency f _t (i) corresponding to the frequency, and a cutoff filter (FI _n ) determined based on the tremor frequency f _t (i), A tremor removing means (noise removing means) 44 for removing the tremor signal from the input signal IN to generate the output signal OUT, and an output means 20 for outputting the output signal OUT generated by the tremor removing means 44. I have. The tremor frequency estimation means 42 includes a main signal removal unit 46 that removes in advance a frequency band that is regarded as a user's main signal from the input signal IN, and a provisional frequency _fe (i) (described later) at each time. And a determination unit 50 that determines the estimated frequency f _t (i) based on the provisional frequency f _e (i) calculated by the provisional frequency calculation unit 48. Note that each part of the tremor removing means 44 and the tremor frequency estimating means 42 constituting the tremor removing system 18 does not have to be a physically independent member, and may be incorporated into one control unit. Good. That is, each part of the tremor removal system 18 may be classified according to each function, rather than being physically distinct.

(About tremor removal means)
The tremor removal means 44 according to the embodiment employs a band stop filter with a small delay as the cutoff filter FI _n . The band stop filter uses a Chbyshev II type with a flat pass band in order to reduce the influence on the main signal. The delay t _delay is set to a delay amount of 0.1 [sec] or less in consideration of deterioration in operability, the removal bandwidth f _stop = 0.3 [Hz], and the difference between the pass band and the removal band f _diff = 0 .1 [Hz] is set as a fixed parameter, and a set of a plurality of cut-off filters FI _n are set while the center frequency f _c of the cut-off band is overlapped by 1/2 of the removal bandwidth f _stop between 1 to 8 [Hz]. (See Equation 1). FIG. 3 shows an example of the cutoff filter FI _n . The designed cutoff filter FI _n is switched according to the estimated frequency f _t (i) determined by the tremor frequency estimating means 42 as will be described later. Then, the tremor removing unit 44 removes the tremor signal from the input signal IN by the determined cutoff filter FI _n to generate the output signal OUT. The removal bandwidth f _stop was confirmed to be about 0.2 [Hz] from the peak frequency of tremor frequency based on the tremor analysis of Parkinson's disease patients performed in advance. Was set to 0.3 [Hz]. Further, the overlap of the filter coefficients is not necessarily limited to 1/2 of the removal bandwidth f _stop , and may be overlapped by 1/3 and 1/4, for example.

(Main signal removal unit)
The main signal removing unit 46 is for preventing a frequency band regarded as a user's main signal from the input signal IN from being input to the provisional frequency calculating unit 48 (notch filter 17 described later). In the embodiment, a high-pass filter is employed as the main signal removal unit 46. That is, the input signal IN input from the force sensor 13 is a colored signal having power (amplitude) at various frequencies including the user's main signal and tremor signal. Therefore, in order to prevent the adaptive algorithm from converging to the highest power frequency (mainly low frequency), the input signal IN is regarded as the main signal by applying a high-pass filter before input to the provisional frequency calculation unit 48. Frequency band to be removed.

  The frequency (cut-off frequency) removed by the main signal removal unit 46 is set to be equal to or lower than the tremor or involuntary movement frequency, and is set to, for example, 1.0 [Hz] based on the analysis result. That is, since the user's meal operation (main signal) has a frequency lower than 1.0 [Hz], the frequency band lower than the frequency is regarded as the main signal and is sent to the provisional frequency calculation unit 48. Input is blocked. However, as will be described later, when the main signal includes a periodic operation of the user higher than the cutoff frequency, the main signal is not cut by the main signal removing unit 46. Therefore, the main signal related to this periodic operation is input to the provisional frequency calculation unit 48 together with the tremor signal. Note that the cut-off frequency can be changed as appropriate in accordance with the degree of tremor of the user and the operation intended by the user.

(Provisional frequency calculator)
As shown in FIG. 4, the provisional frequency calculation unit 48 applies an IIR type second-order all-pass filter APF to the notch filter 17 and calculates the provisional frequency _fe (i) using an adaptive algorithm. . The all-pass filter APF is a filter whose amplitude characteristic is 1 at all frequencies and whose phase is inverted only at a certain frequency. The transfer function of the IIR type second-order all-pass filter APF is expressed by Equation 2. This IIR type second-order all-pass filter APF is used as a notch filter 17 as shown in Equation 3. Between α and ρ and the normalized frequency ω _c whose phase is inverted, there is a relationship as shown in Equation 4, where α is a coefficient (tap coefficient) that determines the frequency (notch frequency) where the phase is inverted, and ρ is It is a coefficient that determines the steepness of phase inversion (steepness of notches). The phase inversion tends to become steep as the value of ρ approaches “1”.

Since the notch frequency of the notch filter 17 is determined by the tap coefficient α, the notch filter 17 updates the tap coefficient α so that the tremor frequency of the tremor signal matches the notch frequency. An adaptive algorithm for updating the tap coefficient α is derived. The basic steepest descent method is expressed by Equation 5. Here, ▽ (i) is a gradient vector at time k, and μ is a step size parameter that determines the width of the update.
From Equation 5, an equation for updating the tap coefficient α is derived. First, Equation 5 becomes Equation 6. However, according to the LMS (Least Mean Square) algorithm, A is the estimated value of ▽ (i), and the estimated value of the evaluation function J = E [e (i) ² ] is B as shown in Equation 7 (E [•] Represents the expected value operation). In addition, using the expressions 8 and 9 from FIG. Therefore, the updating formula of the tap coefficient α is expressed by Equation 11.

This adaptive algorithm updates the coefficient α so as to minimize the error e (i) (the output of the notch filter 17). By updating, the tap coefficient α corresponding to the tremor frequency in the input signal can be obtained. Then, the provisional frequency calculation unit 48 calculates the notch frequency at time i by updating the coefficient α according to the tremor frequency that changes with time, and sets this as the provisional frequency _fe (i) (Equation 12 reference). That is, the provisional frequency f _e (i) is a frequency that is provisionally regarded as the tremor frequency of the tremor signal at the time (the provisional frequency f _e (i) is the estimated frequency f _t (i). Is determined by the determination unit 50 described later). Since the adaptive algorithm diverges depending on the values of the coefficients α and ρ (constant), the threshold needs to be obtained by simulation. In this embodiment, for example, α> −1.99495 when ρ = 0.995. When this estimation method is used, the resolution can be arbitrarily determined by the step size parameter μ and the amount of calculation is small, so that the update can be performed for each sample. However, since the magnitude of μ is in a trade-off relationship between the estimation accuracy and the convergence speed, it is necessary to appropriately select a value suitable for the system.

(About judgment part)
The determination unit 50 functions so as not to remove the main signal when a main signal related to a periodic operation such as writing a character by the user collecting with the spoon 14 is input. As described above, the main signal removal unit 46 removes a signal lower than the cutoff frequency, but when the user performs a periodic operation higher than the cutoff frequency, the signal passes through the main signal removal unit 46. Will pass. Then, a signal related to the periodic operation is input to the provisional frequency calculation unit 48, and the provisional frequency calculation unit 48 calculates the number of provisional frequencies f _e (i) for the signal. Therefore, when the provisional frequency _fe (i) calculated by the provisional frequency calculation unit 48 is caused by a tremor signal, the determination unit 50 determines the provisional frequency _fe (i) as the estimated frequency f _t ( determined as i), if due to the periodic operation is determined estimated determined immediately before the frequency f _t of the (i-1) as the estimated frequency f _t (i) at that time. Specifically, the determination unit 50 controls the provisional frequency _fe (i) at the time i calculated by the provisional frequency calculation unit 48 as shown in the block diagram of FIG. In FIG. 5, h (i) and v (i) are evaluation functions as shown in Equations 13 and 14, respectively.

Here, λ is a forgetting factor. Equation 13 shows a deviation monitoring amount h (i) for monitoring a deviation between the provisional frequency _fe (i) at the time i and the estimated frequency f _t (i-1) just before. Thus, by calculating recursively using the forgetting factor λ, it is possible to have a weight for past data. That is, the discrimination based on the deviation monitoring amount h (i) is generally referred to as “tremor that can be observed for a relatively long time when tremor is discriminated from among a number of peaks in the frequency domain. The tremor discriminant method “can do” is used. Equation 14 represents the variance v (i) of the provisional frequencies _fe (ik) to _fe (i) from time ik to time i (within a predetermined time), and monitors the convergence of the adaptive algorithm. .

That is, the determination unit 50, the deviation monitor the amount h (i) is greater than the first reference value h _r a preset (Yes in step S1 in FIG. 5), and the dispersion v (i) is set in advance been if greater than the second reference value v _r (No in step S2 in FIG. 5), the estimated frequency f _t of the estimated was determined immediately before the time i frequency f _t a (i-1) at time i (i) is determined (step S5 in FIG. 5). Here, the deviation monitor the amount h (i) is greater than the first reference value h _r is the provisional frequency f _e (i) means that changes significantly, a large frequency in the input signal IN varies periodic signal Means that. Meanwhile, the dispersion v (i) may be larger than the second reference value v _r is the provisional frequency f _e (i) is meant to rapidly that changed within the predetermined time. By satisfying both of these conditions, it means that a periodic signal different from the tremor signal is suddenly added within a predetermined time, and the determination unit 50 determines that the periodic signal is based on the periodic motion of the user. As such, the provisional frequency _fe (i) at the time i is not determined as the estimated frequency _ft (i).

Incidentally, when the deviation monitor the amount h (i) is smaller than the first reference value h _r (No in step S1 in FIG. 5), or, the dispersion v (i) if is smaller than the second reference value v _r (FIG. 5 The determination unit 50 determines that the provisional frequency _fe (i) is not based on the periodic motion (is a tremor signal), and determines the provisional frequency _fe (i) as the estimated frequency f. _t (i) (steps S4 and S3 in FIG. 5). That is, when the tremor frequency changes slowly, the determination unit 50 updates the estimated frequency f _t (i) according to the change of the tremor frequency. The first reference value h _r and the second reference value v _r are determined on the basis of the change in the evaluation function at the time when data obtained from the analysis result of the Alzheimer's disease patient is input. When the estimation frequency f _t (i) is determined by the determination unit 50, the tremor removal unit 44 determines n in the equation 1 so that the estimation frequency f _t (i) satisfies the equation 15, and the tremor removal The center frequency f _c (n) of the cutoff region of the means 44 is determined. That is, the tremor removing unit 44 determines the cutoff filter FI _n of the tremor removing unit 44 according to the estimated frequency f _t (i) finally determined by the determination unit 50.

  The output means 20 is connected to the first to sixth motors and outputs the output signal OUT generated by the tremor removing means 44 to the first to sixth motors. Then, the first to sixth motors are driven based on the output signal OUT, and the manipulator 12 (spoon 14) is operated.

(Operation of Example)
Next, the operation of the meal support apparatus 10 according to the embodiment will be described with reference to the flowchart of FIG. 6 together with the tremor removal method (noise removal method) by the tremor removal system 18. When the user grasps the handle 14a of the spoon 14 and applies a force to convey the food squeezed with the spoon 14 to the mouth, the force sensor 13 detects the user's movement and inputs based on this A signal IN is generated (step S1). At this time, the input signal IN includes a tremor signal due to tremor in addition to the main signal of the user.

The input signal IN generated by the force sensor 13 is input to the input unit 15 of the tremor removal system 18 (step S2), and is first input to the main signal removal unit 46 of the tremor frequency estimation unit 42 (step S2). Step S3). Then, since the user's main signal has a lower frequency (lower than the cut-off frequency of 1.0 [Hz]) compared to the tremor signal, the frequency band regarded as the main signal is the main signal removal unit 46. Is removed. Thereby, only the tremor signal (including the main signal when the user performs a periodic operation) is extracted from the input signal IN. Then, the input signal IN from which the frequency band has been removed is input to the provisional frequency calculation unit 48. Then, the provisional frequency calculation unit 48 updates the coefficient α so that the output e (i) of the notch filter 17 is minimized, and calculates the provisional frequency f _e (i) at time i (see step S4 and Expression 12). ).

The provisional frequency f _e (i) calculated by the provisional frequency calculation unit 48 is input to the determination unit 50 (step S5), and the determination unit 50 calculates the tremor frequency (estimated frequency (f _e (i))). determining whether the estimated frequency f _t (i). in other words, the determination unit 50 is larger than the displacement amount of monitoring h (i) a first reference value h _r, and the variance of the provisional frequency f _e (i) v (i) is greater than the second reference value v _r as it is due to (Yes in step S6), and the provisional frequency f _e (i) the user of the cyclic operation (main signal), immediately before the estimated frequency f _t (i-1) determining the estimated frequency f _t (i) at that time (step S7). on the other hand, if the deviation monitor the amount h (i) is smaller than the first reference value h _r, Alternatively, the dispersion v (i) is the is smaller than the second reference value v _r (No in step S6), and the provisional frequency f _e (i) is due to the tremor signal As the determination unit 50 determines provisional frequency f _e in the time the (i) as the estimated frequency f _t (i) (Yes in step S8).

Next, the tremor removal unit 44 determines n satisfying the _equation (15) based on the estimated frequency f _t (i) finally determined by the determination unit 50 (step S9). Based on the determined n, the center frequency f _c (n) of the cutoff region in the tremor removal means 44 is determined (step S10), and the cutoff filter FI _{n of the} tremor removal means 44 is determined (step S11). ). The tremor removing means 44 separates the tremor signal from the input signal IN by the determined cutoff filter FI _n and generates an output signal OUT consisting only of the user's main signal (step S12). The output signal OUT is input to the output means 20 (step S13), and is output from the output means 20 to the first to sixth motors (step S14). As a result, the manipulator 12 (spoon 14) operates in a motion desired by the user from which the influence of tremor has been removed.

As described above, the meal support apparatus 10 according to the embodiment determines the estimated frequency f _t (i) corresponding to the frequency at each time of the tremor signal that the time change system 18 changes over time. It is possible to operate the manipulator 12 as the user intends by removing tremors with individual differences or time-variant tremors. Further, since the main signal removing unit 46 removes the frequency band regarded as the main signal from the input signal IN, it is possible to prevent the main signal from being removed as a tremor signal. Further, if the user has intentionally periodic operation, the determination unit 50 determines estimated determined immediately before the frequency f _t of the (i-1) as the estimated frequency f _t (i) in the time Therefore, it is possible to prevent the main signal related to the periodic operation from being removed.

(Experiment 1)
Next, in order to confirm the effectiveness of the noise removing device, the noise removing method, and the motion correcting device according to the present invention, an experiment using a simulation signal was performed. The results are shown in FIG. 7 and FIG. FIGS. 7A to 7C show an input signal and an output signal to the tremor frequency estimation means 42, and FIG. 8 shows changes in the estimated frequency f _t (i) in each method. Input signal IN is a sine wave with tremor signal (frequency Reference. The following f _r in FIG. 8) in is obtained by adding a sine wave and white noise of different frequencies as a disturbance. (a) is a case where only the estimation by the provisional frequency calculation unit 48 is performed, and (b) is a case where the main signal removal unit 46 having a cutoff frequency of 1.0 [Hz] is added before the input to the provisional frequency calculation unit 48. , (C) is a case where the determination unit 50 is added to (b).

First, the input at T ₁ is only the sine wave and low frequency of the tremor signal. The method (a) converges to a low frequency with a large power, but the methods (b) and (c) converge to _fr , so that the main signal removal unit 46 operates effectively. It can be said that It was then added disturbance frequencies close to f _r at T _2. In this way the (b) method, estimated over the T ₂ through T ₃ frequencies f _t (i) is come off from f _r, the determination unit 50 work effectively in the process of the (c), blocking filter FI _n switching has stopped. At T _4, and varying the f _r little by little. At this time, the estimated frequency f _t (i) follows f _r . That is, when the change of f _r is not steep, there is a high possibility that a minute variation of the tremor frequency, the estimated frequency f _t (i) continues to update in order to follow the f _r.

  Next, the effectiveness was verified by a signal from the force sensor 13 actually used. FIG. 9 is an enlarged view of 15 to 25 [sec] in which the tremor frequency varies particularly greatly. From FIG. 9, it can be seen that even if the tremor frequency changes with time, the tremor removal apparatus and method according to the present invention removes the tremor signal following the fluctuation of tremor.

(Experimental example 2)
Next, as Experiment 2, a verification experiment was performed using a manipulator having a force sensor 13 attached to the tip. In the embodiment, the six-degree-of-freedom manipulator 12 has been described as an example, but in Experimental Example 2, an experiment was performed using a three-degree-of-freedom manipulator. In order to operate manually, the input signal of the force sensor 13 was converted using the following equation.
Here, v _max is the maximum speed of the hand in the eating motion of a healthy person measured by motion capture, and F _max is the maximum input from the force sensor 13. This was calculated for each axis and used as a speed reference v _ref . F _max was 20 [N], and v _max was 80 [mm / sec]. The results in the Z direction are shown in FIG. The upper graph shows the sensor input, and the lower graph shows the result of capturing and tracking the tip position of the spoon 14 attached to the manipulator by motion capture. The filter is switched from OFF to ON from around 17 seconds.

  This result shows that most of the tremor signal has been removed. The operational feeling slightly changed by turning on the filter, but this is considered to be caused by an experiment conducted at an operating frequency of 100 [Hz] for the convenience of the manipulator. The reason why the input is changed before and after the filter is turned on and off is that the manipulator 12 stops moving according to the tremor.

  In addition, although the meal assistance apparatus 10 which removes the tremor as a noise signal from the input signal IN was demonstrated as an operation | movement correction apparatus in the Example, it is not limited to a tremor as a noise signal, For example, a user's It also includes noise due to hand tremors caused by tension and fatigue. Moreover, although the example demonstrated the case of the meal assistance apparatus 10 aiming at meal assistance, as an operation | movement assistance apparatus which concerns on this invention, it is not necessarily limited to meal assistance, For example, a joystick, a tactile sensor, The present invention can also be applied as appropriate to mouse operations on computers. In this case, a joystick or a mouse corresponds to the operated means. Furthermore, the present invention can also be applied to industrial robots and devices that support medical practices such as surgery.

  Further, in the embodiment, the case where the operated means and the motion correction target are the same manipulator 12 is illustrated, but the operated means to which the motion is applied and the motion correction target may be different. For example, it is considered that the user operates the operated means (joystick, etc.) to remove the noise signal and then operate while correcting a member (for example, a writing instrument, a surgical instrument, etc.) away from the user. It is done. Furthermore, after the operation of the operated means by the user is corrected, for example, an operation correction target (such as a mouse cursor) displayed on the screen on the computer may be activated. Further, the operating subject that operates the operated means is not limited to a human as in the embodiment. For example, a robot may operate the operated means to reproduce the operation as a motion correction target. .

In the embodiment, the provisional frequency calculation unit 48 calculates the provisional frequency f _e (i) and the determination unit 50 determines the estimated frequency f _t (i), but the notch frequency obtained from the tap coefficient α is The estimated frequency f _t (i) may be used as it is. That is, the determination unit 50 is not necessarily provided. Further, the main signal removal unit 46 is not limited to the high pass filter as in the embodiment, and a low pass filter or the like may be employed. Furthermore, although the band stop filter is employed as the cutoff filter FI _n in the embodiment, a notch filter or another filter may be employed.

It is a perspective view which shows the whole structure of the meal assistance apparatus which concerns on an Example. It is a block diagram which shows the structure of the tremor removal system provided in the meal assistance apparatus which concerns on an Example. It is a graph which shows an example of the interruption | blocking filter used with a tremor removal means. It is a figure which shows the notch filter employ | adopted as the provisional frequency calculation part. It is a block diagram which shows the determination flow in a determination part. It is a flowchart figure which shows the tremor removal method of a meal assistance apparatus. It is a graph which shows the experimental result of the input signal and output signal of a tremor frequency estimation means, Comprising: When (a) provides only a provisional frequency calculation part, (b) uses a provisional frequency estimation part and a main signal removal part. (C) shows the result when the determination unit is added to (b). It is a graph which shows the change of the estimated frequency estimated in the method of (a)-(c) in FIG. It is an experimental result showing the time change of an input signal and an output signal. It is a graph which shows the experimental result of Experimental example 2, Comprising: The upper graph shows the input signal to a force sensor, and the lower graph shows the result of having image | photographed and tracked the tip position of the spoon by motion capture.

Explanation of symbols

12 Manipulator (Operated means, motion correction target)
14 Spoon (operated means, motion correction target), 13 Force sensor (motion detection means)
15 input means, 17 notch filter, 20 output means 42 tremor frequency estimation means (noise frequency estimation means)
44 tremor removal means (noise removal means), 46 main signal removal section 48 provisional frequency calculation means, 50 determination section, IN input signal APF all-pass filter, α tap coefficient, h (i) deviation monitoring amount v (i) variance , e (i) output (error), f _t (i) estimating the frequency, f _e (i) Preliminary frequency FI _n cutoff filter, OUT an output signal, h _r the first reference value, v _r second reference value

Claims (9)

An input means (15) that is detected by the motion detector (13) and receives an input signal (IN) including a main signal and a noise signal; and
Time-varying noise by updating the tap coefficient (α) that minimizes the output (e (i)) of the input signal (IN) input to the notch filter (17) to which the all-pass filter (APF) is applied. Noise frequency estimation means (42) for determining an estimated frequency (f _t (i)) corresponding to the frequency at each time of the signal;
The cutoff signal (FI _n ) determined based on the estimated frequency (f _t (i)) determined by the noise frequency estimation means (42) removes the noise signal from the input signal (IN) and outputs an output signal ( OUT) to generate noise removal means (44),
An output means (20) for outputting the output signal (OUT) generated by the noise removal means (44).   The noise frequency estimating means (42) includes a main signal removing unit (46) for removing a frequency band regarded as the main signal from the input signal (IN), and the main signal removing unit (46) has a frequency band. The noise signal removing apparatus according to claim 1, wherein the removed input signal (IN) is inputted to the notch filter (17). The noise frequency estimation means (42) includes a provisional frequency calculation means (48) for calculating a provisional frequency (f _e (i)) that is a notch frequency from the tap coefficient (α), and the provisional frequency calculation means (48). A determination unit (50) for determining the estimated frequency (f _t (i)) based on the provisional frequency (f _e (i)) calculated in
The determination unit (50)
The deviation monitoring amount (h (i)) based on the difference between the provisional frequency (f _e (i)) at an arbitrary time and the estimated frequency (f _t (i-1)) determined immediately before the arbitrary time is set in advance. A second reference value that is greater than the set first reference value (h _r ) and has a preset distribution (v (i)) of the provisional frequencies (f _e (ik) to f _e (i)) within a predetermined time when (v _r) greater than, and determines an estimated frequency immediately before is determined to the arbitrary time _{(f t (i-1)} ) as the estimated frequency (f _t (i)) at any time,
When the deviation monitoring amount (h (i)) is smaller than the first reference value (h _r ), or when the variance (v (i)) is smaller than the second reference value (v _r ), The noise signal removal apparatus according to claim 1 or 2, wherein the provisional frequency (f _e (i)) calculated at the arbitrary time is determined as an estimated frequency (f _t (i)) at the arbitrary time. The input signal (IN) including the main signal and noise signal detected by the motion detector (13) is input to the input means (15),
The input signal (IN) input to the input means (15) is input to a notch filter (17) to which an all-pass filter (APF) is applied, and the tap coefficient (α) that minimizes its output (e (i)) ) To determine the estimated frequency (f _t (i)) corresponding to the frequency at each time of the time-varying noise signal,
By a cutoff filter (FI _n ) determined based on the estimated frequency (f _t (i)), a noise signal is removed from the input signal (IN) to generate an output signal (OUT),
An output means (20) outputs the generated output signal (IN).   The noise signal removal method according to claim 4, wherein an input signal (IN) from which a frequency band regarded as the main signal is removed is input to the notch filter (17). Calculate the provisional frequency (f _e (i)) which is a notch frequency from the tap coefficient (α),
The deviation monitoring amount (h (i)) based on the difference between the provisional frequency (f _e (i)) at an arbitrary time and the estimated frequency (f _t (i-1)) determined immediately before the arbitrary time is set in advance. A second reference value that is greater than the set first reference value (h _r ) and has a preset distribution (v (i)) of the provisional frequencies (f _e (ik) to f _e (i)) within a predetermined time when (v _r) greater than, and determines an estimated frequency immediately before is determined to the arbitrary time _{(f t (i-1)} ) as the estimated frequency (f _t (i)) at any time,
When the deviation monitoring amount (h (i)) is smaller than the first reference value (h _r ), or when the variance (v (i)) is smaller than the second reference value (v _r ), The noise signal removal method according to claim 4 or 5, wherein the provisional frequency (f _e (i)) calculated at the arbitrary time is determined as an estimated frequency (f _t (i)) at the arbitrary time. An operation detector (13) for outputting an input signal (IN) including a main signal and a noise signal according to the operation on the operated means (12, 14);
Input means (15) for receiving the input signal (IN) output from the motion detection unit (13),
Time-varying noise by updating the tap coefficient (α) that minimizes the output (e (i)) of the input signal (IN) input to the notch filter (17) to which the all-pass filter (APF) is applied. Noise frequency estimation means (42) for determining an estimated frequency (f _t (i)) corresponding to the frequency at each time of the signal;
The cutoff signal (FI _n ) determined based on the estimated frequency (f _t (i)) determined by the noise frequency estimation means (42) removes the noise signal from the input signal (IN) and outputs an output signal ( OUT) to generate noise removal means (44),
Output means (20) for outputting the output signal (OUT) generated by the noise removing means (44);
An operation correction apparatus comprising: a drive unit that operates the operation correction target (12, 14) based on the output signal (OUT) output from the output unit (20).   The noise frequency estimating means (42) includes a main signal removing unit (46) for removing a frequency band regarded as the main signal from the input signal (IN), and the main signal removing unit (46) has a frequency band. The motion correction device according to claim 7, wherein the removed input signal (IN) is input to the notch filter (17). The noise frequency estimation means (42) includes a provisional frequency calculation means (48) for calculating a provisional frequency (f _e (i)) that is a notch frequency from the tap coefficient (α), and the provisional frequency calculation means (48). A determination unit (50) for determining the estimated frequency (f _t (i)) based on the provisional frequency (f _e (i)) calculated in
The determination unit (50)
The deviation monitoring amount (h (i)) based on the difference between the provisional frequency (f _e (i)) at an arbitrary time and the estimated frequency (f _t (i-1)) determined immediately before the arbitrary time is set in advance. A second reference value that is greater than the set first reference value (h _r ) and has a preset distribution (v (i)) of the provisional frequencies (f _e (ik) to f _e (i)) within a predetermined time when (v _r) greater than, and determines an estimated frequency immediately before is determined to the arbitrary time _{(f t (i-1)} ) as the estimated frequency (f _t (i)) at any time,
When the deviation monitoring amount (h (i)) is smaller than the first reference value (h _r ), or when the variance (v (i)) is smaller than the second reference value (v _r ), The motion correction apparatus according to claim 7 or 8, wherein the provisional frequency (f _e (i)) calculated at the arbitrary time is determined as an estimated frequency (f _t (i)) at the arbitrary time.