Measurement device and measurement method

Abstract

A measurement device (1) includes a first sensor (101), a second sensor (201), a first velocity calculation unit (100), a second velocity calculation unit (200), an observed velocity difference calculation unit (300), an a posteriori estimated velocity difference distribution calculation unit (500), an a priori predicted velocity difference distribution prediction unit (400), an estimated velocity difference calculation unit (600), and a moving body velocity calculation unit (700). The first velocity calculation unit (100) calculates a first velocity of a moving body using a detection value detected by the first sensor (101). The second velocity calculation unit (200) calculates a second velocity of the moving body using a detection value detected by the second sensor (201). The observed velocity difference calculation unit (300) calculates an observed velocity difference on the basis of the first velocity and the second velocity. The a posteriori estimated velocity difference distribution calculation unit (500) calculates an a posteriori estimated velocity difference distribution, which is a distribution of velocity differences. The estimated velocity difference calculation unit (600) calculates an estimated velocity difference. The moving body velocity calculation unit (700) calculates the velocity of the moving body on the basis of the second velocity and the estimated velocity difference.

Description

Measuring device and measuring method

[0001] The present disclosure relates to a measurement device and a measurement method.

[0002] 2. Description of the Related Art A technique is known for measuring the speed of a moving body such as a railway vehicle, using an acceleration detection means for detecting the acceleration of the moving body (see, for example, Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2003-4758

[0004] However, with conventional technology, measurement errors can be large depending on the state of the moving object, making it difficult to measure the speed of the moving object with high accuracy.

[0005] An object of the present disclosure is to solve the above-mentioned problems and to measure the speed of a moving object with high accuracy.

[0006] a first sensor that detects the position, velocity, acceleration, or jerk that is the rate of change of acceleration of a moving object; a second sensor that detects the position, velocity, acceleration, or jerk that is the rate of change of acceleration of the moving object; a first velocity calculation unit that calculates a first velocity of the moving object using a detection value detected by the first sensor; a second velocity calculation unit that calculates a second velocity of the moving object using a detection value detected by the second sensor; an observed velocity difference calculation unit that calculates an observed velocity difference by subtracting the first velocity from the second velocity; a posterior estimated velocity difference distribution calculation unit that calculates a posterior estimated velocity difference distribution that is a distribution of the velocity differences based on a prior predicted velocity difference distribution that is a prediction of the distribution of the velocity differences between the first velocity 103 and the second velocity 203 and the observed velocity difference; a prior predicted velocity difference distribution prediction unit that calculates the prior predicted velocity difference distribution for the next time point based on the posterior estimated velocity difference distribution; and an estimated velocity difference calculation unit that calculates a representative value of the posterior estimated velocity difference distribution as an estimated velocity difference that is an estimate of the velocity difference. and a moving body velocity calculation unit that calculates the velocity of the moving body by subtracting the estimated velocity difference from the second velocity. A measurement method according to one aspect of the present disclosure includes: calculating a first velocity of the moving body using a first sensor; calculating a second velocity of the moving body using a second sensor; calculating an observed velocity difference by subtracting the first velocity from the second velocity; calculating a posterior estimated velocity difference distribution that is a distribution of the velocity differences based on a priori predicted velocity difference distribution that is a prediction of a distribution of the velocity differences between the first velocity 103 and the second velocity 203 and the observed velocity difference; calculating the priori predicted velocity difference distribution for the next time point based on the posterior estimated velocity difference distribution; calculating a representative value of the posterior estimated velocity difference distribution as an estimated velocity difference that is an estimate of the velocity difference; and calculating the velocity of the moving body by subtracting the estimated velocity difference from the second velocity.

[0007] According to the present disclosure, the speed of a moving object can be measured with high accuracy.

[0008] 10 is a block diagram showing a configuration of a measurement device according to a first embodiment of the present disclosure. FIG. 11 is a diagram showing an example of a moving body. FIG. 12 is a diagram showing another example of a moving body. FIG. 13 is a flowchart showing an example of an operation of the measurement device. FIG. 14 is a graph showing an example of a speed calculated by the measurement device according to the first embodiment. FIG. 15 is a block diagram showing a configuration of a measurement device as another example of the measurement device 1. FIG. 16 is a block diagram showing a configuration of a measurement device according to a second embodiment of the present disclosure. FIG. 17 is a graph showing an example of a speed calculated by the measurement device according to the second embodiment. FIG. 18 is a diagram showing an example of a hardware configuration of a measurement device other than a first sensor and a second sensor. FIG. 19 is a diagram showing another example of the hardware configuration shown in FIG.

[0009] In order to explain the present disclosure in more detail, embodiments for implementing the present disclosure will be described below with reference to the accompanying drawings. <Embodiment 1> The configuration of a measurement device 1 according to embodiment 1 will be described. FIG. 1 is a block diagram showing the configuration of the measurement device 1 according to embodiment 1 of the present disclosure. The measurement device 1 includes a first sensor 101, a first velocity calculation unit 100, a second sensor 201, a second velocity calculation unit 200, an observed velocity difference calculation unit 300, a priori predicted velocity difference distribution prediction unit 400, a posterior estimated velocity difference distribution calculation unit 500, an estimated velocity difference calculation unit 600, and a moving object velocity calculation unit 700.

[0010] The first sensor 101 is a sensor that detects the position, velocity, acceleration, or jerk, which is the rate of change of acceleration, of a moving object as a detection value 102 of the first sensor 101. Other quantities related to position may also be detected. The moving object is, for example, a train or an elevator. The first sensor 101 is, for example, a rotation speed detection sensor made up of rollers that rotate in conjunction with the movement of the moving object. The detection value 102 of the first sensor is input to the first velocity calculation unit 100.

[0011] The first velocity calculation unit 100 is a device that calculates a first velocity 103, which is the velocity of the moving object, using a detection value 102 detected by a first sensor. The first velocity 103 is input to the observed velocity difference calculation unit 300.

[0012] The second sensor 201 is a sensor that detects the position, velocity, acceleration, or jerk, which is the rate of change of acceleration, of the moving object as a detection value 202 of the second sensor 201. The second sensor 201 may also detect other quantities as long as they are information related to the position. The second sensor 201 is, for example, an acceleration sensor that detects the acceleration in the moving direction of the moving object. The detection value 202 of the second sensor 201 is input to the second velocity calculation unit 200.

[0013] The second velocity calculation unit 200 is a device that calculates a second velocity 203, which is the velocity of the moving object, using a detection value 202 detected by a second sensor 201. The second velocity 203 is input to the observed velocity difference calculation unit 300 and the moving object velocity calculation unit 700.

[0014] Here, the error in the first velocity 103 and the error in the second velocity 203 are of different types. The first velocity 103 has a large variation error but a small offset error. In other words, the error in the high frequency band is large but the error in the low frequency band is small. Also, it is assumed that a large error can occur instantaneously. For example, this applies to a case where a receiver that detects position using GPS is used as the first sensor 101. The position detected by GPS has a large variation error but a small offset error.

[0015] Furthermore, since the speed is calculated by differentiating the position with respect to time, errors generally become large in the high frequency band. This also applies to a case where a rotation speed detection sensor that detects the number of rotations of the wheels is used as the first sensor 101 in a mobile object that moves on wheels. The number of rotations detected by the rotation speed detection sensor corresponds to the travel distance. Therefore, since the speed is calculated by differentiating the travel distance with respect to time, errors generally become large in the high frequency band. Furthermore, since wheels can generally spin or slide, large errors can occur momentarily. However, the offset error is small.

[0016] On the other hand, the second velocity 203 has a large offset error but a small variation error. In other words, the error in the low frequency band is large but the error in the high frequency band is small. For example, this applies to the case where an acceleration sensor is used as the second sensor 201. Since the velocity is calculated by time-integrating the acceleration detected by the acceleration sensor, errors generally accumulate while variation errors are suppressed. In other words, the error in the low frequency band is large but the error in the high frequency band is small.

[0017] Fig. 2 is a diagram showing an example of a moving body. In the example shown in Fig. 2, the moving body is a train. In the example shown in Fig. 2, a first sensor 101 and a second sensor 201 are mounted on the train. In the example shown in Fig. 2, the first sensor 101 is a rotation speed detection sensor, and the second sensor 201 is an acceleration sensor.

[0018] Fig. 3 is a diagram showing another example of a moving body. The example shown in Fig. 3 is an elevator. In the example shown in Fig. 3, a first sensor 101 and a second sensor 201 are mounted on the elevator (specifically, the elevator car). In the example shown in Fig. 3, the first sensor 101 is a rotation speed detection sensor, and the second sensor 201 is an acceleration sensor.

[0019] The observed velocity difference calculation unit 300 is a device that calculates the observed velocity difference 301 using the first velocity 103 and the second velocity 203 .

[0020] The posterior estimated speed difference distribution calculation unit 500 is a device that calculates a posterior estimated speed difference distribution 501, which is the distribution of the speed difference between the first speed 103 and the second speed 203, using the observed speed difference 301 and a prior predicted speed difference distribution 401, which is a prediction of the distribution of the speed difference between the first speed 103 and the second speed 203. The posterior estimated speed difference distribution 501 is input to the prior predicted speed difference distribution prediction unit 400 and the estimated speed difference calculation unit 600.

[0021] The prior predicted speed difference distribution prediction unit 400 is a device that calculates a prior predicted speed difference distribution 401 for the next time based on a posterior estimated speed difference distribution 501 .

[0022] The estimated speed difference calculation unit 600 is a device that calculates an estimated speed difference 601, which is an estimate of the speed difference between the first speed 103 and the second speed 203, based on the posterior estimated speed difference distribution 501. The estimated speed difference 601 is input to the moving object speed calculation unit 700.

[0023] The moving object speed calculation unit 700 is a device that calculates the speed of the moving object by subtracting the estimated speed difference 601 from the second speed 203 and outputs the calculated speed to the outside.

[0024] The first speed calculation unit 100, the second speed calculation unit 200, the observed speed difference calculation unit 300, the posterior estimated speed difference distribution calculation unit 500, the preliminary predicted speed difference distribution prediction unit 400, the estimated speed difference calculation unit 600, and the moving object speed calculation unit 700 are configured by a processor or the like that executes a program stored in a memory. The processor is a processing circuit such as a CPU or a system LSI. Alternatively, the above functions may be realized by a combination of multiple processors and multiple memories.

[0025] The operation of the measurement device 1 according to the first embodiment will be described. FIG. 4 is a flowchart showing an example of the operation of the measurement device 1. The first sensor 101 detects the position, velocity, acceleration, or jerk of a moving object. The first velocity calculation unit 100 calculates a first velocity 103, which is the velocity of the moving object, using the detection value 102 detected by the first sensor 101 (step S1). The second sensor 201 detects the position, velocity, acceleration, or jerk of the moving object. The second velocity calculation unit 200 calculates a second velocity 203, which is the velocity of the moving object, using the detection value 202 detected by the second sensor 201 (step S2). The observed velocity difference calculation unit 300 calculates an observed velocity difference 301 by subtracting the first velocity 103 from the second velocity 203 (step S3).

[0026] The posterior estimated speed difference distribution calculation unit 500 calculates the posterior estimated speed difference distribution 501 based on the prior predicted speed difference distribution 401 and the observed speed difference 301 (step S4). If a discrepancy occurs between the observed speed difference 301 and the prior predicted speed difference distribution 401, the posterior estimated speed difference distribution 501 is calculated while disregarding the calculation result of the observed speed difference 301. If a high-frequency band error or an instantaneous error due to a skid or slide occurs in the first speed 103, this will cause a similar error in the observed speed difference 301. In this case, a discrepancy occurs between the observed speed difference 301 and the prior predicted speed difference distribution 401, so the posterior estimated speed difference distribution calculation unit 500 calculates the posterior estimated speed difference distribution 501 while prior predicted speed difference distribution 401 is given more importance, thereby suppressing the error.

[0027] The prior prediction speed difference distribution prediction unit 400 inputs and stores the posterior estimated speed difference distribution 501, and when calculating the speed of the moving body at the next time, calculates the prior prediction speed difference distribution 401 based on a predetermined time evolution update rule (step S5).

[0028] As a specific example of a method for calculating the posterior estimated speed difference distribution 501, the posterior estimated speed difference distribution calculation unit 500 may calculate a weighted average using a predetermined weight for the averages of the observed speed difference 301 and the prior predicted speed difference distribution 401, and calculate the weighted average as the average of the posterior estimated speed difference distribution 501. In this case, the average of the posterior estimated speed difference distribution 501 is calculated as a weighted average of the average of the prior predicted speed difference distribution 401 and the observed speed difference 301.

[0029] The average of the prior predicted speed difference distribution 401 may be calculated as the average of the posterior estimated speed difference distribution 501 at the previous time. Alternatively, the variance of the prior predicted speed difference distribution 401 may be updated according to a predetermined update rule, and the variance of the posterior estimated speed difference distribution 501 may be calculated.

[0030] The posterior estimated speed difference distribution 501 is output based on the mean and variance calculated by the above procedure. The operation of calculating the mean of the prior predicted speed difference distribution 401, which corresponds to the output at the previous time, and the weighted mean of the observed speed difference 301, which corresponds to the input at the current time, is nothing but a low-pass filter using an IIR digital filter. Therefore, the above operation removes high-frequency band errors that occur in the observed speed difference 301. In addition, instantaneous errors due to skidding or sliding are also removed because they contain a large amount of high-frequency band errors.

[0031] The estimated speed difference calculation unit 600 calculates a representative value of the posterior estimated speed difference distribution 501 as an estimated speed difference 601, which is an estimated value of the speed difference between the first speed 103 and the second speed 203 (step S6). The representative value is an average value, a median value, a mode value, or a value calculated by another predetermined method. The estimated speed difference 601 is calculated, for example, by a Kalman filter using the observed speed difference 301 as an observed value.

[0032] The moving object speed calculation unit 700 calculates the speed of the moving object by subtracting the estimated speed difference 601 from the second speed 203 (step S7). The posterior estimated speed difference distribution 501 removes high-frequency band variation errors and instantaneous errors caused by skidding or sliding, but does not remove low-frequency band errors. Therefore, although low-frequency band errors occur in the second speed 203, the low-frequency band errors are removed by subtracting a representative value of the posterior estimated speed difference distribution 501, which has passed through only the low-frequency band errors.

[0033] The estimated speed difference 601, the posterior estimated speed difference distribution 501, and the anterior predicted speed difference distribution 401 may be calculated by, for example, a Kalman filter. In this case, the Kalman gain of the Kalman filter can be calculated by dividing the anterior error covariance matrix of the estimated speed difference 601 by the sum of the anterior error covariance matrix and the variance of the observed speed difference 301.

[0034] Here, an operation to which a Kalman filter is applied will be described as a specific example of the operation of the measurement device 1. First, a Kalman filter is derived within the framework of Bayesian estimation. The state at time t is expressed as an n-dimensional vector X t , the observations are expressed as an m-dimensional vector y tAs such, consider the linear discrete-time state space model shown in the following equation.

[0035] where H is an observation matrix with m rows and n columns. X,t is the system noise of an n-dimensional vector that follows a Gaussian distribution with mean 0 and variance Q. ｙ,ｔ is the observation noise of an m-dimensional vector that follows a Gaussian distribution with mean 0 and variance R. t The distribution of p(X t ), X t When y is given t The conditional distribution p(y t │X t ), y t The distribution of p(y t ) is expressed as follows:

[0036] Here, x ｔ t is X t Expected value of Σ ｔ is X t By Bayes' theorem, y t When X is obtained t The conditional distribution p(X t │y t ) is expressed as follows:

[0037] p(y t │X t ) p(X t The exponent part ψ of ( ) can be transformed as follows:

[0038] Here, we introduce Π shown in the following equation.

[0039] In this case, −2ψ can be transformed into the following equation:

[0040] Therefore, p(X t │y t ) is expressed as follows:

[0041] Here, K shown in the following formula ｔ Introduce.

[0042] The K introduced above ｔ Using the above, the Kalman filter is written as follows: 1) Initialization

[0043] 2) Observation update step

[0044] 3) Time update step

[0045] 4) Set t←t+1 and return to 2).

[0046] The following values ​​are the X values ​​calculated by the Kalman filter: ｔ , Σ t is an estimate of

[0047] Here, the first velocity 103 and the second velocity 203 are respectively expressed as v １，ｔ , v ２，ｔ Let's say. １，ｔ The error in the high frequency band is large, while the error in the low frequency band is small. ２，ｔ In this case, the error in the low frequency band is large, while the error in the high frequency band is small. ｔ is the difference between the two v ２，ｔ -v １，ｔ is the observable, X t is the difference between the two v ２，ｔ -v １，ｔ As the true state of t Consider estimating

[0048] In equations (1.1) and (1.2), the error of the low frequency component is the system noise w Ｘ，ｔ The error of the high frequency component is considered to be the observation noise w ｙ，ｔ It is considered that X ｔ , v １，ｔ , v ２，ｔ Among the errors in １，ｔ The error of v is removed, and the error of the low frequency component is ２，ｔ Therefore, the velocity of the moving object (Equation 1.23 described later) can be estimated using a Kalman filter as follows:

[0049] 1) Initialization

[0050] 2) Observation update step

[0051] 3) Moving object speed estimation step

[0052] 4) Time update step

[0053] 4) Set t←t+1 and return to 2).

[0054] The observed velocity difference 301 is calculated by y ｔ Corresponds to.

[0055] The mean of the posterior estimated speed difference distribution 501 corresponds to the following value in equation (1.21):

[0056] The variance of the posterior estimated speed difference distribution 501 corresponds to the following value in equation (1.22):

[0057] The mean of the predicted speed difference distribution 401 corresponds to the following value in equation (1.24):

[0058] The variance of the predicted speed difference distribution 401 corresponds to the following value in equation (1.25):

[0059] The estimated velocity difference corresponds to the following value in equation (1.21):

[0060] The velocity of the moving body corresponds to the following value in equation (1.23):

[0061] The velocity of the moving object is calculated by the following formula: ２，t This corresponds to the operation of removing low-frequency band errors from the calculated velocity. Therefore, the velocity of the moving object can be calculated with high accuracy.

[0062] Consider a case where a rotation detection sensor and an acceleration sensor are applied to the first sensor 101 and the second sensor 201, respectively. １，ｔ , v ２，ｔ are calculated by the rotation detection sensor and the acceleration sensor, respectively, and are calculated by the following equations:

[0063] Here, N, ΔL, a ｔ , Δt are the number of rotations detected by the rotation detection sensor, the moving distance per rotation, the acceleration detected by the acceleration sensor, and the measurement period of both sensors, respectively.

[0064] 5 is a graph showing an example of the velocity calculated by the measurement device 1 according to the first embodiment. The velocity of the moving object increases at a constant pace from a stationary state until 250 seconds, then moves at a constant speed of 25 m / s from 250 seconds to 750 seconds, decreases at a constant pace from 750 seconds to 1000 seconds, and finally comes to a standstill. The velocity v of the first sensor 101 is １，ｔ A variation error, which is a high frequency band error, occurs in the velocity v of the second sensor 201. Also, an instantaneous error occurs at the time point of 500 seconds. ２，ｔ It can be seen that an offset error, which is an error in the low frequency band, occurs in v and increases over time. ２，ｔ The offset error is removed from the velocity v of the first sensor 101, and the velocity is estimated with high accuracy. １，ｔ , it can be seen that the variation error is suppressed and the instantaneous error at the time point of 500 seconds is also reduced.

[0065] The advantages of the measurement device 1 according to the first embodiment will be described. The posterior estimated speed difference distribution calculation unit 500 removes high-frequency band errors that occur in the observed speed difference 301 and outputs a posterior estimated speed difference distribution 501 that allows low-frequency band errors to pass through. The moving object speed calculation unit 700 then calculates the speed of the moving object by subtracting the estimated speed difference 601 from the second speed 203. Although low-frequency band errors occur in the second speed 203, the low-frequency band errors are removed by subtracting a representative value of the posterior estimated speed difference distribution 501 that allows only low-frequency band errors to pass through. Therefore, the measurement device 1 can measure the speed of the moving object with high accuracy.

[0066] The first sensor 101 and the second sensor 201 are not limited to a combination of a rotation speed detection sensor and an acceleration sensor, and various other sensor combinations are possible. For example, the first sensor 101 and the second sensor 201 may be a combination of a receiver that detects position using GPS and an acceleration sensor, and applied to a train. The first sensor 101 and the second sensor 201 may also be an absolute position sensor that detects absolute position by placing a code tape indicating position information on the path of the moving object, such as a linear encoder, and placing a code tape reader on the moving object, and a method of detecting height (position) from air pressure using an air pressure sensor, and applied to an elevator car. Other sensor combinations may also be applied to other moving objects.

[0067] If the detected values ​​102 and 202 are positions, the velocity is calculated by dividing the distance traveled in a predetermined time by the predetermined time, as described above. If the detected values ​​102 and 202 are velocities, the velocity is calculated. If the detected values ​​102 and 202 are accelerations, the velocity is calculated by integrating the value obtained by multiplying each detected value from a predetermined time by the measurement period. If the detected values ​​102 and 202 are jerk values, the acceleration is calculated by integrating the value obtained by multiplying each detected value from a predetermined time by the measurement period, and the velocity is calculated from the acceleration.

[0068] 6 is a block diagram showing the configuration of measurement device 2, another example of measurement device 1. As shown in FIG. 6, measurement device 2 may further include a moving object position calculation unit 800 in addition to the configuration of measurement device 1. The moving object position calculation unit 800 calculates the moving distance of the moving object from a predetermined point based on the velocity of the moving object calculated by the moving object velocity calculation unit 700. For example, the moving object position calculation unit 800 can calculate the position of the moving object by integrating the value obtained by multiplying the velocity of each moving object from a predetermined point by the measurement period.

[0069] <Embodiment 2> The configuration of a measurement device 3 according to embodiment 2 will be described. Fig. 7 is a block diagram showing the configuration of the measurement device 3 according to embodiment 2 of the present disclosure. The measurement device 3 includes a first sensor 101, a first velocity calculation unit 100, a second sensor 201, a second velocity calculation unit 200, an observed velocity difference calculation unit 300, an outlier probability calculation unit 900, a priori predicted velocity difference distribution prediction unit 400, a posterior estimated velocity difference distribution calculation unit 510, an estimated velocity difference calculation unit 600, and a moving object velocity calculation unit 700.

[0070] The first sensor 101, the first speed calculation unit 100, the second sensor 201, the second speed calculation unit 200, the observed speed difference calculation unit 300, the advance predicted speed difference distribution prediction unit 400, the estimated speed difference calculation unit 600, and the moving body speed calculation unit 700 are the same as the corresponding components of the measuring device 1.

[0071] The observed speed difference 301 is input to the outlier probability calculation unit 900 and the posterior estimation speed difference distribution calculation unit 510. The prior prediction speed difference distribution 401 is input to the posterior estimation speed difference distribution calculation unit 510 and the outlier probability calculation unit 900.

[0072] The outlier probability calculation unit 900 is a device that calculates an outlier probability 901, which is the probability that the observed speed difference 301 will be an outlier, based on the prior predicted speed difference distribution 401 and the observed speed difference 301. The outlier probability 901 is input to the posterior estimated speed difference distribution calculation unit 510.

[0073] The posterior estimated speed difference distribution calculation unit 510 is a device that calculates a posterior estimated speed difference distribution 511 based on the observed speed difference 301 , the predicted speed difference distribution 401 , and the outlier probability 901 .

[0074] The outlier probability calculation unit 900 and the posterior estimated speed difference distribution calculation unit 510 are configured with a processor that executes a program stored in a memory. The processor is a processing circuit such as a CPU or a system LSI. Furthermore, multiple processors and multiple memories may work together to achieve the above functions.

[0075] The following describes the operation of the measurement device 3 according to embodiment 2. The operations of the first sensor 101, the first velocity calculation unit 100, the second sensor 201, the second velocity calculation unit 200, and the observed velocity difference calculation unit 300 are similar to the operations of the corresponding components of the measurement device 1.

[0076] The outlier probability calculation unit 900 calculates the outlier probability 901 based on the difference between the predicted speed difference distribution 401 and the observed speed difference 301. When the difference between the two is small, the outlier probability 901 is small, and when the difference between the two is large, the outlier probability 901 is large. For example, the outlier probability 901 can be a monotonically increasing function with the difference between the average of the predicted speed difference distribution 401 and the observed speed difference 301 as a variable.

[0077] The outlier probability 901 may be calculated by assuming that the distribution of the observed speed difference 301 is a mixed distribution of a distribution of normal values ​​and a distribution of outliers, and that the distribution of normal values ​​is a Gaussian distribution and the distribution of outliers is a uniform distribution.

[0078] Furthermore, based on the difference between the average of the predicted speed difference distribution 401 and the observed speed difference 301, an outlier probability that a large measurement error has occurred in the observed speed difference 301 due to the influence of a skid or slide of the moving body or the like may be calculated, and this may be set as the outlier probability 901. Furthermore, the calculation method for the outlier probability 901 may be any other method as long as it is designed as a measure for the purpose of calculating the possibility that a large error has occurred in the measurement error of the observed speed difference 301.

[0079] The operation of the advance speed difference distribution prediction unit 400 is similar to the operation of the corresponding component of the measurement device 1 .

[0080] The posterior estimated speed difference distribution calculation unit 500 calculates the posterior estimated speed difference distribution 501 based on the prior predicted speed difference distribution 401, the observed speed difference 301, and the outlier probability 901. As an example of a specific calculation method, the posterior estimated speed difference distribution calculation unit 510 may calculate a weighted average of the observed speed difference 301 and the average of the prior predicted speed difference distribution 401 using a predetermined weight set in consideration of the outlier probability 901, and calculate the weighted average as the average of the posterior estimated speed difference distribution 511.

[0081] When the outlier probability 901 is small, a weight is assigned that places importance on the observed speed difference 301. When the outlier probability 901 is large, the weight on the average of the prior predicted speed difference distribution 401 is increased and the weight on the observed speed difference 301 is decreased, thereby placing importance on the average of the prior predicted speed difference distribution 401 and calculating the average of the posterior estimated speed difference distribution 511. Furthermore, the variance of the prior predicted speed difference distribution 401 is updated according to a predetermined update rule, and the variance of the posterior estimated speed difference distribution 511 is calculated.

[0082] The a posteriori estimated velocity difference distribution 511 is output based on the mean and variance calculated by the above processing. As a result, like the measurement device 1, high-frequency band errors occurring in the observed velocity difference 301 are removed and low-frequency band errors are allowed to pass through.

[0083] Furthermore, by using a predetermined weight set in consideration of the outlier probability 901, the measurement device 3 calculates the average of the posterior estimated speed difference distribution 511 by placing particular emphasis on the average of the prior predicted speed difference distribution 401 while particularly disregarding the observed speed difference 301, since the outlier probability 901 becomes large when a large measurement error occurs in the observed speed difference 301 due to the influence of the moving body's skidding or sliding, etc. This eliminates the influence of variance errors in the high frequency band and instantaneous errors caused by skidding or sliding.

[0084] The estimated speed difference 601, the posterior estimated speed difference distribution 511, and the prior predicted speed difference distribution 401 are calculated by, for example, a Kalman filter. In this case, the Kalman gain of the Kalman filter can be calculated by multiplying the prior error covariance matrix of the estimated speed difference 601 by the probability that it will be a normal value, and dividing the result by the sum of the product of the prior error covariance matrix and the probability that it will be a normal value and the variance of the observed speed difference. The probability of it being a normal value can be calculated by subtracting the outlier probability from 1.

[0085] The operations of the estimated speed difference calculation unit 600 and the moving object speed calculation unit 700 are similar to the operations of the corresponding components of the measurement device 1. Here, as a specific example of the operation of the measurement device 3, an operation to which a Kalman filter is applied will be described.

[0086] In the measurement device 3, the Kalman filter explained in the operation of the measurement device 1 is applied, and a Kalman filter that takes outliers into account is considered. For this purpose, a Kalman filter that takes outliers into account is first derived. ｔ The probability that is an outlier is Z ｔ Then, equation (1.2) can be transformed into the following equation.

[0087] In formula (2.1), the arbitrary value is an arbitrary value. ｔ When y is given ｔ The conditional distribution p(y ｔ │X ｔ ) is expressed as follows:

[0088] In this case, p(y ｔ │X ｔ ), Jensen's inequality gives us

[0089] In this case, the difference between both sides of the inequality is expressed as follows:

[0090] Here, the last equation of equation (2.4) uses the Kullback-Leibler divergence KL(P|Q) for given distributions P and Q. Here, the following approximation is introduced:

[0091] In this case, if equations (2.5) and (2.6) are used, equation (2.4) becomes 0, and logp(y ｔ │X ｔ ) can be expressed as follows:

[0092] y ｔ When p(y ｔ │OUT,X ｔ ) = 1. In this case, logp(y ｔ │X ｔ ) can be transformed into the following equation:

[0093] Therefore, φ ｔ = p(in│y ｔ , Xｔ ) the following equation is obtained:

[0094] By the way, φ ｔ can be transformed into the following equation:

[0095] Here, the prior distribution of outliers is p(out|X ｔ ) = ν, then p(in | X ｔ ) = 1 - ν, and φ ｔ can be further transformed into the following equation:

[0096] Here, p(y ｔ │in,X ｔ ) is expressed as follows:

[0097] At this time, φ ｔ and p(y ｔ │X ｔ ) can be expressed as follows:

[0098] Therefore, p(X ｔ │y ｔ ) is expressed as follows:

[0099] Here, K shown in the following formula t ' is introduced.

[0100] Introduced K t A Kalman filter that takes outliers into account using ' is written as follows:

[0101] 1) Initialization

[0102] 2) Observation update step

[0103] 3) Time update step

[0104] 4) Set t←t+1 and return to 2).

[0105] Therefore, the speed of the moving body v t can be estimated using a Kalman filter as follows:

[0106] 1) Initialization

[0107] 2) Observation update step

[0108] 3) Moving object speed estimation step

[0109] 4) Time update step

[0110] 5) Set t←t+1 and return to 2).

[0111] Here, the observed velocity difference 301 is y ｔ The outlier probability 901 corresponds to 1-φ in equation (2.26). ｔ Corresponds to.

[0112] The mean of the posterior estimated speed difference distribution 511 corresponds to the following value in equation (2.28):

[0113] The variance of the posterior estimated speed difference distribution 511 corresponds to the following value in equation (1.29):

[0114] The mean of the predicted speed difference distribution 401 corresponds to the following value in equation (2.31):

[0115] The variance of the predicted speed difference distribution 401 corresponds to the following value in equation (2.32):

[0116] The estimated velocity difference corresponds to the following value in equation (2.28):

[0117] The velocity of the moving body corresponds to the following value in equation (2.30):

[0118] The measuring device 3 is ｔ By introducing y ｔ If the difference between this and the value below is large, the value below will be output with priority.

[0119] As a result, even if a large measurement error occurs in first velocity 103 due to the influence of spinning or sliding of the moving body, the measurement error can be suppressed and the velocity of the moving body can be output.

[0120] 8 is a graph showing an example of the velocity calculated by the measuring device 3 according to the second embodiment. The velocity of the moving object increases at a constant pace from a stationary state until 250 seconds, then moves at a constant speed of 25 m / s from 250 seconds to 750 seconds, decreases at a constant pace from 750 seconds to 1000 seconds, and finally comes to a standstill. The velocity v of the first sensor 101 is １，ｔ A dispersion error, which is an error in the high frequency band, occurs in the time period t1. Also, an instantaneous error occurs at 500 seconds.

[0121] On the other hand, the velocity v of the second sensor 201 ２，ｔ It can be seen that an offset error, which is an error in the low frequency band, occurs in the measurement device 3 and increases over time. ２，ｔ The offset error is removed from the velocity v of the first sensor 101, and the velocity is estimated with high accuracy. １，ｔ It can be seen that the variation error is suppressed and that there is almost no effect of the instantaneous error at 500 seconds. The velocity of the moving body calculated by measurement device 1 is somewhat affected, but the velocity of the moving body calculated by measurement device 3 is hardly affected.

[0122] The advantages of the measurement device 3 according to the second embodiment will be described. The posterior estimated speed difference distribution calculation unit 500 calculates the average of the posterior estimated speed difference distribution 511 by using a predetermined weight set in consideration of the outlier probability 901. When a large measurement error occurs in the observed speed difference 301 due to the influence of the mobile body spinning or sliding with respect to the first speed 103, the outlier 901 becomes large. Therefore, the average of the prior predicted speed difference distribution 401 is emphasized in calculating the average of the posterior estimated speed difference distribution 511.

[0123] According to the second embodiment, in addition to the effects of the measuring device 1, the measuring device 3 can measure the speed of a moving body with high accuracy even when a large momentary error occurs in the first sensor.

[0124] Fig. 9 is a diagram illustrating an example of a hardware configuration other than the first sensor 101 and the second sensor 201 in the measurement device 1, 2, or 3. Fig. 10 is a diagram illustrating another example of the hardware configuration illustrated in Fig. 9.

[0125] In the measuring device 1, 2, or 3, components other than the first sensor 101 and the second sensor 201 are configured, for example, as a control unit 42. In this case, the control unit 42 is configured, for example, as at least one processor 42a and at least one memory 42b. The processor 42a is, for example, a central processing unit (CPU) that executes a program stored in the memory 42b. In this case, the functions of the measuring device 1, 2, or 3 are realized by software, firmware, or a combination of software and firmware. The software and firmware can be stored as a program in the memory 42b. With this configuration, the program for realizing the functions of the control unit 42 is executed by a computer.

[0126] The memory 42b is a computer-readable recording medium, and is, for example, a volatile memory such as a Random Access Memory (RAM) or a Read Only Memory (ROM), a nonvolatile memory, or a combination of a volatile memory and a nonvolatile memory.

[0127] The control unit 42 may have a plurality of processors 42a and a plurality of memories 42b. In this case, the functions of the measurement device 1, 2, or 3 are realized by the plurality of processors 42a and the plurality of memories 42b.

[0128] The control unit 42 may be configured with a processing circuit 42c as dedicated hardware, such as a single circuit or a composite circuit. The processing circuit 42c may be, for example, a system LSI. In this case, the functions of the measurement devices 1, 2, and 3 are realized by the processing circuit 42c.

[0129] The features of the above-described embodiments can be combined with each other.

[0130] 1, 2, 3... Measurement device 100... First speed calculation unit 101... First sensor 102... Detected value of first sensor 103... First speed 200... Second speed calculation unit 201... Second sensor 202... Detected value of second sensor 203... Second speed 300... Observed speed difference calculation unit 301... Observed speed difference 400... Pre-estimated speed difference distribution prediction unit 401... Pre-estimated speed difference distribution 500, 510... Posterior estimated speed difference distribution calculation unit 501, 511... Posterior estimated speed difference distribution 600... Estimated speed difference calculation unit 601... Estimated speed difference 700... Mobile body speed calculation unit 800... Mobile body position calculation unit 900... Outlier probability calculation unit 901... Outlier probability

Claims

1. A first sensor that detects the position, velocity, acceleration, or jerk that is the rate of change of acceleration of a moving object; A second sensor that detects the position, velocity, acceleration, or jerk that is the rate of change of acceleration of the moving object; A first velocity calculation unit that calculates a first velocity of the moving object using a detection value detected by the first sensor; A second velocity calculation unit that calculates a second velocity of the moving object using a detection value detected by the second sensor; An observed velocity difference calculation unit that calculates an observed velocity difference by subtracting the first velocity from the second velocity; A posterior estimated velocity difference distribution calculation unit that calculates a posterior estimated velocity difference distribution that is a distribution of the velocity differences based on the observed velocity difference and a priori predicted velocity difference distribution that is a prediction of the distribution of the velocity differences between the first velocity and the second velocity; A priori predicted velocity difference distribution prediction unit that calculates the priori predicted velocity difference distribution at the next time point based on the posterior estimated velocity difference distribution; An estimated velocity difference calculation unit that calculates a representative value of the posterior estimated velocity difference distribution as an estimated velocity difference that is an estimate of the velocity difference; A moving object velocity calculation unit that calculates the velocity of the moving object by subtracting the estimated velocity difference from the second velocity. A measuring device comprising:

2. The measurement device according to claim 1, wherein the mean of the posterior estimated speed difference distribution is calculated by a weighted average of the mean of the prior predicted speed difference distribution and the observed speed difference.

3. The measurement device according to claim 1 or 2, further comprising an outlier probability calculation unit that calculates an outlier probability, which is the probability that the observed speed difference will be an outlier, based on the a priori predicted speed difference distribution and the observed speed difference, and wherein the posterior estimated speed difference distribution calculation unit calculates the posterior estimated speed difference distribution based on the observed speed difference, the a priori predicted speed difference distribution, and the outlier probability.

4. The measurement device according to claim 3, wherein the distribution of observed speed differences is a mixed distribution of a distribution of normal values ​​and a distribution of outliers, and the outlier probability is calculated by assuming the distribution of normal values ​​to be a Gaussian distribution and the distribution of outliers to be a uniform distribution.

5. The measurement device according to claim 1 or 2, characterized in that the estimated speed difference, the posterior estimated speed difference distribution, and the prior predicted speed difference distribution are calculated by a Kalman filter, and the Kalman gain of the Kalman filter is calculated by dividing the prior error covariance matrix of the estimated speed difference by the sum of the prior error covariance matrix and the variance of the observed speed difference.

6. The measurement device according to claim 3 or 4, wherein the estimated speed difference, the posterior estimated speed difference distribution, and the prior predicted speed difference distribution are calculated by a Kalman filter, and the Kalman gain of the Kalman filter is calculated by multiplying the prior error covariance matrix of the estimated speed difference by the probability that it will be a normal value, and dividing the result by the sum of the product of the prior error covariance matrix by the probability that it will be a normal value and the variance of the observed speed difference, and the probability of it being a normal value is calculated by subtracting the outlier probability from 1.

7. A measuring device according to any one of claims 1 to 6, characterized in that the first sensor is a rotation speed detection sensor consisting of a roller that rotates in conjunction with the movement of the moving body, and the second sensor is an acceleration sensor that detects acceleration in the direction of movement of the moving body.

8. A measurement device according to any one of claims 1 to 7, further comprising a moving body position calculation unit that calculates the distance traveled by the moving body from a predetermined point based on the speed of the moving body calculated by the moving body speed calculation unit.

9. A measurement method comprising: calculating a first velocity of a moving body using a first sensor; calculating a second velocity of the moving body using a second sensor; calculating an observed velocity difference by subtracting the first velocity from the second velocity; calculating a posterior estimated velocity difference distribution, which is a distribution of the velocity difference, based on a prior predicted velocity difference distribution, which is a prediction of the distribution of the velocity difference between the first velocity and the second velocity, and the observed velocity difference; calculating the prior predicted velocity difference distribution for the next time point based on the posterior estimated velocity difference distribution; calculating a representative value of the posterior estimated velocity difference distribution as an estimated velocity difference, which is an estimate of the velocity difference; and calculating the velocity of the moving body by subtracting the estimated velocity difference from the second velocity.

Images