STF parameter calculation device and STF parameter calculation method

Abstract

The target tracking device includes an STF parameter setting unit (21) that acquires observed values ​​and predicted values ​​of a target and dynamically sets an STF parameter ρ based on the acquired observed values ​​and predicted values.

Description

[Technical Field]

[0001] The present disclosure relates to target tracking techniques. [Background technology]

[0002] When tracking targets that are difficult to fully model using sensors such as radar or optical cameras, a multi-model approach is often used, in which the reliability of each model is calculated based on the likelihood of multiple models and then weighted integration is performed to achieve tracking. However, multi-model approaches require a high computational load because multiple models are operated, and it is known to be difficult to perform correlation processing on multiple observations. Therefore, there is a demand for a target tracking filter that can maintain tracking accuracy equivalent to that of a multi-model with a single model. One document that discloses a technology that meets these expectations is Non-Patent Document 1, which proposes a Strong Tracking Filter (STF). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] DF Crouse, “Strong Tracking Filters: Derivation and Improved Heuristic,” in 2019 22nd International Conference on Information Fusion (FUSION), Jul. 2019, pp. 1-8. Summary of the Invention [Problem to be solved by the invention]

[0004] The STF is a filter that achieves a certain level of tracking accuracy by adaptively expanding the prediction error in the tracking filter according to the difference between the predicted and observed values, based on the premise that it is difficult to accurately estimate the amount of error for a single model in advance. The degree to which the STF prediction error is expanded depends on the STF parameter ρ, which is set in advance to a fixed value between 0 and 1. However, the optimal value of this STF parameter ρ differs depending on the tracking situation, which presents a problem in that the value of the set STF parameter ρ cannot be dynamically adjusted.

[0005] The present disclosure has been made to solve such problems, and aims to provide a target tracking technology that allows the value of the STF parameter ρ to be dynamically set. [Means for solving the problem]

[0006] One aspect of a target tracking device according to an embodiment of the present disclosure includes an STF parameter setting unit that acquires observed values ​​and predicted values ​​of a target, and dynamically sets an STF parameter ρ based on the acquired observed values ​​and predicted values. [Effects of the Invention]

[0007] According to the target tracking device according to the embodiment of the present disclosure, the value of the STF parameter ρ can be dynamically set. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a functional block diagram showing an example of the configuration of a target tracking device. [Figure 2A] FIG. 2 is a diagram illustrating an example of the hardware configuration of a target tracking device. [Figure 2B] FIG. 2 is a diagram illustrating an example of the hardware configuration of a target tracking device. [Figure 3] 4 is a flowchart showing the operation of the target tracking device. [Figure 4] FIG. 2 is a functional block diagram showing a configuration example of an STF parameter setting unit. DETAILED DESCRIPTION OF THE INVENTION

[0009] Various embodiments of the present disclosure will be described in detail below with reference to the drawings. In the drawings, identical or similar parts are designated by identical or similar reference numerals, and redundant explanations of such parts will be omitted. In addition, in this disclosure, the term "or" is used to mean an inclusive logical OR unless otherwise specified.

[0010] Embodiment 1 <Configuration> A target tracking device according to a first embodiment of the present disclosure will be described with reference to FIG. 1. FIG. 1 is a functional block diagram illustrating an example configuration of a target tracking device. As shown in FIG. 1, the target tracking device includes an observation processing unit 11, a prediction processing unit 12, an STF parameter setting unit 21, an STF processing unit 16, and a time update processing unit 17. The STF parameter setting unit 21 is a functional unit that sets STF parameters used in STF processing. To achieve this function, the STF parameter setting unit 21 includes, for example, an average residual calculation processing unit 13, a distance calculation processing unit 14, and an STF parameter determination processing unit 15.

[0011] (Observation processing unit) The observation processing unit 11 receives signals from sensors such as radar or optical cameras (not shown), detects the state of the target from the received signals, and generates observation values ​​z(t) as vectors. Examples of the state of the target include physical quantities such as position or velocity, or image features. The observation processing unit 11 supplies the generated observation values ​​z(t) to the mean residual calculation processing unit 13, the STF processing unit 16, and the time update processing unit 17.

[0012] (Prediction processing unit) The prediction processor 12 predicts the state of the target at the next time t. Specifically, similar to a known Kalman filter, the prediction processor 12 predicts the state x as a vector at the current time t-1. t-1, the state transition matrix, and the noise vector to generate a predicted value x^(t|t-1) and a prediction error covariance matrix P~(t|t-1) as vectors. The prediction processing unit 12 supplies the generated predicted value x^(t|t-1) and prediction error covariance matrix P~(t|t-1) to the average residual calculation processing unit 13. Note that the generated prediction error covariance matrix P~(t|t-1) is data used by the distance calculation processing unit 14, so the prediction processing unit 12 may supply the generated prediction error covariance matrix P~(t|t-1) directly to the distance calculation processing unit 14.

[0013] (Average residual calculation processing unit) The average residual calculation processing unit 13 receives the observed value z(t) from the observation processing unit 11, and receives the predicted value x^(t|t-1) and the prediction error covariance matrix P~(t|t-1) from the prediction processing unit 12. The average residual calculation processing unit 13 calculates the average e(t) of the difference between the predicted value and the observed value using the predicted value x^(t|t-1) and the observed value z(t). The average of the difference between the predicted value and the observed value is calculated, for example, according to the following equation (1). TIFF0007734873000001.tif16170

[0014] where H is the observation matrix and n is 2 or more Each represents a design parameter with an arbitrary integer value.

[0015] The average residual calculation processing unit 13 supplies the prediction error covariance matrix P~(t|t-1) received from the prediction processing unit 12 and the calculated average e(t) of the differences between the predicted values ​​and the observed values ​​to the distance calculation processing unit 14. When the prediction processing unit 12 supplies the prediction error covariance matrix P~(t|t-1) to the distance calculation processing unit 14, the average residual calculation processing unit 13 does not need to supply the prediction error covariance matrix P~(t|t-1) to the distance calculation processing unit 14.

[0016] (Distance calculation processing unit) The distance calculation processing unit 14 receives the prediction error covariance matrix P~(t|t-1) from the prediction processing unit 12 or the average residual calculation processing unit 13, and receives the average e(t) of the differences between the predicted values ​​and the observed values ​​and the observation matrix H from the average residual calculation processing unit 13. The distance calculation processing unit 14 performs processing to calculate the square of the Mahalanobis distance as the statistical distance using the received prediction error covariance matrix P~(t|t-1), the average e(t) of the differences between the predicted values ​​and the observed values, and the observation matrix H.

[0017] The distance calculation processing unit 14 first calculates the observation residual covariance matrix S(k) by the following equation (2). TIFF0007734873000002.tif7166

[0018] In equation (2), R represents the observation covariance matrix, and T represents the transpose of the matrix.

[0019] Next, the distance calculation processing unit 14 calculates the square value d(t) of the Mahalanobis distance using the observation residual covariance matrix S(k) according to the following equation (3). d(t)=e(t) T S(t) -1 e(t) (3)

[0020] The distance calculation processing unit 14 supplies the square value d(t) of the calculated Mahalanobis distance to the STF parameter determination processing unit 15.

[0021] (STF parameter determination processing unit) The STF parameter determination processor 15 determines the STF parameter ρ according to the following equation (4). ρ(t)=min(1,P / d(t)) (4)

[0022] Here, P is an arbitrary parameter determined from a chi-square distribution according to the number of dimensions of the observed or predicted tracking values. The parameter P is a parameter determined by the number of dimensions (or the magnitude of the vector) of the observed or predicted values. Here, the dimensions (state quantities) of the observed and predicted values ​​are not limited to specific dimensions (e.g., the position or velocity of the target, the pixel value or size, or the shape feature quantity, etc.) and may include any dimension. The STF parameter determination processing unit 15 obtains the value of the parameter P from the observed value z(t). Furthermore, min(·,·) is a function that returns the value of the smaller of its arguments. Therefore, when P / d(t) is a value equal to or greater than 1, the value of 1 is used as the value of the STF parameter ρ, and when P / d(t) is a value less than 1, the value of P / d(t) is used as the value of the STF parameter ρ. Therefore, the value of the STF parameter ρ can be dynamically and appropriately set so that the value of P is not taken into consideration when it is an outlier, and the value of P is taken into consideration when it is not an outlier.

[0023] The STF parameter determination processing unit 15 supplies the determined STF parameter ρ to the STF processing unit 16 .

[0024] (STF processing unit) The STF processor 16 is a functional unit that receives the observed value z(t) from the observation processor 11 and the STF parameter ρ from the STF parameter determination processor 15, and performs STF processing using the observed value z(t) and the STF parameter ρ. Briefly, the STF processing calculates a residual covariance matrix using a prediction error covariance matrix, the cross product of the observed prediction error at the current time, and the STF parameter ρ (Equation (60) in Non-Patent Document 1). The STF processing calculates an expansion coefficient λ of the prediction error covariance matrix using the observation error covariance matrix and the calculated residual covariance matrix (Equation (71) in Non-Patent Document 1). The calculated expansion coefficient λ is applied to the prediction error covariance matrix to update the prediction error covariance matrix (Equation (76) in Non-Patent Document 1), and a new prediction error covariance matrix P(t|t-1) is calculated. For details of the STF processing, please refer to known documents such as Non-Patent Document 1. The STF processor 16 outputs the updated prediction error covariance matrix P(t|t-1) to the time update processor 17.

[0025] (Time update processing unit) When the time update processing unit 17 receives the observation value z(t) from the observation processing unit 11 and the updated prediction error covariance matrix P(t|t-1) from the STF processing unit 16, it updates the time and supplies the updated time, observation value z(t), and updated prediction error covariance matrix P(t|t-1) to the prediction processing unit 12.

[0026] Next, an example of the hardware configuration of a target tracking device will be described with reference to Figures 2A and 2B. Each function of the target tracking device is realized by a processing circuitry. The processing circuitry may be a dedicated processing circuit 100a as shown in Figure 2A, or a processor 100b that executes a program stored in a memory 100c as shown in Figure 2B.

[0027] When the processing circuitry is a dedicated processing circuit 100a, the dedicated processing circuit 100a may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof. The functions of the target tracker may be realized by separate processing circuits, or the functions of the target tracker may be realized together in a single processing circuit.

[0028] When the processing circuitry is a processor 100b, the functions of the target tracking device are realized by software, firmware, or a combination of software and firmware. The software and firmware are written as programs and stored in the memory 100c. The processor 100b realizes the functions of the target tracking device by reading and executing the programs stored in the memory 100c. Here, examples of the memory 100c include non-volatile or volatile semiconductor memories such as random access memory (RAM), read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), and electrically erasable programmable read-only memory (EEPROM), as well as magnetic disks, flexible disks, optical disks, compact disks, minidisks, and DVDs.

[0029] It is also possible to realize some of the functions of the target tracking device using dedicated hardware, and other functions using software or firmware. In this way, the processing circuit can realize the functions of the target tracking device using hardware, software, firmware, or a combination of these.

[0030] <Operation> Next, the operation of the target tracking device according to the first embodiment will be described with reference to FIG.

[0031] (Step ST11) In step ST11, the observation processing unit 11 receives a signal from a sensor such as a radar or an optical camera (not shown), detects the state of the target from the received signal, and generates an observation value z(t).

[0032] (Step ST12) In step ST12, the prediction processing unit 12 predicts the state of the target at the next time t. Specifically, the prediction processing unit 12 predicts the state x at the current time t-1. t-1 , the state transition matrix, and the noise vector are used to generate the predicted value x̂(t|t-1) and the prediction error covariance matrix P̂(t|t-1).

[0033] (Step ST13) In step ST13, the average residual calculation processing unit 13 calculates the average e(t) of the difference between the predicted value and the observed value using the predicted value x^(t|t-1) and the observed value z(t).

[0034] (Step ST14) In step ST14, the distance calculation processing unit 14 uses the prediction error covariance matrix P~(t|t-1), the average e(t) of the differences between the predicted values ​​and the observed values, and the observation matrix H to calculate the square of the Mahalanobis distance.

[0035] (Step ST15) In step ST15, the STF parameter determination processing unit 15 determines the STF parameter ρ using equation (4).

[0036] (Step ST16) In step ST16, the STF processing unit 16 performs STF processing using the observed value z(t) and the STF parameter ρ.

[0037] (Step ST17) In step ST17, the time update processing unit 17 updates the time.

[0038] According to the target tracking device of embodiment 1 described above, the value of the STF parameter ρ is dynamically set in accordance with equation (4). As shown in equation (4), the STF parameter ρ is determined based on an arbitrary parameter P determined from a chi-square distribution according to the number of dimensions of the observed or predicted tracking values, so the value of the STF parameter ρ can be set appropriately depending on the tracking situation.

[0039] <Modification> Various modifications of the first embodiment will be described below.

[0040] <Variation 1> The following equations (5) and (6) may be used instead of equation (1) in embodiment 1. That is, the average may be a weighted average. e(0)=0 (5) TIFF0007734873000003.tif11166

[0041] <Variation 2> Equation (3) in the first embodiment may be replaced by finding a statistical test quantity such as Kullback-Leibler divergence, Fisher information, or t-value using e or e and S. In that case, P in equation (4) is changed to a numerical value corresponding to each value.

[0042] <Variation 3> The results obtained by equation (4) do not necessarily have to be applied to STF processing. For example, the tracking prediction error can be directly adjusted by using ρ(t) obtained by equation (4) as the prediction error correction amount and applying it to the parameters related to the drive noise of the Kalman filter. In this case, the value range of the parameter to be adjusted can be adjusted by multiplying ρ(t) by an arbitrary coefficient k.

[0043] Embodiment 2 The target tracking device may include an STF parameter setting unit 21A as shown in Fig. 4, instead of the STF parameter setting unit 21 according to embodiment 1. As shown in Fig. 4, the STF parameter setting unit 21A includes a target detection accuracy calculation processing unit 22 and an STF parameter determination processing unit 15A.

[0044] (Target detection accuracy calculation processing unit) The target detection accuracy calculation processing unit 22 is a functional unit that calculates the target detection accuracy according to known technology. For example, when tracking is performed using radar, the target detection accuracy may be derived from a known model such as a radar equation, or may be calculated from the error from the true value using Monte Carlo simulation. Furthermore, the target detection accuracy may be calculated from measurement results in a real environment, or, if the true value is unknown, may be expressed as its expected value and variance, etc. The target detection accuracy calculation processing unit 22 supplies the calculated target detection accuracy to the STF parameter determination processing unit 15A.

[0045] (STF parameter determination processing unit) When the STF parameter determination processor 15A acquires the calculated target detection accuracy, it references a ρ table (not shown) in which values ​​of the target detection accuracy and values ​​of the STF parameter ρ are previously associated with each other, and acquires the value of the STF parameter ρ corresponding to the calculated target detection accuracy. The ρ table is stored in the memory 100c or a storage device (not shown).

[0046] For example, in the case of radar, the accuracy of distance measurement changes depending on the distance to the target. Therefore, if there is a correlation between the target detection accuracy of the sensor and ρ, as in the second embodiment, a ρ table may be prepared in advance to hold values ​​of the STF parameter ρ corresponding to the target detection accuracy, and the value of ρ may be changed according to the ρ table.

[0047] It is possible to combine the embodiments, and to modify or omit each embodiment as appropriate. [Industrial Applicability]

[0048] The target tracking device of the present disclosure can be used as a device that tracks a target using a sensor such as a radar or an optical camera. [Explanation of symbols]

[0049] 11 observation processing unit, 12 prediction processing unit, 13 average residual calculation processing unit, 14 distance calculation processing unit, 15 STF parameter determination processing unit, 15A STF parameter determination processing unit, 16 STF processing unit, 17 time update processing unit, 21 STF parameter setting unit, 21A STF parameter setting unit, 22 target detection accuracy calculation processing unit, 100a processing circuit, 100b processor, 100c memory.

Claims

1. An STF parameter setting unit acquires target observed and predicted values, and dynamically sets the STF parameter ρ based on the mean residual of the acquired observed and predicted values. A target tracking device equipped with [a specific feature].

2. The STF parameter setting unit is A mean residual calculation processing unit that calculates the mean residual difference between the observed value and the predicted value, A distance calculation processing unit calculates the statistical distance of the mean residuals from the mean residuals and the residual covariance matrix, An STF parameter determination processing unit that determines the STF parameter ρ based on the calculated statistical distance, A target tracking device according to claim 1, comprising:

3. The aforementioned statistical distance is the square of the Mahalanobis distance, The STF parameter determination processing unit determines the STF parameter ρ to be the smaller of the calculated squared Mahalanobis distance, the value of an arbitrary parameter determined from the chi-squared distribution, and 1. A target tracking device as described in claim 2.

4. A target tracking method performed by a target tracking device equipped with an STF parameter setting unit, The STF parameter setting unit performs the steps of acquiring target observed and predicted values, and calculating the mean residual between the acquired observed and predicted values. The STF parameter setting unit performs the step of calculating the statistical distance of the mean residuals from the mean residuals and the residual covariance matrix, The STF parameter setting unit determines the STF parameter ρ to be the smaller of the calculated squared Mahalanobis distance, the value of an arbitrary parameter determined from the chi-squared distribution, and 1. A target tracking method comprising the following features.

Images