JPH0477251B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a state estimation system that integrates different types of sensors, among state estimation systems using multi-sensors, and an estimation method thereof.

[Background of the invention]

A multi-sensor state estimation system that estimates the state of a target such as the movement of a moving body using multiple sensors and estimation mechanisms is superior to a state estimation system using a single sensor in the following points: .

(1) Even if one sensor fails, data from other sensors can be used, increasing the functional reliability of the system.

(2) The observation area or monitoring area becomes wider.

(3) The accuracy of state estimation can be improved.

For this reason, a state estimation system using multiple sensors has come to be considered. Regarding multi-sensor systems, integration of sensor estimation mechanisms of the same type has been considered and put into practical use. That is,
This is a system that integrates the same type of sensor and estimation mechanism with the same type of state estimation method. In this case, an index A _i of the magnitude of the estimation error is calculated corresponding to the estimated state amount X _i of each sensor. A _i corresponds to the estimation error covariance P in estimation using a Kalman filter.
Here, if A _i A _j , the state estimate X _i of sensor i is better, and if A _i A _j , the state estimate X _j of sensor j is better, so that A _i is better for each sensor. Calculated. In order to integrate the state estimates from each sensor, if we compare the index A _i of the estimation error size,
It will be a good thing.

However, if sensor i and sensor j are different types and have different state estimation mechanisms, the above method cannot be applied. This is because, although the index A _i of the magnitude of estimation error in each sensor is not calculated on the same scale, there are some sensors for which A _i is not calculated completely.

[Purpose of the invention]

The present invention provides a multi-sensor state estimation system and estimation method for integrating estimation data in a multi-sensor state estimation system that integrates different types of sensors and estimation mechanisms based on a priori information regarding the state estimation data of each sensor. Our goal is to provide the following.

[Summary of the invention]

In order to achieve such an objective, in the present invention, a priori information regarding state estimation data of each sensor is expressed as a fuzzy quantity, the effectiveness of each sensor is evaluated using fuzzy inference, and this effectiveness evaluation is performed. The feature is that data is integrated based on the results. That is, compared to conventional methods, the present invention is characterized by the integration of estimated data using a priori information possessed by humans.

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a state estimation system using multi-sensors, which is the object of the present invention. Each sensor site consists of an observation mechanism (sensor) 2 and an estimation mechanism 3. Sensor 2 observes the movement of target 5,
Send observation data Y _i to the state estimation mechanism 3. The state estimating mechanism 3 estimates the state of motion of the target from Y _i , sets this as X _i , and calculates an estimation error index A _i that is a measure of the magnitude of the state estimation error. Each sensor site is a conventional single target state observation and estimation system. Estimated data X _i and estimated error index A _i are transferred from these sensor sites to the estimated data integration system 4. The estimated data integration system 4
This is a system that implements the state estimation data integration method according to the present invention.

First, a method for integrating data from different types of sensors and estimation mechanisms, which is the principle of the present invention, will be explained.

For estimated data from each sensor site,
It is possible to obtain a priori information such as "The effective distance of sensor A is about 20 km, and measurement accuracy decreases if it is longer than that." or "Sensor A has higher measurement accuracy than sensor B in a specific area." shall be. According to these types of information, the target state is defined as X=X ₁ x… _x X _N
is given in the direct product space of For example, X ₁ is the set of magnitudes of the estimation error index A _i , and X ₂ is . It depends on the distance from the sensor to the target. Among the sets of individual sets X _k , fuzzy sets A _l ^k , j=
1. Define n _k . For example, A ₁ ² is information such as "the distance from the sensor to the target is small", and A ₂ ² is information such as "the distance from the sensor to the target is normal". Under such a fuzzy set A _jk , X _i
As a priori information, the degree of effectiveness when selecting U _ij ^k
It shall be given by U _ij ^k is one of the fuzzy sets {U _p , p=1, ..., m} representing effectiveness, and U _p is a set U= representing the degree of effectiveness.
It is a fuzzy set on {u ₁ ,...u _M }. U _p is, for example, a set of types such as ``very effective'' and ``normal.''

Let the membership function of A _k ^j be μ _Aj ^k (x), and the membership function of U _p be μ _e (u). Examples of membership functions and fuzzy sets are shown below.

As an example of a priori information, it is assumed that X ₁ : Size of estimation error index X ₂ : Distance from sensor to target X ₃ : Time interval from target observation time at each sensor to data integration.

The membership function for X ₁ is shown in Figure 2.

Here, VB, B, M, S, and VS represent VB: very large, B: large, M ₁ : normal, S: small, VS: very small. Similarly, the membership functions are shown in FIG. 3 for X ₂ and in FIG. 4 for X ₃ .

Here, N: near, M ₂ : normal, D: far, S ¹ : short, M ₃ : normal, L: long. Also, the fuzzy set U _p representing the effectiveness
The membership function of is shown in FIG. Here, VL: Very low effectiveness L: Low effectiveness M ₄ : Normal H: High effectiveness VH: Very high effectiveness.

Now, as a priori information, when X ₁ is VB, the validity VH X ₂ is the validity L of M ₂ . This information is compiled for each sensor side to create a table of a priori information as shown in FIG. Since this a priori information differs depending on the target system, it must be created based on the knowledge of experts and professionals involved in the actual work. Now, from this knowledge, the degree E _i indicating the effectiveness corresponding to the target estimated data X _i is calculated as follows.

Step 1: First, calculate the vector x = (x ₁ ,...x _N ) corresponding to X = X ₁ x...xX _N from X _i Step 2: For each x _k , calculate the membership of the fuzzy set A _j ^k Find the value μ _Aj ^k (xk) of the function. This indicates that the degree of belonging of xk to A _j ^k is μ _Aj ^k (xk). Writing this as (μ _Aj ^k (x _k ), A _j ^k ), x _k
In the above, {(μ _Aj ^k (x _k ), A _j ^k )} ⁽ⁱ⁾ _j=1 ,...n _k pairs are created.

Step 3: When _{A j} ^k , it is given from a priori information that it has validity U _ij ^k , so the degree of membership of U _ij ^k of x _k is {(μ ^k _Aj (x _k ), U _ij ^k )} ⁽ⁱ⁾ _j=1 ,...n _j ^k . Since U _ij ^k is one of {U _p , _p=1 ,... _n }, evaluating this with U _p yields {(μ _e ^k , U _e )} ⁽ⁱ⁾ _e=1 ,... _n . Here, μ _e ^k is {U _ij ^k , j=1,...n _k }
It is calculated as the sum of the membership functions of U _ij ^k , which is U _e in . That is, it is calculated as μ _e ^k = μ _Aj ^k ₁ (x _k )…μ _Ajn ^k (x _k ) f ₁ f ₂ = f ₁ + f ₂ − _{f 1} *+ ₂ .

Step 4. Since the fuzzy set U _e is U _e = _M 〓 ⁱ⁼¹ μ U _e (u _i )/u _i , evaluate this on the set U = {u ₁ ,..., u _M } and write {(u _o ^*k , u _o )} ⁽ⁱ⁾ Find _o=1 ,..., _M. Here, u _o ^*k is calculated as μ _o ^*k = (μ ₁ K∧ μ _U1 (U _o )…(μ _n K∧ μ _Un (u _n )). ∧ is (f ₁ ∧f ₂ ) = Let min(f ₁ , f ₂ ).

Step 5.X A function U ^k indicating the effectiveness of _k is given by U ^k = {(μ _o ^*k , U _o )} ⁽ⁱ⁾ _o=1 ,... _M . The effectiveness U~ of all X is evaluated, and it is assumed that U~ ⁽ⁱ⁾ is given by a linear combination of the effectiveness ^Uk of each set _Xk . That is, utility independence for each X ₁ ,...X _N is assumed. At this time, U~ ⁽ⁱ⁾ is calculated as U~ ⁽ⁱ⁾ = _N 〓 ^k=1 α _k U ^k . Here, U~ ⁽ⁱ⁾ is the validity membership function of the data from the i-th sensor and can be written as U~ ⁽ⁱ⁾ _M 〓 ^e=1 μU~ ⁽ⁱ⁾ (u _e )/u _e . can. From this, the effectiveness E _i when the state estimator X^ _i is selected can be calculated using E _i =μU~ ⁽ⁱ⁾ (u ₁ )∧...∧μU~ ⁽ⁱ⁾ (u _M ).

The above is the method for calculating the effectiveness E _i for the state estimation amount X _i from each sensor. Using this, if you want to select the most effective sensor from each sensor, you can select it by X^=X _i* st max _i=i* E _i , or you can use linear combination to When integrating data, X∧=1/E ₁ +E ₂ +...E _NN 〓 ⁱ⁼¹ E _i X _i .

The specific configuration and processing flow of the state estimation data integration system 4 based on the above integration method are shown in FIGS. 7 and 8. FIG. 7 is a block diagram of the state estimation data integration system, and FIG. 8 is a flowchart showing the flow of processing performed by the processing device 7. The receiving device 6 receives data from each sensor and stores it in a buffer memory within the receiving device 6. The processing device 7 reads this data (Step 11 in Figure 8), performs the above-mentioned effectiveness evaluation estimation data integration process (Steps 12 and 13 in Figure 8),
Calculate estimated data X∧. At this time, a priori information stored in the storage device 8 is used. The display device 9 for man-machine response performs situation monitoring performed by the processing device 7 (step 14 in FIG. 8) and man-machine response processing for correcting a priori information (step 16 in FIG. 8). The modified a priori information is stored in the storage device 8.

〔Effect of the invention〕

As described above, according to the present invention, it is possible to configure a multi-sensor system that integrates not only the conventional homogeneous sensor estimation mechanism but also different types of sensors and estimation mechanisms. Because state estimation data can be integrated using such information,
Highly accurate state estimation can be performed.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an example of a multi-sensor state estimation system according to the present invention, FIGS. 2 to 5 are diagrams showing membership functions of fuzzy variables, FIG. 6 is a diagram showing a table of a priori information, and FIG. FIG. 7 is a block diagram of an example of the state estimation data integration system according to the present invention, and FIG. 8 is a flowchart of the processing of the processing device shown in FIG. 1: Sensor site, 2: Sensor, 3: Estimation mechanism, 4: State estimation data integration system, 5: Target, 6: Receiving device, 7: Processing device, 8: Storage device, 9: Display device.

Claims (1)

[Claims] 1. In a multi-sensor state estimation system that estimates the motion state of a target by integrating sensors and estimation mechanisms that obtain different types of data, the data obtained for each sensor and its estimation mechanism is expressed as a first fuzzy quantity, and the a storage device that represents the validity of the obtained data as a second fuzzy quantity and stores the second fuzzy quantity in correspondence with the first fuzzy quantity as a priori information of the sensor and its estimation mechanism; means for integrating the first fuzzy quantities stored in the storage device corresponding to the observation data of the respective sensors to obtain estimated data regarding the motion state of the target; and means corresponding to the observation data of the respective sensors. means for determining the effectiveness of the estimated data by integrating the second fuzzy quantities representing the effectiveness of each of the first fuzzy quantities; and determining the effectiveness of the estimated data from the obtained estimated data and the effectiveness of the estimated data. A state estimation system comprising: means for obtaining final estimated data regarding the motion state of the vehicle.