JP6541538B2 - Target tracking device and radar device - Google Patents

Description

  The present invention relates to a target tracking technique for estimating a track of a target to be observed, such as a moving object, based on observation information from a sensor.

  There is a target tracking technique of detecting a moving target such as an airplane and estimating a track of the moving target based on observation information by an active sensor such as a radar sensor or a passive sensor such as an infrared sensor. For example, Non-Patent Document 1 (Donald B. Reid, "An Algorithm for Tracking Multiple Targets") discloses an algorithm that enables simultaneous tracking of a plurality of moving targets.

  However, if ambiguity exists in sensor observation values such as distance, angle (elevation angle and azimuth angle) and Doppler velocity with respect to the moving target, the estimation accuracy of the track is reduced and errors other than the target track are made. There is a problem of generating a wake.

  For example, in the case of MPRF (Middle PRF) radar and HPRF (High PRF) radar, which are pulse radars having relatively high pulse repetition frequency (PRF), the pulse repetition period (PRI) is relatively short. Therefore, ambiguity is likely to occur in the observed value of the distance to the target or the relative velocity of the target. Here, PRF is the reciprocal of the pulse repetition period (PRI).

  On the other hand, in the case of LPRF (Low PRF) radar which is a pulse radar whose PRF is relatively low, the pulse repetition period is longer than the time when the pulse wave reciprocates the assumed maximum detection distance. There is no guilty, but ambiguity is likely to occur in the Doppler velocity observation. Therefore, it is difficult to directly use the Doppler velocity observation value for target tracking, and under the environment where the ambiguity of the Doppler velocity observation value (hereinafter also referred to as "doppler ambiguity") occurs, the correlation performance of the target tracking Was a problem.

  The prior art of patent document 1 (Unexamined-Japanese-Patent No. 2005-83867) exists with respect to this problem. The target tracking device disclosed in Patent Document 1 includes a Doppler velocity information extraction unit that narrows down the Doppler velocity candidate from detection data obtained by observation with a pulse compression radar. The Doppler velocity information extraction unit obtains a first Doppler velocity candidate from the frequency deviation between the transmission frequency and the reception frequency of the pulse compression radar, and performs distance observation by alternately executing up-chirping and down-chirping. The second Doppler velocity is calculated from the distance deviation of the second velocity, and the first Doppler velocity is compared with the second Doppler velocity to narrow down the first Doppler velocity candidates. As a result, good quality Doppler velocity observation values can be obtained, which makes it possible to improve the aforementioned problems caused by Doppler ambiguities.

JP 2005-83867 A (for example, paragraphs 0045 to 0069)

Donald B. Reid, "An Algorithm for Tracking Multiple Targets," IEEE Transactions on Automatic Control. Vol. AC-24. N0. 6, December, 1979.

  However, in the prior art of Patent Document 1, although it is possible to estimate the target track with high accuracy under an environment where only Doppler ambiguity occurs, an environment where ambiguity occurs in the distance observation value or the angle observation value Below, the estimation accuracy of the track is not sufficient, and a false track may occur. Not only Doppler ambiguity, but also highly versatile and efficient, which can determine the target track with high accuracy even in the environment where any ambiguity of various sensor observation values such as distance observation value and angle observation value occurs Good algorithms are required.

  In view of the above, it is an object of the present invention to provide a target tracking device and a radar device that can determine a target track efficiently and with high accuracy regardless of the type of sensor observation value even in an environment where ambiguity occurs. It is to provide.

The target tracking device according to the first aspect of the present invention is a target tracking device that performs tracking processing on the moving target using a sensor that observes the moving target, and the position of the moving target is determined based on the output of the sensor. A signal processing unit for generating detection data, a temporary detection data generation unit for generating a plurality of temporary detection data from the detection data, an observation value forming an existing track obtained by the tracking process in the past, and the temporary detection data And a tracking operation for determining the track of the moving target from among the track candidates using the likelihood calculation unit for calculating a correlation likelihood indicating the likelihood of the track candidate formed by combining with the track candidate and the correlation likelihood comprising a part, and the detection data to each sensor observation value of the observation error distribution determined error distribution data storage unit error distribution data is stored that is included, the observation error distribution, a plurality of probability min There is a mixed distribution in which the temporary detection data generation unit shifts the respective sensor observation values included in the detection data by the representative value of the plurality of probability distributions based on the error distribution data. And generating the plurality of temporary detection data.

  A radar device according to a second aspect of the present invention is characterized by including the sensor for observing the moving target, and the target tracking device.

  According to the present invention, a plurality of temporary detection data is generated by shifting each sensor observation value by the representative value of the plurality of probability distributions in the mixture distribution. A track candidate is generated based on the plurality of temporary detection data, and the track of the moving target can be determined from among the track candidates. Therefore, even in an environment where ambiguities occur in sensor observation values, the target track can be determined efficiently and with high accuracy regardless of the type of the sensor observation values. This reduces the occurrence of erroneous navigation and enables early establishment of the navigation.

BRIEF DESCRIPTION OF THE DRAWINGS It is a block diagram which shows schematic structure of the radar apparatus of Embodiment 1 which concerns on this invention. 5 is a flowchart illustrating an example of a procedure of tracking processing according to Embodiment 1; FIG. 5 is a view showing an example of a mixed distribution according to Embodiment 1; FIG. 6 is a diagram for explaining a method of calculating provisional detection data according to the first embodiment. 5 is a view showing another example of the mixture distribution according to Embodiment 1. FIG. FIG. 10 is a diagram showing still another example of the mixed distribution according to Embodiment 1. It is a figure for demonstrating the correlation between the existing track and temporary detection data. FIG. 2 is a diagram illustrating an example of a hardware configuration according to the first embodiment. It is a block diagram which shows schematic structure of the radar apparatus of Embodiment 2 which concerns on this invention. It is a figure for demonstrating the correlation between the existing track and temporary detection data. FIG. 16 is a flowchart illustrating an example of the procedure of tracking processing according to Embodiment 2. FIG. It is a block diagram which shows schematic structure of the target tracking apparatus of Embodiment 3 which concerns on this invention. FIG. 16 is a flowchart illustrating an example of the procedure of tracking processing according to Embodiment 3. FIG. It is a flowchart which shows the detail of a two-step process (step ST21) shown in FIG.

  Hereinafter, various embodiments according to the present invention will be described in detail with reference to the drawings. Note that components given the same reference numerals throughout the drawings have the same configuration and the same function.

Embodiment 1
FIG. 1 is a block diagram showing a schematic configuration of a radar device 1 according to a first embodiment of the present invention. As shown in FIG. 1, the radar device 1 detects a movement target (hereinafter, also simply referred to as “target”) 10, and detection data Od indicating the position of the target based on the output of the sensor 10. A signal processing unit 20 to generate, an error distribution data storage unit 21 storing error distribution data sets 21d, 21a, and 21v that determine observation error distribution, and a tracking process that performs tracking processing on a movement target based on detection data Od And a unit 30. The target tracking device according to the present embodiment includes a signal processing unit 20, an error distribution data storage unit 21, and a tracking processing unit 30.

  The sensor 10 pulse-modulates a high-frequency signal to generate and output a pulse-like transmission wave, and the transmission antenna unit 12 radiates the transmission wave toward space at every pulse repetition period (PRI). A receiving antenna unit 13 for receiving a reflected wave that has been reflected by a target in the space and received, and a receiving circuit 14 for detecting an output wave of the receiving antenna unit 13 to generate an analog reception signal There is. The reception circuit 14 A / D converts an analog reception signal into a digital reception signal and outputs the digital reception signal to the signal processing unit 20.

The signal processing unit 20 analyzes the phase and power of the digital received signal, and performs sensor observation such as distance observation value to the target, Doppler velocity observation value (hereinafter, also simply referred to as “Doppler observation value”), and angle observation value. Calculate the value. Here, the angle observation value consists of a set of elevation angle observation value and azimuth angle observation value. The signal processing unit 20 is an observation vector z consisting of a set of sensor observation values (distance observation value, Doppler velocity observation value, angle observation value) at each sampling time t _k (hereinafter, also referred to as “time t _k ”). Detection data Od indicating ⁰ _k (k is an integer for specifying time t _k ) is output to the tracking processing unit 30. Further, the signal processing unit 20 acquires, from the error distribution data storage unit 21, the error distribution data Pd corresponding to the sensor observation value of the observation vector z ⁰ _k and supplies the error distribution data Pd to the tracking processing unit 30. . The error distribution data storage unit 21 includes a data set 21d of an observation error distribution group of distance observation values, a data set 21a of an observation error distribution group of angle observation values (elevation angle observation value and direction observation value), and Doppler observation values. A data set 21d of the observation error distribution group is stored. The data sets 21 d, 21 a, 21 d include data of a normal observation error distribution and data of an observation error distribution representing the ambiguity of the sensor observation value.

  In the sensor 10, the transmission circuit 11 controls the high frequency signal between pulse hits or between detections during ranging or speed measurement in order to suppress the appearance probability of the ambiguity of the distance observation value or the Doppler speed observation value. It is desirable to perform intra-pulse code modulation or PRF (pulse repetition frequency) modulation. In addition, also in angle measurement, in order to suppress the appearance probability of the ambiguity of the angle observation value, the transmission circuit 11 can execute transmission frequency modulation between pulse hits or between detections. Furthermore, in order to suppress the appearance probability of the ambiguity at the time of angle measurement, it is desirable to optimize the arrangement of the antenna element groups constituting the transmitting antenna unit 12 and the receiving antenna unit 13 in advance.

As shown in FIG. 1, the tracking processing unit 30 includes an ambiguity control unit 31 that derives a plurality of temporary observation values from each sensor observation value included in the input detection data Od. As described later, the ambiguity control unit 31 generates a plurality of temporary observation vectors {z ^s _k } from one observation vector z ⁰ _k to generate the plurality of temporary observation vectors {z ^s _k }. Has a function of generating a provisional detection data group TOd indicating. Here, the superscript s is an integer number that uniquely identifies each temporary observation vector z ^s _k . Each temporary observation vector z ^s _k is configured by a set of temporary observation values of a plurality of types (distance, Doppler velocity, and angle), similarly to the observation vector z ⁰ _k . In this specification, it is assumed that one temporary observation vector z ^s _k constitutes one temporary detection data. Also, the tracking processing unit 30 determines whether each temporary detection data correlates with the existing track (that is, the track obtained in the past tracking processing), and the past determining the existing track A likelihood operation unit 32 for calculating a correlation likelihood indicating the likelihood of a new track (hereinafter also referred to as a “track candidate”) formed by combining the observation vector of 1 and the temporary observation vector z ^s _k ; A track specification calculator 33 for calculating track specifications used for the track, a tracking track determination unit 34 for determining a target track from among track candidates using correlation likelihood, and a track related data storage 35 It is comprised including. The track related data storage unit 35 is an area for storing track related data such as detection data Pd used for processing in the correlation determination unit 41 and the track specification calculation unit 33.

  Hereinafter, the configuration and operation of the tracking processing unit 30 will be described with reference to FIG. FIG. 2 is a diagram schematically illustrating an example of the procedure of the tracking process according to the first embodiment.

Referring to FIG. 2, the ambiguity control unit 31 stands by until the detection data Od is input (NO in step ST1). When the detection data Od is input at the sampling time t _k (YES in step ST1), the ambiguity control unit 31 uses the detection data Od and the error distribution data Pd simultaneously input with the detection data Od. A plurality of temporary detection data are generated from the (step ST2). Specifically, the ambiguity control unit 31 causes each of the sensor observation values (the distance observation value, the Doppler observation value, and the angle observation value) included in the detection data Od to be a plurality of shift amounts included in the error distribution data Pd. By shifting only, it is possible to generate the temporary detection data group TOd corresponding to the detection data Od. As shift amounts, representative values of a plurality of probability distributions (hereinafter also referred to as "components") mixed in the observation error distribution are used. Here, “a plurality of probability distributions mixed in the observation error distribution” means that, for example, as shown in the following equation, the observation error distribution P _y is formed by weighted linear combination of a plurality of probability distributions Means

Here, y is a symbol indicating one of the sensor observation value types (distance, Doppler velocity, elevation angle or azimuth angle), P _y (i) is the i-th probability distribution, and w _y (i ) Is a mixing ratio weighted to this probability distribution P _y (i). The probability distribution P _y (i) can be a normal distribution.

FIG. 3 is a view showing an example of an observation error distribution of an angle observation value (one of an elevation observation value and an azimuth observation value). In the observation error distribution shown in FIG. 3, the horizontal axis indicates the angle, and the vertical axis indicates the frequency. This observation error distribution is regarded as a mixed distribution in which three components P _θ (−1), P _θ (0), and P _θ (+1) are mixed. The mixed distribution P _θ can be expressed by the following equation.

Here, w _θ (−1), w _θ (0) and w _θ (1) are mixing ratios of the components P _θ (−1), P _θ (0) and P _θ (+1), respectively. The central component P _θ (0) is a component distributed around the zero angle in the absence of ambiguity in the angular observation, and the component P _θ (−1) on the left is the angular observation Is the component distributed around the angle -Δθ _-1 when assuming that the ambiguity exists in, and the component P _θ (+1) on the right side is the angle when it is assumed that the ambiguity exists in the angle observation value It is a component distributed around [Delta] [theta] ₁ . Expected values which are representative values of these components P _θ (−1), P _θ (0) and P _θ (+1) are −Δθ ₋₁ , 0, + Δθ ₁ respectively. In this case, the ambiguity control unit 31 generates temporary observation values K ₋₁ , K ₀ , and K ₁ of the following three angles from one angle observation value θ _k constituting the observation vector z ⁰ _k .
K ₋₁ = θ _k −Δθ ₋₁ ; K ₀ = θ _k ; K ₁ = θ _k + Δθ ₁

FIG. 4 is a view schematically showing the relationship between the mixture distribution of the angle observation values and the temporary observation values K ₋₁ , K ₀ and K ₁ of the angles with respect to the existing track T1.

Next, FIG. 5 is a diagram showing an example of the observation error distribution of Doppler observation values. In the observation error distribution shown in FIG. 5, the horizontal axis indicates the Doppler velocity, and the vertical axis indicates the frequency. This observation error distribution is regarded as a mixed distribution in which four components P _v (−1), P _v (0), P _v (1) and P _v (2) are mixed. This mixed distribution P _v can be expressed by the following equation.

Here, w _v (-1), w _v (0), w _v (1), w _v (2) are components P _v (-1), P _v (0), P _v (1), respectively. , P _v (2). The component P _v (0) is a component on the assumption that no ambiguity exists in the Doppler observation value, and the other components P _v (−1), P _v (1) and P _v (2) are It is a component group when it is assumed that ambiguity exists in the Doppler observation value. Expected values which are representative values of these components P _v (-1), P _v (0), P _v (1) and P _v (2) are respectively −Δv ₋₁ , 0, + Δv ₁ and + Δv ₂ is there. In this case, the ambiguity control unit 31 generates four Doppler velocity provisional observation values from one Doppler observation value that constitutes the observation vector z ⁰ _k .

Next, FIG. 6 is a view showing an example of the observation error distribution of the distance observation value. In the observation error distribution shown in FIG. 6, the horizontal axis represents distance, and the vertical axis represents frequency. This observation error distribution is regarded as a mixed distribution in which three components P _r (0), P _r (1) and P _r (2) are mixed. The mixed distribution P _r can be expressed by the following equation.

Here, w _r (0), w _r (1) and w _r (2) are mixing ratios of the components P _r (0), P _r (1) and P _r (2), respectively. The component P _r (0) is a component under the assumption that no ambiguity exists in the distance observation value, and the other components P _r (1) and P _r (2) are ambiguity in the distance observation value It is a component group when it is assumed that exists. Expected values which are representative values of these components P _r (0), P _r (1) and P _r (2) are 0 and Δr ₁ and Δr ₂ respectively. In this case, the ambiguity control unit 31 generates temporary observation values of three distances from one distance observation value constituting the observation vector z ⁰ _k .

  As shown in FIG. 1, the likelihood calculation unit 32 includes a correlation determination unit 41, a track likelihood calculation unit 42, and an ambiguity likelihood calculation unit 43.

The correlation determination unit 41 generates predicted observation values of the target distance, Doppler velocity, and angle at the current time t _k based on the existing track (step ST3), and then each of the input temporary detection data is It is determined whether it correlates with the existing track of the target (step ST4). Specifically, the correlation determination unit 41 determines that the temporary detection data is likely to be observed based on the target predicted observation vector at time t _k predicted from the existing track and the residual covariance matrix. Set the gate range which is the area. The predicted observation vector is composed of a set of predicted observation values calculated in step ST3. Correlation determining section 41, the temporary observation vector z ^s _k included in the gate range is determined to correlate with the existing track (i.e. the gate), the temporary observation vector that is not included in the gate range is correlated with the existing track It is determined that it is not (that is, outside the gate). The gate range is expressed, for example, by the following equation (1).

Here, z _k is an arbitrary one of the temporary observation vector z ^s _k , z _{k | k−1} is a predicted observation vector, d is a parameter for determining the size of the gate range, and T is , Is a symbol representing transposition. Also, S _k ^-1 is the inverse of the residual covariance matrix S _k at time t _k .

The predicted observation vector z _{k | k-1} is calculated by the following equation (2).

Here, x _{k | k−1} is a target predicted state vector, which is calculated by equation (8) described later. The predicted state vector x _{k | k−1} may be composed of a set of predicted values of position coordinates x, y, z of the target in the orthogonal coordinate system and predicted values of the velocity Vx, Vy, Yz of the target. Further, f (x _{k | k−1} ) is a function vector representing an observation process at time t _k , and is expressed by the following equation (2a) (P is a positive integer).

Also, the residual covariance matrix S _k is expressed by the following equation (3).

Here, _{Pk | k-1} is a prediction error covariance matrix. Also, H _k is an observation matrix, and H _k ^T is a transposed matrix with respect to the observation matrix H _k . Also, R _k is an observation error covariance matrix. The observation matrix H _k is a Jacobian matrix of multivariable functions f ₁ (x), f ₂ (x),..., F _P (x), and a function representing an observation process as shown in the following equation (3a) The vector f (x) can be generated by partially differentiating around the predicted state vector _{xk | k-1} .

The prediction error covariance matrix P _{k | k−1} is expressed by the following equation (4).

Here, _{Pk-1 | k-1} is a smooth error covariance matrix at sampling time tk _-1 .

For temporary observation vector z ^s _k which is determined not to be included in the gate range, track formed by the combination of the temporary observation vector z ^s _k an existing track is as "there is no possibility of correlation" Excluded from consideration. Things On the other hand, the temporary observation vector z ^s _k which is determined to be included in the gate range, track formed by the combination of the temporary observation vector z ^s _k with the existing track is that "there is a possibility of the correlation" It is considered as a "track candidate". FIG. 7 is a diagram illustrating the relationship between three temporary observation vectors z ⁰ _k , z ¹ _k and z ² _k and the gate range Gt. The temporary observation vector z ⁰ _k is an observation vector corresponding to a shift amount of zero.

Next, the track likelihood calculation unit 42 calculates the track likelihood g indicating the likelihood of the track candidate when assuming that no ambiguity exists in each sensor observation value (step ST5). The track likelihood g is expressed by the following equation (5).

Here, detS _k is a determinant of residual covariance matrix S _k .

Next, the ambiguity likelihood calculation unit 43 calculates the ambiguity likelihood as the correlation likelihood by correcting the track likelihood by using the mixing ratio of the probability distribution corresponding to the track candidate among the mixture distribution. To do (step ST6). Specifically, the ambiguity likelihood calculation unit 43 can calculate the ambiguity likelihood by weighting the mixing probability of the probability distribution corresponding to the track candidate to the track likelihood. For example, the ambiguity likelihood L can be calculated by the following equation (6).

Here, the weighting factor w _θ (p) is a mixing ratio of the p-th probability distribution forming a mixture distribution of angular observation values, and the weighting factor w _v (q) is a mixing distribution q of Doppler speed observation values The weighting factor w _r (u) is the mixing ratio of the u th probability distribution, which is the mixing distribution of distance observation values. The data Ld indicating the ambiguity likelihood L is supplied to the tracking track determination unit 34. Now, assuming that the number of temporary observation values derived from one distance observation value (ie, the number of probability distributions forming a mixed distribution of distance observation values) is Nr, the number of temporary observation values derived from one angle observation value (ie, angle Let Na be the number of probability distributions making up the mixture of observed values, and Nv be the number of tentatively observed values derived from one Doppler observation (ie the number of probability distributions making up the mixture of Doppler observations): The ambiguity likelihood L is generated as many as the number (= Nr × Na × Nv) of combinations of three types of tentatively observed values of distance, angle, and Doppler velocity. Therefore, the number of track candidates formed by the combination of the existing track and the temporary observation vector may be Nr × Na × Nv at maximum.

  Next, the tracking track determination unit 34 determines a target track from track candidates based on the correlation determination result Ad supplied from the correlation determination unit 41 and the data Ld (step ST8). For example, the tracking track determination unit 34 may determine a track by assigning temporary detection data corresponding to the highest ambiguity likelihood among the temporary detection data determined to have a correlation to the existing track. At this time, the tracking track determination unit 34 uses a correlation algorism such as GNN (Global Nearest Neighbor) method, PDA (Probabilistic Data Association) method, JPDA (Joint Probabilistic Data Association) method or JIPDA (Joint Integrated Probabilistic Data Association) method. Track can be determined.

Thereafter, the track specification operation unit 33 executes smoothing processing to calculate track specifications to be used for tracking processing at the next sampling time t _{k + 1} (step ST9). As one of the track specifications, for example, the smooth vector x _{k | k} is calculated by the following equation (7).

Here, x _{k | k−1} is the predicted state vector of the target, z _{k, j} is the temporary observation vector at time t _k , and the subscript j indicates the j-th observation point ing. Also, K _k is an extended Kalman gain matrix at time t _k .

The predicted state vector x _{k | k-1} is calculated by the following equation (8).

The extended Kalman gain matrix K _k is calculated by the following equation (9).

  The smoothing process (step ST9) may be performed before steps ST2 to ST7, or may be performed in parallel with steps ST2 to ST7.

  Next, referring to FIG. 2, when the tracking processing unit 30 continues target tracking (YES in step ST10), the tracking process returns to step ST1. On the other hand, when the tracking processing unit 30 does not continue the target tracking (NO in step ST10), the tracking process ends.

  The above-described signal processing unit 20 and tracking processing unit 30 may be configured by, for example, a computer with a built-in CPU (Central Processing Unit) such as a work station or a mainframe. In addition, the signal processing unit 20 and the tracking processing unit 30 may be realized by a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof. The error distribution data storage unit 21 may be configured using a non-volatile memory such as a ROM (Read Only Memory), and the track related data storage unit 35 may be configured using a memory such as an SDRAM.

  FIG. 8 is a functional block diagram showing a schematic configuration of an information processing apparatus 60 which is a hardware configuration example of the signal processing unit 20, the error distribution data storage unit 21 and the tracking processing unit 30. As shown in FIG. 8, the information processing apparatus 60 includes a processor 61 including a CPU 61C, a random access memory (RAM) 62, a ROM 63, a large capacity recording medium 64, and an input / output interface 65. The processor 61, the RAM 62, the ROM 63, the large capacity recording medium 64, and the input / output interface 65 are connected to one another via a bus 66. Here, the input / output interface 65 corresponds to a circuit (not shown in FIG. 1) that physically connects the sensor 10 and the signal processing unit 20 shown in FIG. The large capacity recording medium 64 constitutes the error distribution data storage unit 21 and the track related data storage unit 35 shown in FIG. The processor 61 loads a software program from the ROM 62 or the large-capacity recording medium 64 and can operate according to the software program to implement the functions of the signal processing unit 20 and the tracking processing unit 30. The large-capacity recording medium 64 may be, for example, a hard disk (magnetic disk), an optical disk, or a flash memory.

  As described above, the radar apparatus 1A according to the first embodiment shifts the sensor observation values included in the detection data by the representative value of the plurality of probability distributions in the mixture distribution, thereby performing a plurality of provisional detections on the respective detection data. Generate data. A track candidate is generated based on the plurality of temporary detection data, and the track of the moving target can be determined from among the track candidates. Therefore, even in an environment where ambiguities occur in sensor observation values, the target track can be determined efficiently and with high accuracy regardless of the type of the sensor observation values. This reduces the occurrence of erroneous navigation and enables early establishment of the navigation. For example, in the prior art, even in a situation where a target may be missed due to the influence of ambiguity, a plurality of temporary detection data are generated in this embodiment, so that there is an advantage that the probability of false detection is reduced. . For this reason, it is possible to establish a wake early.

Second Embodiment
Next, Embodiment 2 according to the present invention will be described. FIG. 9 is a block diagram showing a schematic configuration of a radar device 1A including the target tracking device of the second embodiment.

  Like the radar device 1 according to the first embodiment, the radar device 1A includes the sensor 10, the signal processing unit 20, and the error distribution data storage unit 21, and further includes the tracking processing unit 30A. The configuration of the tracking processing unit 30A of the present embodiment is the same as the configuration of the tracking processing unit 30 of the first embodiment except for the tracking track determination unit 34A. The target tracking device according to the present embodiment includes a signal processing unit 20, an error distribution data storage unit 21, and a tracking processing unit 30A.

The tracking track determination unit 34A of the present embodiment includes a track search unit 45 and a hypothesis restriction unit 46. The track search unit 45 can determine the track of the moving target while holding a plurality of hypotheses in accordance with a known MHT (Multiple Hypothesis Tracking) method. Here, the hypothesis is a set of tracks, and each hypothesis is consistent with each other (that is, a track that does not share the same temporary detection data that is a set of sensor observation values at the same observation point at the same time t _k ) Configured by choosing For the known MHT, for example, the method disclosed in Non-Patent Document 1 may be used. Also, the hypothesis limiting unit 46 rejects hypotheses in which a plurality of tracks are correlated with two or more pieces of temporary detection data generated from the same detection data. This allows the number of hypotheses to be limited so that existing tracks do not correlate with two or more pieces of temporary detection data generated from the same detection data.

FIG. 10 illustrates the relationship between gate ranges Gt1 and Gt2 respectively corresponding to two existing tracks T1 and T2 and a plurality of temporary observation vectors. In the example of FIG. 10, from one observation vector (detection data) z ⁰ _k (1), three temporary observation vectors (temporary detection data) z ⁰ _k (1), z ¹ _k (1), z ² _k (1) is generated, and three other temporary observation vectors (temporary detection data) z ⁰ _k (2), z ¹ _k (2) from the other one observation vector (detection data) z ⁰ _k (2) , Z ² _k (2) are generated. Here, the temporary observation vector z ⁰ _k (1) is an observation vector corresponding to a shift amount of zero, and the temporary observation vector z ⁰ _k (1) is an observation vector corresponding to a shift amount of zero. Since the temporary observation vector z ⁰ _k (2), z ¹ _k (1) does not enter either of the gate ranges Gt1 and Gt2 and is outside the gate, the correlation determination unit 41 determines that the temporary observation vector z ⁰ _k (2 , Z ¹ _k (1) is determined not to be correlated with the existing track T1, T2. The hypothesis limiting unit 46 correlates the existing track T1 with the temporary observation vector z ² _k (2) and the existing track T 2 with the temporary observation vector z ¹ _k (2), that is, from the same detection data It rejects hypotheses in the case where a plurality of existing hypotheses T1 and T2 are correlated with the generated two or more tentative detection data. This is based on the principle that at most one existing track is correlated with one piece of detection data.

FIG. 11 is a flowchart schematically showing an example of the procedure of the tracking process of the second embodiment. Steps ST1 to ST6 in FIG. 11 are the same as steps ST1 to ST6 in FIG. In this embodiment, track search unit 45, in step ST7, in accordance with the following equation (10), the normalized hypothesis likelihood (reliability) is calculated beta _i.

Here, M is an integer value indicating the number of hypotheses at the current time. Further, γ _i is the likelihood of a non-normalized hypothesis and is given by the following equation (11).

Here, β _{P (i)} is the reliability of the pre-update hypothesis at the previous time, and N _TRK is the number of existing tracks in the pre-update hypothesis. Further, _NDT is an integer value indicating the number of existing tracks in the pre-update hypothesis and the number of tracks correlated with temporary detection data of the latest observation time. Furthermore, N _FT is the number of unnecessary signals included in the hypothesis, N _NT is the number of new tracks included in the hypothesis, and L is the ambiguity likelihood of the above equation (6).

  Next, in step ST11, the track searching unit 45 and the hypothesis limiting unit 46 determine the tracking track based on a plurality of hypotheses according to the MHT method. Then, the track specification operation unit 33 performs a smoothing process on each of the plurality of hypotheses to calculate the track specification (step ST12). When the tracking processing unit 30A continues target tracking (YES in step ST13), the tracking process returns to step ST1. On the other hand, when the tracking processing unit 30A does not continue the target tracking (NO in step ST13), the tracking process of the present embodiment ends.

  The tracking track determination unit 34A may execute tracking processing using a known Lagrange relaxation method in order to solve a three-dimensional correlation problem for the existing track, detection data, and temporary detection data. .

  Further, as in the first embodiment, the signal processing unit 20 and the tracking processing unit 30A of the present embodiment may be configured by a computer with a built-in CPU, such as a workstation or a mainframe. Also, the signal processing unit 20 and the tracking processing unit 30A may be realized by a DSP, an ASIC, an FPGA, or a combination thereof.

  As described above, in the present embodiment, it is possible to improve tracking correlation performance by performing tracking processing while holding a plurality of hypotheses. In addition, by limiting the possibility that two or more wakes are correlated with one piece of detection data by the hypothesis limiting unit 46, it is expected to effectively suppress the occurrence of a false wake.

Third Embodiment
Next, a third embodiment of the present invention will be described. FIG. 12 is a block diagram showing a schematic configuration of a radar device 1B including the target tracking device of the third embodiment.

  This radar device 1B includes the sensor 10, the signal processing unit 20, and the error distribution data storage unit 21 similarly to the radar device 1 of the first embodiment, and further includes the tracking processing unit 30B. The configuration of the tracking processing unit 30B of the present embodiment is the same as the configuration of the tracking processing unit 30 of the first embodiment except for the tracking track determination unit 34B. As shown in FIG. 12, the tracking track determination unit 34B of the present embodiment includes a first correlation search unit 48, a second correlation search unit 49, and a hypothesis database 50. The target tracking device according to the present embodiment includes a signal processing unit 20, an error distribution data storage unit 21, and a tracking processing unit 30B.

  FIG. 13 is a diagram schematically illustrating an example of the procedure of the tracking process according to the third embodiment. Steps ST1 to ST7 in FIG. 13 are the same as steps ST1 to ST7 in FIG. The present embodiment is characterized in that a two-step process is performed in step ST21. FIG. 14 is a flowchart schematically showing an example of the procedure of the two-step process of step ST21.

Referring to FIG. 14, the first correlation search unit 48 is a likelihood that is a two-dimensional array of the highest log likelihood for temporary detection data among the logarithm values of hypothesis likelihood (hereinafter referred to as “log likelihood”). A matrix is generated (step ST31). Here, the hypothesis likelihood is either the likelihood (reliability) of the normalized hypothesis given by the above equation (10) or the likelihood of the unnormalized hypothesis given by the above equation (11) It is good. Since the log likelihood is an amount given by combining the existing track, the detection data, and the temporary detection data derived from the detection data, the hypothesis likelihood is said to be an amount specified in the three-dimensional space it can. Therefore, when the existing track number is N ₁ , the number indicating detection data is N ₂ , and the number indicating temporary detection data is N ₃ , log likelihood is represented by B (N ₁ , N ₂ , N ₃ ) Can. In step ST31, the first correlation search unit 48 selects, for each set of (N ₁ , N ₂ ), the highest hypothesis likelihood from among the different log likelihoods of temporary detection data. By this, it is possible to obtain a likelihood matrix Γ = {Γ (N ₁ , N ₂ )} (Γ (N ₁ , N ₂ ) is a matrix element of the N _1st row and the N _2nd column) of a two-dimensional array it can.

Next, the first correlation search unit 48 searches the top N (N is a positive integer) hypotheses based on the likelihood matrix 用 い using a known N best solution search method (step ST32). For the N-best solution search method, for example, Non-patent documents: Yasushi Kojima, System Justice, Shingo Shindo, Yoshio Koshiba, "Speeding up wake-type MHT with N-best solution search algorithm (1)", 2001 Electronic information communication General Conference, B-2-32, Sep. 2001, it is disclosed. The likelihood matrix Γ is a matrix including, as matrix elements Γ (N ₁ , N ₂ ), log likelihoods in the case of assigning the N ₁ -th temporary detection data to the N ₂ -th existing track. In the N best solution search method, the sum of a plurality of matrix elements is used as a score, and the top N hypotheses are determined in descending order of the score. Here, one hypothesis corresponds to one combination of a plurality of matrix elements.

  Then, the first correlation search unit 48 stores data of the top N hypotheses in the hypothesis database 50 (step ST33). Thereby, the correlation search process of the first stage is completed.

Next, the second correlation search unit 49 executes a second stage correlation search process. That is, the second correlation search unit 49 focuses on one of the top N hypotheses (step ST34), and searches for a hypothesis adjacent to the focused hypothesis (hereinafter, also referred to as “adjacent hypothesis”) (step S34). ST 35). Specifically, while the combination of the existing track and the detection data in the target hypothesis is fixed, the second correlation search unit 49 searches for a plurality of temporary detection data derived from the detection data, and the score of the combination is obtained. Search hypotheses that maximize (correlation), that is, adjacent hypotheses. For example, in the case of FIG. 10 described above, the existing track T1 and the temporary observation vector (temporary detection data) z ⁰ _k (1) are mutually correlated, and the existing track T2 and the temporary observation vector (temporary detection data) z ¹ _k Assuming that the hypothesis that (2) is correlated with each other is the attention hypothesis, the existing track T1 and the temporary observation vector (temporary detection data) z ² _k (1) are mutually correlated, and the existing track T2 and the temporary observation The hypothesis that the vectors (provisional detection data) z ¹ _k (2) correlate with each other can be the adjacent hypotheses.

  Then, if the score of the adjacent hypothesis is less than or equal to the score of the focused hypothesis (NO in step ST36), the second correlation search unit 49 focuses on the next hypothesis in the hypothesis database 50 (step ST39) and processes step ST35 and subsequent steps. Run repeatedly.

  On the other hand, when the score of the adjacent hypothesis is larger than the score of the focused hypothesis (YES in step ST36), the second correlation search unit 49 updates the hypothesis database 50 by replacing the focused hypothesis with the adjacent hypothesis (step ST37). .

  In the case where the score of the adjacent hypothesis is less than or equal to the score of the focused hypothesis (NO in step ST36) or in the case where the hypothesis database 50 is updated (step ST37) (The YES of step ST38), the 2nd correlation search part 49 performs the process after step ST35 paying attention to the following hypothesis. On the other hand, when the processing is completed for all hypotheses (YES in step ST38), the second correlation search unit 49 ends the above two-stage processing, and then shifts the processing to step ST22 in FIG.

Referring to FIG. 13, in step ST22, the track data computing unit 33 executes the same smoothing process as step ST9 (FIG. 2) of the above-mentioned first embodiment, and performs tracking process at the next sampling time t _{k + 1.} The track specifications to be used are calculated (step ST22). When the tracking processing unit 30B continues the target tracking (YES in step ST23), the tracking process returns to step ST1. On the other hand, when the tracking processing unit 30B does not continue the target tracking (NO in step ST23), the tracking process ends.

  As described above, in the third embodiment, two-stage correlation processing is performed using a known N-best solution search method, so a method of solving a three-dimensional correlation problem between existing tracks, detection data and temporary detection data The computational load can be reduced compared to.

  As mentioned above, although various embodiments according to the present invention have been described with reference to the drawings, these embodiments are merely examples of the present invention, and various embodiments other than these embodiments can be adopted. For example, in the target tracking devices according to the first to third embodiments, an active sensor is used as the sensor 10. However, the present invention is not limited to this, and a passive sensor such as an infrared sensor may be used.

  In the scope of the present invention, free combinations of Embodiments 1, 2 and 3 above, deformation of any component of each embodiment, or omission of any component of each embodiment is possible. is there.

  1, 1A, 1B target tracking device, 10 sensor, 11 transmitting circuit, 12 transmitting antenna unit, 13 receiving antenna unit, 14 receiving circuit, 20 signal processing unit, 21 data storage unit, 30 tracking processing unit, 30A tracking processing unit, 31 ambiguity control unit (temporary detection data generation unit), 32 likelihood operation unit, 33 track specification unit, 34 tracking track determination unit, 34A tracking track determination unit, 35 track related data storage unit, 41 correlation determination unit , 42 track likelihood calculation unit, 43 ambiguity likelihood calculation unit, 45 track search unit, 46 hypothesis limiting unit, 48 first correlation search unit, 49 second correlation search unit, 50 hypothesis database.

Claims (18)

A target tracking device that performs tracking processing on a moving target using a sensor that observes the moving target,
A signal processing unit that generates detection data indicating the position of the movement target based on the output of the sensor;
A temporary detection data generation unit that generates a plurality of temporary detection data from the detection data;
A likelihood calculating unit that calculates a correlation likelihood indicating the likelihood of a track candidate formed by combining the temporary track observation value and the observation value forming the existing track obtained in the past tracking processing;
A tracking track determination unit that determines the track of the moving target from among the track candidates using the correlation likelihood;
An error distribution data storage unit storing error distribution data for determining observation error distribution of each sensor observation value included in the detection data ;
The observation error distribution is a mixed distribution in which a plurality of probability distributions are mixed,
The temporary detection data generation unit shifts the plurality of temporary detection data from the detection data by shifting each sensor observation value included in the detection data by the representative value of the plurality of probability distributions based on the error distribution data. Generate,
A target tracking device characterized in that.   The target tracking device according to claim 1, wherein each of the plurality of probability distributions is a normal distribution, and the representative value is an expected value of the normal distribution.   The target tracking device according to claim 1 or 2, wherein the detection data includes a plurality of types of sensor observation values, and the plurality of temporary detection data includes each of the plurality of types of sensor observation values. A target tracking device generated by shifting only the representative value of the probability distribution of. The target tracking device according to claim 3, wherein
The likelihood computing unit
A track likelihood calculation unit that calculates a track likelihood indicating the likelihood of the track candidate when assuming that no ambiguity exists in each of the sensor observation values;
And an ambiguity likelihood calculating unit that calculates the correlation likelihood by correcting the track likelihood using the mixing ratio of the probability distribution corresponding to the track candidate among the plurality of probability distributions. A target tracking device characterized in that. The target tracking device according to claim 4, wherein
The case where it is assumed that no ambiguity exists in each sensor observation value is a case where it is assumed that the error of each sensor observation value is represented by a single probability distribution,
The track likelihood calculation unit calculates the track likelihood using the single probability distribution;
The ambiguity likelihood calculation unit calculates the correlation likelihood by weighting a mixing ratio of the probability distribution corresponding to the track candidate to the track likelihood.
A target tracking device characterized in that.   The target tracking device according to claim 5, wherein the single probability distribution is a normal distribution.   The target tracking device according to any one of claims 1 to 6, wherein the detection data is selected from a distance observation value, an angle observation value and a Doppler velocity observation value of the moving target. A target tracking device comprising at least one kind of observation value. The target tracking device according to claim 4, wherein
The likelihood calculation unit further includes a correlation determination unit that determines whether each of the plurality of temporary detection data correlates with the existing track of the movement target,
The track likelihood calculation unit is a new formed by combining the previously observed values forming the existing track and the temporary detection data determined to be correlated with the existing track among the plurality of temporary detection data. A target tracking device characterized by using only a track as the track candidate. The target tracking device according to claim 8, wherein
It further comprises a track specification operation unit that predicts the observed value of the moving target based on the past observed value forming the existing track;
The correlation determination unit determines whether each of the plurality of temporary detection data is correlated with the existing track based on a residual between the predicted observation value and each of the plurality of temporary detection data. A target tracking device characterized by determining. The target tracking device according to any one of claims 1 to 9, wherein
The tracking track determination unit holds a plurality of hypotheses formed based on a combination of a plurality of existing tracks obtained by past tracking processing and the plurality of temporary detection data according to a MHT (Multiple Hypothesis Tracking) method. A target tracking device characterized in that the track of the moving target is determined.   The target tracking device according to any one of claims 1 to 9, wherein the tracking track determination unit is configured to obtain a plurality of existing tracking processes obtained in the past according to a Multiple Hypothesis Tracking (MHT) method. Determining a track of the moving target while holding a plurality of hypotheses formed on the basis of a combination of the plurality of temporary detection data and the plurality of temporary detection data, and the two or more temporary detections generated from the same corresponding detection data A target tracking device characterized in that the number of hypotheses is limited so that the existing track is not correlated with data. The target tracking device according to any one of claims 1 to 9, wherein
The tracking track determination unit
The top N (N is a positive integer) hypotheses are searched from among a plurality of hypotheses formed based on a combination of a plurality of existing tracks obtained by past tracking processing and the plurality of temporary detection data 1 correlation search unit,
Focusing on each hypothesis among the top N hypotheses, searching for a plurality of temporary detection data derived from the detection data while fixing the combination of the existing track and the detection data in each hypothesis And a second correlation search unit for finding preliminary detection data that maximizes the correlation of the combination.
The target tracking device characterized in that the second correlation search unit replaces each of the hypotheses with an adjacent hypothesis newly formed based on the found temporary detection data. A sensor that observes the moving target,
A radar apparatus comprising the target tracking device according to any one of claims 1 to 12.   The radar apparatus according to claim 13, wherein the sensor transmits a pulse-like observation wave toward the movement target, and receives the reflection wave reflected by the movement target to receive the movement target. A radar apparatus characterized in that it is an active sensor for observation.   The radar apparatus according to claim 14, wherein the sensor changes the transmission frequency of the observation wave for each pulse hit.   The radar apparatus according to claim 14, wherein the sensor changes the transmission frequency of the observation wave for each detection so as to change the average value of the angular error distribution for each detection.   The radar apparatus according to claim 13 or 14, wherein the sensor changes a pulse repetition frequency for each pulse hit.   The radar apparatus according to claim 13, wherein the sensor changes a pulse repetition frequency for each detection.

Images