A false alarm elimination method for point cloud imaging radar using phase information

Abstract

The application discloses a false alarm elimination method of point cloud imaging radar using phase information, and based on a coherent factor weighting grating lobe suppression method, the variance of the target side lobe / main lobe amplitude ratio is calculated after the target angle power spectrum is weighted by the coherent factor or the phase coherent factor. According to the discrimination threshold drawn by the algorithm model trained by a large amount of data, the variance of the angle detection target side lobe / main lobe amplitude ratio is judged. If the variance is greater than the threshold, it is a false alarm; if the variance is less than or equal to the threshold, it is a real target. The method can achieve a good classification effect and effectively eliminate false alarms.

Description

Technical Field

[0001] This invention relates to a method for eliminating false alarms in point cloud imaging radar using phase information, belonging to the field of radar monitoring. Background Technology

[0002] Radar's application in surveillance is a relatively new one, demonstrating its outstanding advantages in moving target detection. However, this advantage can also lead to false alarms. In many application environments, there are often numerous trees and weeds, and in windy weather, the swaying of trees can be detected by radar, resulting in false alarms. This is especially true in the field of point cloud imaging radar, where, in order to achieve the desired imaging effect and present a more detailed outline of the target, the detection threshold must be set very low, making false alarms more likely.

[0003] In the radar field, target detection rate and false alarm rate are contradictory and cannot be simultaneously achieved at the desired levels. In practical applications, maximizing the target detection rate is the ultimate goal, inevitably leading to an increase in the false alarm rate. Without considering reducing the detection probability, it's necessary to suppress false alarms from multiple perspectives to reduce them to a level that minimizes the false alarm probability. Existing methods for false alarm elimination are numerous, including those based on target micro-Doppler characteristics, logical management strategies, and multi-sensor fusion. However, these cannot guarantee that radar will be completely free of false alarms. More false alarm elimination methods are needed to refine the theory and further reduce the false alarm probability. Summary of the Invention

[0004] This invention aims to provide a false alarm elimination method for point cloud imaging radar that utilizes phase information. Based on the coherence factor weighted grating lobe suppression method, after coherence factor weighting, combined with the fluctuation judgment of the amplitude ratio of the angle measurement sidelobe and the main lobe, a new false alarm elimination method is proposed to further reduce the false alarm probability.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a method for false alarm rejection of point cloud imaging radar using phase information, the method being as follows:

[0006] Step 1), acquire two-dimensional echo sampling data;

[0007] Step 2): After performing two sequential discrete Fourier transforms on the data obtained in Step 1), a two-dimensional detection matrix is ​​obtained.

[0008] Step 3) Using constant false alarm rate (CFAR) detection technology, the target point information of the target in the two-dimensional detection matrix is ​​found. The target point information includes distance information, velocity information, and intensity information. The distance information and velocity information correspond to the coordinates of the target point in the two-dimensional detection matrix, and the intensity information corresponds to the amplitude of the target point in the two-dimensional detection matrix.

[0009] Step 4): Using the coordinates of the target point as an index, extract the corresponding amplitude from the two-dimensional detection matrix to form an array Y(i), where 1≤i≤M; the array Y(i) is subjected to Fourier transform to obtain the target angular power spectrum P; where M is the number of receiving antennas of the radar;

[0010] Step 5), calculate the coherence factor CF using the array Y(i):

[0011]

[0012] Step 6): Perform a weighting operation on the target angle power spectrum P to obtain the weighted target angle power spectrum P. CF :

[0013] P CF =P·CF

[0014] Step 7), find the target angular power spectrum P CF The maximum value in is denoted as A. L A L The corresponding target angular power spectrum index is the angular position of the target, denoted as L;

[0015] Step 8), repeat steps 1) to 7) for K detections. The variance of the sidelobe-to-mainlobe amplitude ratio at the target point is calculated as follows:

[0016]

[0017]

[0018] Where Ratio(k) is the amplitude ratio of the sidelobe to the main lobe of the target point in the k-th detection, 1≤k≤K. Let S be the mean of the sidelobe-to-mainlobe amplitude ratios for K detections of the target point, and let S be the variance representing the fluctuation of the sidelobe-to-mainlobe amplitude ratio. Let be the amplitude corresponding to the position of Li in the weighted target angular power spectrum after the k-th detection, i = -N,...,N, and i ≠ 0, where N is an integer;

[0019] Step 9) Collect two-dimensional echo sampling data of Z groups containing false alarms and real targets. Label the two-dimensional echo sampling data, label the false alarm data with false alarm labels, and label the real target data with real target labels to obtain a data training set. Repeat steps 2) to 8) for the data training set to obtain the variance of the real target data. Select an appropriate discrimination threshold Th so that the number of variances of the real target data in Z groups that are less than Th exceeds a set value.

[0020] Step 10): For the test target data, repeat steps 2) to 8). If the calculated variance is greater than the threshold Th, the test target is considered a false alarm; otherwise, the test target is considered a real target.

[0021] S>Th→False Alarm

[0022] S≤Th→True Target

[0023] The coherence factor weighted grating lobe suppression method is mainly used to suppress grating lobes generated by radar during angle measurement and has no relation to radar false alarm elimination. In a certain experiment, the following pattern was accidentally discovered: after applying coherence factor weighting to all detected target points, under multiple detections, the amplitude of the angle measurement sidelobe main lobe of the detected points for trees and noise-induced false alarms was larger than the fluctuation, while the amplitude of the angle measurement sidelobe main lobe of the detected points for the monitored targets of interest, such as people and vehicles, was much smaller than the fluctuation. From the perspective of physical laws and mathematical formulas, the reason for the above phenomenon is that real-world noise is mostly Gaussian noise with no fixed variation. After a tree sways, the speeds of its branches and trunk are not the same, and the speed distribution is relatively chaotic. Therefore, their phase coherence is poor. After applying coherence factor weighting, the amplitude of the angle measurement sidelobe main lobe of the detected results in different frames is larger than the fluctuation. The human body only has the speed of the torso and limbs, and it swings back and forth in a regular manner. Vehicles are rigid objects, and different parts have the same speed. Therefore, the phase consistency between the human body and the vehicle is better reflected. After coherence factor weighting, the amplitude ratio of the angle measurement sidelobe and main lobe in different frame detection results fluctuates much less. Based on the above analysis, it can be seen that target category classification can be performed by weighting the parameters of false alarms and true targets and calculating the variance of the angle measurement sidelobe and main lobe amplitude ratio. By defining a discrimination threshold, target category identification can be performed based on the obtained threshold, thereby accurately filtering out false alarm targets.

[0024] According to embodiments of the present invention, the present invention can be further optimized, and the optimized technical solution is as follows:

[0025] In one preferred embodiment, step 5) is replaced by: calculating the phase coherence factor (PCF).

[0026] PCF = [1 - σ(Y)] Q

[0027] Where Q is a constant greater than zero;

[0028]

[0029] Step 6) is replaced with: performing a weighting operation on the target angle power spectrum P to obtain the weighted target angle power spectrum P. CF :

[0030] P CF=P·PCF

[0031] In one preferred embodiment, K ≥ 6.

[0032] In one preferred embodiment, N takes the value range [5, 15].

[0033] Based on the same inventive concept, the present invention also provides an electronic device, the electronic device including a memory and one or more processors; the memory stores one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the steps of any of the above methods.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] This invention, by weighting the parameters of the detected target and determining the variance of the sidelobe-to-mainlobe amplitude ratio based on a defined discrimination threshold, identifies targets exceeding the threshold as false alarms and targets less than or equal to the threshold as true targets. Analysis of experimental results shows that this method achieves good classification performance and effectively eliminates false alarms. Attached Figure Description

[0036] Figure 1 This is a flowchart of the false alarm rejection method provided in the embodiments of the present invention;

[0037] Figure 2 This is an experimental result diagram of one embodiment of the present invention. Detailed Implementation

[0038] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0039] Coherence factor weighted grating lobe suppression method:

[0040] Assuming the radar has M receiving antennas, during a single detection operation, the radar data is first processed using signal processing techniques to obtain M two-dimensional detection matrices. Specifically, the radar acquires the two-dimensional raw data, and after two sequential discrete Fourier transforms, a two-dimensional frequency domain matrix is ​​obtained, which can also be called the two-dimensional detection matrix [Reference: Li Nan, ed. Research on Signal Processing of Linear Frequency Modulated Continuous Wave Radar [D]. Xi'an University of Electronic Science and Technology, 2012.]. The target point information in the two-dimensional detection matrix is ​​found using constant false alarm rate (CFAR) detection technology. The target point information includes range information, velocity information, and intensity information. The range and velocity information correspond to the target point's coordinates (x, y) in the two-dimensional detection matrix, and the intensity information corresponds to the target point's amplitude in the two-dimensional detection matrix [Reference: Zhao Likai, ed. Research on Constant False Alarm Rate Detection Algorithm for Radar Targets [D]. Beijing Institute of Technology, 2016.]. Using the coordinates (x, y) of the target point as indices, extract the corresponding amplitudes (in complex number format) from the M two-dimensional detection matrices to form an array Y(i), where i = 1, ..., M. After Fourier transform, the array Y(i) yields the target angular power spectrum P, which is a one-dimensional array whose length is determined by the number of points used in the Fourier transform [Reference: He Zishu, Xia Wei, et al. Modern Digital Signal Processing and Applications [M]. Beijing: Tsinghua University Press, 2009, 287-338].

[0041] The coherence factor-weighted grating lobe suppression method uses the array Y(i) to calculate the following formula:

[0042]

[0043] Where CF is the coherence factor to be calculated. After obtaining the coherence factor CF, a weighting operation is performed:

[0044] P CF =P·CF

[0045] The target angular power spectrum P after coherence factor weighting is obtained CF .

[0046] Side lobe to main lobe amplitude ratio fluctuation:

[0047] Find P CF The largest amplitude is denoted as A. L The A L Corresponding P CF The sequence number represents the angular position of the target, denoted as L. The side lobe-to-main lobe amplitude ratio is calculated as follows:

[0048]

[0049] in, Let be the amplitude corresponding to the position of Li in the weighted target angular power spectrum after the k-th detection, i = -N,...,N, and i ≠ 0. N is usually an integer between 5 and 15, and its specific value is related to M: M represents the number of receiving channels, i.e., the number of receiving antennas. The number of receiving antennas and the antenna spacing (conventional radars use half-wavelength spacing arrays) determine the array pattern of the radar receiver; if the array pattern is fixed, the positions of the main lobe and side lobes are fixed. The calculation of the array pattern is a fundamental knowledge in the field of array antennas. The rule for the values ​​of N and M is: the larger M is, the smaller the value of N will be;

[0050] By repeating the detection K times, the variance of the sidelobe-to-mainlobe amplitude ratio at the target point is calculated.

[0051]

[0052]

[0053] Where Ratio(k) is the amplitude ratio of the sidelobe to the main lobe of the target point in the k-th detection, k = 1,...,K. Let S be the mean of the sidelobe-to-mainlobe amplitude ratio of the target point in K detections, and let S be the variance, which mathematically represents the fluctuation of the sidelobe-to-mainlobe amplitude ratio.

[0054] The method steps provided in this embodiment of the invention are as follows:

[0055] (1-1) Assuming the radar has M receiving antennas, during a single detection operation, the radar data is first processed using signal processing techniques to obtain M two-dimensional detection matrices. Specifically, the radar acquires the two-dimensional raw data, and after two sequential discrete Fourier transforms, a two-dimensional frequency domain matrix is ​​obtained, which can also be called the two-dimensional detection matrix [Reference: Li Nan, ed. Research on Signal Processing of Linear Frequency Modulated Continuous Wave Radar [D]. Xi'an University of Electronic Science and Technology, 2012.]. The target point information in the two-dimensional detection matrix is ​​found using constant false alarm rate (CFAR) detection technology. The target point information includes range information, velocity information, and intensity information. The range and velocity information correspond to the target point's coordinates (x, y) in the two-dimensional detection matrix, and the intensity information corresponds to the target point's amplitude in the two-dimensional detection matrix [Reference: Zhao Likai, ed. Research on Constant False Alarm Rate Detection Algorithm for Radar Targets [D]. Beijing Institute of Technology, 2016.]. (x, y) are also called distance and velocity points, i.e., the coordinates of the target in the two-dimensional detection matrix. The index of one dimension (row or column) of the two-dimensional detection matrix corresponds to the distance information of the coordinates, and the index of the other dimension (column or row) corresponds to the velocity information of the coordinates. Using the coordinates (x, y) of the target point as the index, the corresponding amplitudes (in complex number format) in the M two-dimensional detection matrices are extracted to form an array Y(i), where i = 1,...,M. After Fourier transform, the target angular power spectrum P is obtained from the array Y(i). P is a one-dimensional array, and the length of the array is determined by the number of points used in the Fourier transform [Reference: He Zishu, Xia Wei, et al., eds. Modern Digital Signal Processing and Applications [M]. Beijing: Tsinghua University Press, 2009, 287-338]. The target angular power spectrum P obtained from the array Y(i) can also be achieved through other commonly used signal processing methods.

[0056] (1-2) There are two ways to perform a weighting operation on the target angular power spectrum P:

[0057] The first method is to calculate the coherence factor (CF):

[0058]

[0059] After obtaining the coherence factor, a weighting operation is performed:

[0060] P CF =P·CF

[0061] The target angular power spectrum P after coherence factor weighting is obtained CF .

[0062] The second method involves calculating the phase coherence factor (PCF):

[0063] PCF = [1 - σ(Y)] Q

[0064] Q is a constant greater than zero, usually taken as 1. The value of Q acts as an amplifier, affecting the values ​​of the main lobe and side lobes, but it does not affect the determination of the variance of the fluctuation.

[0065]

[0066] After obtaining the phase coherence factor, a weighting operation is performed:

[0067] P CF =P·PCF

[0068] The target angular power spectrum P obtained after phase coherence factor weighting CF .

[0069] The first method and the second method calculate P CF There are differences, but they do not affect the subsequent judgment of fluctuations in the side lobe-to-main lobe amplitude ratio. See the examples section for detailed comparison results.

[0070] (1-3) Obtain P CF Then, find P CF The largest amplitude is denoted as A. L The A L Corresponding P CF Let L be the angular position of the target. The side lobe-to-main lobe amplitude ratio is calculated as follows:

[0071]

[0072] in, Let be the amplitude corresponding to the position of Li in the weighted target angular power spectrum after the k-th detection, i = -N,...,N, and i ≠ 0. N is usually an integer between 5 and 15, and its specific value is related to M: M represents the number of receiving channels, i.e., the number of receiving antennas. The number of receiving antennas and the antenna spacing (conventional radars use half-wavelength spacing arrays) determine the array pattern of the radar receiver; if the array pattern is fixed, the positions of the main lobe and side lobes are fixed. The calculation of the array pattern is a fundamental knowledge in the field of array antennas. The rule for the values ​​of N and M is: the larger M is, the smaller the value of N will be;

[0073] (1-4) Repeat the detection K times and calculate the variance of the sidelobe-to-mainlobe amplitude ratio at the target point.

[0074]

[0075]

[0076] Where Ratio(k) is the amplitude ratio of the sidelobe to the main lobe of the target point in the k-th detection, k = 1,...,K. Let K be the mean of the sidelobe-to-mainlobe amplitude ratios from K detections of the target point, and let S be the variance representing the fluctuation of the sidelobe-to-mainlobe amplitude ratio. It is generally recommended that K be greater than or equal to 8; however, in practical applications, it will vary depending on the specific needs. For example, when a higher recognition accuracy is required, K can be set to 10 or a larger value. When a faster recognition result is desired, K can be set to 7 or 6, but this will result in a loss of recognition accuracy.

[0077] (1-5) Using sensor equipment, Z sets of two-dimensional raw data containing false alarms and real targets are collected in the field. The two-dimensional raw data are labeled, that is, false alarm data are labeled with false alarm labels and real target data are labeled with real target labels, thus obtaining a data training set. After repeating the above experiment on the data training set, the labeled variance S is obtained; an appropriate discrimination threshold Th is selected so that the number of variances less than Th in the variance S of the Z sets of real target data exceeds a set value; the set value is 2 / 3*Z by default.

[0078] (1-6) For the test data, if S is greater than the threshold Th, the target is considered a false alarm due to trees or noise; otherwise, the target is considered a real target and is retained.

[0079] S>Th→False Alarm

[0080] S≤Th→True Target

[0081] Figure 2 This is the result of processing vehicle and swaying tree data using the method provided in this embodiment of the invention. Two methods, CF weighting and PCF weighting, were used to process the vehicle and swaying tree data respectively. Based on the experimental results, it can be found that: (1) the variances of swaying trees and vehicles are significantly different, with swaying trees having a larger variance and vehicles a smaller variance; (2) the results obtained for CF weighting and PCF weighting are slightly different, but this does not change the conclusion in (1); (3) based on the conclusions of (1) and (2), an appropriate threshold can be set. In this experiment, the threshold was set to 0.5. The variance results of trees and vehicles were compared with the threshold multiple times. In 10 comparisons, the variance of trees was greater than the set threshold at least 8 times, while the variance of vehicles was less than the set threshold in all 10 comparisons. Therefore, applying the method provided in this embodiment of the invention to the detection target can effectively eliminate false alarms.

[0082] The 10 calculation results in Example 1 refer to repeating steps (1-1) to (1-4) for 10 sets of experiments. In these 10 sets of experiments, the sensor installation location or the model of the observed target was changed to improve the comprehensiveness of data coverage. In these 10 sets of experiments, the variance of trees exceeded the defined threshold value 8 times, while the variance of vehicles was always less than the defined threshold value. This indicates that the threshold value is set relatively reliably and is unlikely to produce false positives; however, it does not mean that the variance of vehicles will always be less than the threshold value.

[0083] In steps (1-4), the K detections refer to the radar performing K detections in a single experiment. This means that the training dataset collected in a single experiment has a time length and includes a certain time span. The flowchart of the method provided in this embodiment is as follows: Figure 1 As shown.

[0084] The above embodiments should be understood as being used only to illustrate the present invention more clearly, and not to limit the scope of the present invention. After reading the present invention, any modifications of the present embodiments by those skilled in the art will fall within the scope defined by the appended claims.

Claims

1. A method for false alarm rejection in point cloud imaging radar utilizing phase information, characterized in that, Includes the following steps: Step 1), acquire two-dimensional echo sampling data; Step 2): After performing two sequential discrete Fourier transforms on the data obtained in Step 1), a two-dimensional detection matrix is ​​obtained. Step 3) Using constant false alarm rate (CFAR) detection technology, the target point information of the target in the two-dimensional detection matrix is ​​found. The target point information includes distance information, velocity information, and intensity information. The distance information and velocity information correspond to the coordinates of the target point in the two-dimensional detection matrix, and the intensity information corresponds to the amplitude of the target point in the two-dimensional detection matrix. Step 4): Using the coordinates of the target point as an index, extract the corresponding amplitude from the two-dimensional detection matrix to form an array Y(i), where 1≤i≤M; the array Y(i) is subjected to Fourier transform to obtain the target angular power spectrum P; where M is the number of receiving antennas of the radar; Step 5), calculate the coherence factor CF using the array Y(i): Step 6): Perform a weighting operation on the target angle power spectrum P to obtain the weighted target angle power spectrum P. CF : P CF =P·CF Step 7), find the target angular power spectrum P CF The maximum value in is denoted as A. L A L The corresponding target angular power spectrum index is the angular position of the target, denoted as L; Step 8), repeat steps 1) to 7) for K detections. The variance of the sidelobe-to-mainlobe amplitude ratio at the target point is calculated as follows: Where Ratio(k) is the amplitude ratio of the sidelobe to the main lobe of the target point in the k-th detection, 1≤k≤K. Let S be the mean of the sidelobe-to-mainlobe amplitude ratios for K detections of the target point, and let S be the variance representing the fluctuation of the sidelobe-to-mainlobe amplitude ratio. Let be the amplitude corresponding to the position of Li in the weighted target angular power spectrum after the k-th detection, i = -N,...,N, and i ≠ 0, where N is an integer; Step 9) Collect two-dimensional echo sampling data of Z groups containing false alarms and real targets. Label the two-dimensional echo sampling data, label the false alarm data with false alarm labels, and label the real target data with real target labels to obtain a data training set. Repeat steps 2) to 8) for the data training set to obtain the variance of the real target data. Select an appropriate discrimination threshold Th so that the number of variances of the real target data in Z groups that are less than Th exceeds a set value. Step 10): For the data of the test target, repeat steps 2) to 8). If the calculated variance is greater than the threshold Th, the test target is identified as a false alarm; otherwise, the test target is identified as a real target.

2. The method for false alarm rejection of point cloud imaging radar using phase information according to claim 1, characterized in that, Step 5) is replaced with: Calculate the phase coherence factor (PCF): PCF=[1-σ(Y)] Q Where Q is a constant greater than zero; Step 6) is replaced with: performing a weighting operation on the target angle power spectrum P to obtain the weighted target angle power spectrum P. CF : P CF =P·PCF。 3. The method for false alarm rejection of point cloud imaging radar using phase information according to claim 1, characterized in that, K≥6。 4. The method for false alarm rejection of point cloud imaging radar using phase information according to claim 1, characterized in that, The value of N is in the range [5, 15].

5. An electronic device, characterized in that, The electronic device includes a memory and one or more processors; the memory stores one or more programs, which, when executed by the one or more processors, cause the one or more processors to perform the steps of the method according to any one of claims 1 to 4.

Images