JP6773606B2 - Radar device and its radar signal processing method - Google Patents

Description

The present embodiment relates to a radar device and a radar signal processing method thereof.

In the conventional radar method, when detecting a target with a low RCS (Radar Cross Section), the threshold of CFAR (Constant False Alarm Rate) is reduced to improve the sensitivity. However, if the threshold is reduced, false detections will increase. Also, when observing a low-altitude target, false positives increase due to the influence of clutter. As described above, when the number of false positives increases, a large number of verification beams (dedicated verify beams) for discriminating whether or not the target is obtained after the detection are required in order to establish a track for tracking. As a result, the constraint of scheduling (beam management) of the transmission / reception beam for efficiently observing the target within a predetermined time becomes large, and problems such as not being able to observe all the targets occur.

CFAR processing, Yoshida,'Revised radar technology', Institute of Electronics, Information and Communication Engineers, pp.87-89 (1996) Cluster Analysis, Sebastian Raschka, Python,'Machine Learning Programming', Impress, pp.297-319 (2016) Least squares estimation, Nakamizo,'Signal analysis and system identification processing', Corona Publishing Co., Ltd., pp.10-17 (1987) DBSCAN (Density-based Spatial Clustering of Applications with Noise) Sebastian Raschka, Python Machine Learning Programming, Impress, pp.319-323 (2016) Correlation Tracking, Yoshida,'Revised Radar Technology', Institute of Electronics, Information and Communication Engineers, pp.254-259 (1996) Neural network, Okaya,'Deep learning', Kodansha, pp.7-26 (2014) Convolutional Neural Network, Okaya,'Deep Learning', Kodansha, pp.79-110 (2014) Recursive Neural Network, Okaya,'Deep Learning', Kodansha, pp.111-130 (2014)

As described above, in the conventional radar method, if the CFAR threshold is reduced for high sensitivity or the false detection due to the influence of the clutter in the case of low-altitude target observation increases, it is discriminated whether it is the target or not after the detection. A large number of test beams are required for this purpose, and the restrictions on beam management become large, causing problems such as the inability to observe all targets.

The present embodiment has been made in view of the above problems, and an object of the present invention is to provide a radar device capable of suppressing erroneous detection and extracting only a target with high accuracy, and a radar signal processing method thereof.

In order to solve the above problem, according to the present embodiment, when N (N ≧ 2) frames are observed with the unit for observing the radar observation range M (M ≧ 2) times as one frame, M for each frame is observed. arranged batch of observations in a three-dimensional or two-dimensional coordinates, extracts the observations arranged on the coordinate, the cluster of the target candidate I row cluster analysis due to the difference of the density distribution of the erroneous detection target coordinate axis Then, the observed value or the representative value in the cluster for N frames is output from the target candidate cluster .

That is, in the radar device according to the present embodiment, false detection can be suppressed and only the target can be extracted by performing cluster analysis based on the difference between the target and the false detection density distribution on the coordinate axes.

The block diagram which shows the schematic structure of the radar apparatus which concerns on 1st Embodiment. In the first embodiment, the figure which shows the state of sliding for each 3D data of a plurality of scans and performing cluster analysis for each divided frame. In the first embodiment, the figure which shows the state of the false detection suppression and the target detection by the cluster distribution utilizing the difference of the density distribution. The block diagram which shows the schematic structure of the radar apparatus which concerns on 2nd Embodiment. The figure which shows the state of the correlation tracking in the 2nd Embodiment. In the second embodiment, it is a figure which shows a mode in which a predetermined gate is set around a continuous smooth value for every frame. In the second embodiment, it is a figure which shows how the lack is suppressed when there is a missing observation value at the time of cluster detection. The block diagram which shows the schematic structure of the radar apparatus which concerns on 3rd Embodiment. The figure which shows the state of performing the cluster analysis by changing the parameter of the cluster analysis again in the 3rd Embodiment. The block diagram which shows the schematic structure of the radar apparatus which concerns on 4th Embodiment. The figure which shows the mode of setting the target range by the interpolation curve in 4th Embodiment. The block diagram which shows the schematic structure of the radar apparatus which concerns on 5th Embodiment. The figure for demonstrating the DBSCAN method in 5th Embodiment. The block diagram which shows the schematic structure of the radar apparatus which concerns on 6th Embodiment. The figure for demonstrating the neural network scheme in the sixth embodiment. The figure for demonstrating the unit used for the neural network in 6th Embodiment. In the sixth embodiment, a diagram showing an example of plotting at three-dimensional coordinate (X, Y, Z) points based on the coordinates of the target candidate extracted by cluster distribution detection and target candidate extraction as input data. ..

Hereinafter, embodiments will be described with reference to the drawings. In the description of each embodiment, the same parts are designated by the same reference numerals, and duplicate description will be omitted.

(First Embodiment) -As a cluster distribution radar device, in the case of pulse modulation or continuous wave, and as pulse modulation, LPRF (Pulse Repetition Interval) has ambiguity in speed according to the classification of PRI (Pulse Repetition Interval). Low Pulse Repetition Frequency (Low Pulse Repetition Frequency), MPRF (Medium Pulse Repetition Frequency) with ambiguity in distance and speed, HPRF (High Pulse Repetition Frequency) with ambiguity in distance It can be widely applied to the case of frequency), but here, for the sake of simplicity, the case of pulse modulation of LPRF will be described.

The radar device according to the first embodiment will be described with reference to FIGS. 1 to 3. FIG. 1 is a block diagram showing the schematic configuration, FIG. 2 is a diagram showing a state in which cluster analysis is performed for each divided frame by sliding each three-dimensional data of a plurality of scans, and FIG. 3 is a cluster distribution using the difference in density distribution. It is a figure which shows the state of the false detection suppression and the target detection.

As shown in FIG. 1, the radar device according to the present embodiment constitutes a monopulse antenna by a plurality of antenna elements, and transmits a transmission signal through an antenna 1 that forms a transmission / reception beam of Σ and Δ and a target. The distance and angle of the target are observed from the transmitter / receiver 2 that receives the reflected signal from the transmitter / receiver 2 and the receiver signal obtained by the transmitter / receiver 2, the three-dimensional position coordinates are calculated, and the detection process based on the cluster distribution is performed to obtain the target. It includes a signal processor 3 that outputs observed values.

The transmitter / receiver 2 includes a transmitter / receiver 21 and a modulation control unit 22. The transmission / reception unit 21 transmits a monopulse pulse modulation wave generated by the LPRF pulse modulation of the modulation control unit 22 through the antenna 1, receives the reflected wave of the transmission output, and receives signals of the Σ beam and the Δ beam (the reception signal of the Σ beam and the Δ beam. Hereinafter referred to as Σ signal and Δ signal) are output.

The signal processor 3 includes a Σ observation processing unit 31, a CFAR processing unit 32, a distance measuring unit 33, a Δ observation processing unit 34, a cell extraction unit 35, an angle measuring unit 36, a three-dimensional position coordinate output unit 37, and a cluster analysis unit. 38 is provided.
The Σ observation processing unit 31 inputs a Σ reception signal by a monopulse beam, performs signal processing such as pulse compression and FFT according to the transmission waveform using CPI data, and obtains range-doppler (RD) data. ..
The CFAR processing unit 32 uses the RD data of the Σ signal generated by the monopulse output of the antenna 1 to set a predetermined threshold by CFAR and detect the cell at the reflection point.
The distance measuring unit 33 measures the target distance by converting the time axis into the distance axis for the cells of the reflection points obtained by CFAR (distance measuring).

The Δ observation processing unit 34 inputs a Δ reception signal by a monopulse beam, and similarly to the Σ observation processing unit 31, performs signal processing such as pulse compression and FFT according to the transmission waveform using CPI data. Obtain range-doppler (RD) data.
The cell extraction unit 35 extracts the same cell as the Σ signal detection cell by using the processing result of the CFAR processing unit 33.
The angle measuring unit 36 outputs the angles of AZ and EL for each extracted cell by performing monopulse angle measuring processing with the target distance obtained by the distance measuring unit 33 for the cells extracted by the cell extraction unit 35. To do.

The three-dimensional position coordinate output unit 37 calculates and outputs the three-dimensional position coordinates of the target from the distance and angle of the target obtained through the angle measuring unit 36.

The cluster analysis unit 38 performs cluster analysis using the density difference that the cell density near the target having a large amplitude is high and is small near the false detection, and when the number of reflection points is not sufficient in one CPI process, The observed value obtained by adding the detection cells of multiple CPI times is output.

In the above configuration, the radar signal processing method according to the present embodiment will be described below.

First, the pulse-modulated wave is transmitted and received through the antenna 1 (2), and signal processing such as pulse compression and FFT is performed according to the transmitted waveform using the output of the Σ signal by the monopulse antenna beam to obtain a range-doppler signal. (31), a predetermined threshold is set by CFAR (Non-Patent Document 1), a target is detected (32), and a target distance is output by the detected range cell (33).

Further, the same signal processing is performed on the Δ signal of the monopulse antenna beam (34), the same cell as the detection cell of the Σ signal is extracted (35), and the angles of AZ and EL by the monopulse angle measurement processing are output. (36). If the distance and angle are known, the three-dimensional position coordinates can be output (37).

Cluster analysis of this three-dimensional coordinate data is performed (Non-Patent Document 2) (38). The cluster analysis is a method of grouping and classifying input values according to a predetermined condition. In this embodiment, it is used to discriminate between a target and false detection due to clutter or noise and extract only the target. Discrimination between a target and a false positive utilizes, for example, that the density of reflection points near the target is higher than in the case of false positives.

When observing multiple scans, as shown in FIG. 2, each of the three-dimensional data of scans M, M + 1, ..., M + m is divided into sliding frames N, N + 1, ..., And cluster analysis is performed for each frame. .. FIG. 3 shows the suppression of false detection and the target detection by the cluster distribution using the difference in the density distribution. That is, as shown in FIG. 3A, erroneous detection for a plurality of scans and cluster analysis based on the density difference are performed for the targets, and as shown in FIG. 3B, clusters containing only high-density targets are extracted.

When the target cluster is extracted, the observed value in the cluster is output as it is, or the center of gravity value by the following equation using the mean value and the amplitude of the three-dimensional coordinates in the cluster as the representative value is output as the observed value.

As described above, the radar device according to the present embodiment has M for each frame when observing N (N ≧ 2) frames with the unit for observing a predetermined observation range M (M ≧ 2) times as one frame. The observed values of the times are arranged in three-dimensional or two-dimensional coordinates, cluster analysis is performed, the clusters of the target candidates are extracted, and the observed values or representative values in the cluster for N frames are output. That is, the false detection can be suppressed and only the target can be extracted by the cluster analysis based on the difference between the target and the false detection density distribution on the three-dimensional coordinate axes.

(Second Embodiment) -Suppression of False Detection by Correlation Tracking The radar device according to the second embodiment will be described with reference to FIGS. 4 to 7. FIG. 4 shows a schematic configuration thereof, FIG. 5 shows a state of correlation tracking, FIG. 6 shows a state of setting a predetermined gate centered on a continuous smooth value for each frame, and FIG. 7 shows a missing state at the time of cluster detection. It shows how to suppress the omission when there is an observed value.

In the configuration of the first embodiment, false positives, which are observed values other than the extracted clusters, may remain. In this embodiment, the countermeasures will be described.

The radar device according to the present embodiment includes a correlation tracking unit 39 as shown in FIG. 4, detects a cluster by the same method as in the first embodiment, and then performs a correlation tracking process (non-patent) based on a representative value of the cluster. Document 5) is performed. In the correlation tracking, as shown in FIG. 5, first, the representative value of the first scan is set as the predicted value, the representative value closest to the predicted value is selected from the representative values in the predetermined gate, and the next predicted value is selected. Is calculated. A predetermined gate is set as shown in FIG. 6 centering on a continuous smooth value for each frame, and the observation value inside the gate is extracted while suppressing (deleting) the observation value outside the gate as a false detection. In this case, as shown in FIG. 7 (a), even if there are missing observation values at the time of cluster detection, by re-extracting the observation values in the gate of correlation tracking as shown in FIG. 7 (b), The omission can be suppressed.

As described above, the radar device according to the second embodiment suppresses erroneous detection by performing correlation tracking processing using clusters of target candidates for N (N ≧ 1) frames. By performing correlation tracking processing using the extracted clusters in this way, residual false positives can be suppressed.

(Third Embodiment) -Multiple times of cluster analysis The radar device according to the third embodiment will be described with reference to FIGS. 8 and 9. FIG. 8 shows the schematic configuration, and FIG. 9 shows how the cluster analysis is performed by changing the parameters of the cluster analysis again.

In the first and second embodiments, false positives may remain. The radar device according to the present embodiment is provided as a countermeasure, and as shown in FIG. 8, the cluster analysis unit 38a performs a plurality of cluster analyzes with different parameters. That is, in a state where the distribution as shown in FIG. 9A is obtained, first, cluster detection is performed by the same method as in the first or second embodiment. As a result, when there is a residual false detection as shown in FIG. 9B, the density distribution of the observed values changes. Therefore, in the present embodiment, as shown in FIGS. 9C and 9D, the cluster analysis is sequentially performed by changing the parameters. The parameters include, for example, in the case of the DBSCAN method (Non-Patent Document 4) described in the fifth embodiment, the gate radius, the score of the observed value in the gate, and the like. By repeating this cluster analysis a predetermined number of times (≧ 2), erroneous detection can be suppressed. At this time, although it is assumed that the observed value of the target is suppressed, the observed value of the missing target can be restored by using the method described in the second or fourth embodiment.

As described above, the radar device according to the present embodiment performs the cluster analysis process P (P ≧ 2) a plurality of times to suppress erroneous detection. This makes it possible to suppress residual false positives that cannot be suppressed by a single cluster analysis.

(Fourth Embodiment) -Target extraction (fitting) by cluster interpolation
The radar device according to the fourth embodiment will be described with reference to FIGS. 10 and 11. FIG. 10 shows a schematic configuration thereof, and FIG. 11 shows how a target range is set by an interpolation curve.

In the first to third embodiments, a method for suppressing false positives has been described. At this time, the target cluster may be missing. Although the method of suppressing the influence of omission by the correlation tracking process of the second embodiment has been described, other measures will be described in the present embodiment.

As shown in FIG. 10, the radar device according to the present embodiment has a configuration in which the target interpolation unit 3A is arranged between the cluster analysis unit 31a and the correlation tracking unit 39 of the second embodiment. That is, in the present embodiment, as shown in FIGS. 11A to 11D, after the clusters are extracted, the target candidate range is extracted using the observed values in the L (L ≧ 1) frame. At this time, an interpolation curve is extracted using a least squares curve (Non-Patent Document 3) or the like.

Here, the calculation method of the interpolation function by the least squares method will be described. When the explanatory variable is represented by the vector x and the objective variable is represented by the vector y, the following equation is obtained.

To calculate the least squares curve of the data in Eq. (2), it corresponds to calculating the parameter θ that minimizes the following equation.

When f is a polynomial, θ is a coefficient of each term as shown in the following equation.

The θ (ap in the case of a polynomial) in Eq. (3) can generally be calculated as a nonlinear estimation problem under the conditions of the following Eq. (Non-Patent Document 3).

Equation (5) can generally be calculated by repeating the following equation.

A region having a predetermined width shown in FIG. 11D is set around the fitting curve calculated by the above processing, and all the observed values in the region are extracted. As a result, the observed value of the target that was missing at the time of cluster detection can be restored.

In the method of this embodiment, the data of (xq, yq) for the L frame is used when calculating the interpolation curve, but while sliding the data of the L frame, 1 to L, 2 to L + 1, ... If the data is selected as such, a continuous interpolation curve can be obtained after the data after the first L frame is acquired, and the missing data can be corrected by the cluster analysis. The target interpolation value extracted in this way is output as an observed value (X, Y, Z) as it is, or the observed value is output after the correlation tracking process (39) in order to further suppress the false detection.

As described above, the radar device according to the fourth embodiment extracts the target candidates for N frames, calculates the fitting curve using the observed values, and extracts the observed values in the region having a predetermined width. Then, the missing target observations in the area of the target cluster are interpolated. In this way, even if the extraction of the target observed value is missing due to the cluster analysis, the target cluster range is interpolated by the interpolation process, so that the missing target observed value can be suppressed.

(Fifth Embodiment) -DBSCAN Method The radar device according to the fifth embodiment will be described with reference to FIGS. 12 and 13. FIG. 12 shows a schematic configuration thereof, and FIG. 13 shows processing by the DBSCAN method.

In this embodiment, a method using DBSCAN (Non-Patent Document 4) as a cluster analysis method will be described.

As shown in FIG. 12, the radar device according to the present embodiment adopts the DBSCAN method in the cluster analysis unit 38b, and as shown in FIG. 13, the circle range of the radius ε and the number of observed values MinPts in this circle range are measured. Set and classify the clusters according to the difference in density. Multiple clusters can be generated for each cluster that satisfies the set radius ε and the number of observed values MinPts, and the others can be classified as noise.

Specifically, points with at least the set number of observed values MinPts adjacent points within the set radius ε are regarded as core points, and points within the radius ε that have adjacent points less than MinPts are regarded as border points. Is done. Points that are neither core points nor border points are called noise points. Thereby, the reflection point (core point and border point) and the false detection (noise point) due to the target can be discriminated (Non-Patent Document 4).

As described above, the radar device according to the present embodiment uses DBSCAN as the cluster analysis method. As a result, cluster analysis based on the difference between the target and the density distribution of false positives can be efficiently performed.

(Sixth Embodiment) -Neural Network The radar device according to the sixth embodiment will be described with reference to FIGS. 14 to 17. FIG. 14 shows the schematic configuration thereof, FIG. 15 shows the concept of the neural network method, FIG. 16 shows the configuration of the unit used in the neural network, and FIG. 17 shows the configuration of the unit used in the neural network, and FIG. An example of plotting at three-dimensional coordinate (X, Y, Z) points based on the extracted target candidate coordinates is shown.

In this embodiment, as a cluster detection method, attention is paid to the density difference between the target and the false detection. Further, in order to enhance the ability to discriminate between the target and the false detection, a method of identifying by using a neural network (Non-Patent Document 6) can be applied by inputting coordinate data in a predetermined range centered on the selected target cluster as an input. As a specific example of the neural network (NN), a convolutional neural network (Non-Patent Document 6), a recursive neural network (Non-Patent Document 7), and the like can be used. In this case, in addition to the density difference between the target and the false detection, the difference in the characteristics of the spatial spread can be extracted as the feature amount and discriminated.

As shown in FIG. 14, the radar device according to the present embodiment has a configuration in which the target candidate extraction filter 3B is arranged between the target interpolation unit 3A and the correlation tracking unit 39.

The target candidate extraction filter 3B includes a three-dimensional data generation unit 3B1, a column vector conversion unit 3B2, an NN (neural network) processing unit 3B3, and a filter 3B4, and is extracted from the detection result by the cluster distribution in the target interpolation unit 3A. Based on the target candidate, the neural network NN identifies whether it is a target or a false detection, and extracts only the target candidate. A configuration example of the NN used for that purpose is shown in FIG. 15 (multilayer perceptron method), and the unit configuration thereof is shown in FIG. 16 (Non-Patent Document 6). Each unit constituting the NN is formulated by the following equation.

As the activation function, various applications such as a sigmoid function (Non-Patent Document 6) can be applied.

As the input data, based on the coordinates of the target candidate extracted by the detection 31b by the cluster distribution and the target candidate extraction 33, plotting is performed at the points of the three-dimensional coordinates (X, Y, Z) as shown in FIG. The three-dimensional data (Nx, Ny, Nz) is converted into a one-dimensional column vector (1 to Nx × Ny × Nz) including the points in the space where the target does not exist, and input to the NN. For learning the weight of NN, a simulator or an actually measured value is used to generate known three-dimensional learning data of a true value (a teacher signal of a target or a false detection), and the generated signal (X, Y, Z). ) And the teacher signal (target or false positive) to learn in advance. Discrimination between target and false positive corresponds to binary classification and multi-class classification (number of classes = 2).

Generally, in the case of multi-class classification, training data for learning can be defined by the following equation.

Further, the error function E is defined by the squared error of the following equation.

Select w so that this is the smallest. For this purpose, a stochastic gradient descent method (Non-Patent Document 6) or the like may be used.

In the actual processing after learning, the NN of FIG. 15 is operated using a fixed weight, the target or false detection or the target candidate filter 3B4 is passed, and the correlation tracking unit 39 uses only the target candidates. Input to and output the target observation values (X, Y, Z). As described above, the observed value of the target signal, which is less affected by false detection, can be output.

As described above, the radar device according to the sixth embodiment plots in three-dimensional coordinates using the observed value coordinates of the target cluster candidate extracted by the cluster analysis, and inputs the three-dimensional plot coordinates as input to the neural network ( NN) is used to filter the target or false positive. As a result, the target and the erroneous detection can be discriminated by using the NN process, and the process that suppresses the erroneous detection can be performed.

The present invention is not limited to the above embodiment as it is, and at the implementation stage, the components can be modified and embodied within a range that does not deviate from the gist thereof. In addition, various inventions can be formed by an appropriate combination of the plurality of components disclosed in the above-described embodiment. For example, some components may be removed from all the components shown in the embodiments. In addition, components across different embodiments may be combined as appropriate.

1 ... Antenna,
2 ... transmitter / receiver, 21 ... transmitter / receiver, 22 ... modulation control unit,
3 ... Signal processor, 31 ... Σ observation processing unit, 32 ... CFAR processing unit, 33 ... Distance measuring unit, 34 ... Δ observation processing unit, 35 ... Cell extraction unit, 36 ... Angle measuring unit, 37 ... 3D position coordinates Output unit, 38, 38a, 38b ... Cluster analysis unit, 39 ... Correlation tracking unit, 3A ... Target interpolation unit, 3B ... Target candidate extraction filter, 3B1 ... 3D data generation unit, 3B2 ... Column vector conversion unit, 3B3 ... NN (Neural network) Processing unit, 3B4 ... Filter.

Claims (7)

Radar observation area as one frame unit to observe M (M ≧ 2) times, when observing N (N ≧ 2) frames, arranging the observed value of M times per frame to the three-dimensional or two-dimensional coordinate Arrangement means and
For the observed values arranged in the coordinates, cluster analysis is performed based on the difference between the target and the density distribution of false positives on the coordinate axes to extract the target candidate clusters, and the observed values in the cluster for N frames from the target candidate cluster. Alternatively, a radar device including a cluster analysis means for outputting a representative value. The radar device according to claim 1, further comprising a correlation tracking means that performs a correlation tracking process using clusters of target candidates for N (N ≧ 1) frames extracted by the cluster analysis means . The radar device according to claim 1, wherein the cluster analysis means performs the cluster analysis process P (P ≧ 2 ) times with different parameters . Further, a fitting curve is calculated using the observed values in the cluster of the target candidates for the N frames extracted by the cluster analysis means, and the observed values in the region having a predetermined width are extracted from the fitting curve. The radar device according to claim 1, further comprising an interpolation means for interpolating the lack of observed values in the region of the target candidate cluster. The radar device according to claim 1, wherein the cluster analysis means uses a DBSCAN (Density-based Spatial Clustering of Applications with Noise) method as a cluster analysis method. Furthermore, the coordinates of the observed values of the clusters of the target candidates extracted by the cluster analysis are plotted in three-dimensional coordinates, and the three-dimensional plot coordinates are used as input, and whether the target or false detection is performed using a neural network (NN). The radar device according to claim 1, further comprising a filtering means for performing filtering. Radar observation area as one frame unit to observe M (M ≧ 2) times, N (N ≧ 2) when frame observed arranging observations of M times per frame to the three-dimensional or two-dimensional coordinate ,
For the observed values arranged in the coordinates, cluster analysis is performed based on the difference between the target and the density distribution of false positives on the coordinate axes to extract the target candidate clusters, and the observed values in the cluster for N frames from the target candidate cluster. Or a radar signal processing method of a radar device that outputs a representative value.

Images