Multi-radar system error correction method based on DBSCAN clustering

Abstract

The invention discloses a multi-radar system error correction method based on DBSCAN clustering, and relates to the field of radar target tracking. Multi-radar plot fusion can improve the target tracking precision, but different radars inevitably generate system errors due to differences between installation and performance parameters, resulting in uncertainty of target measurement; meanwhile, in leak detection and false alarm scenes, system errors are difficult to accurately estimate. The method comprises the following steps: classifying inter-radar offsets (namely vector difference values between reference radar measurement extrapolation values and radar measurement to be corrected) in w continuous timestamps by utilizing DBSCAN clustering, extracting an offset cluster with the highest density, taking a mean value as a real offset, and eliminating false offsets generated by missing detection and false alarm; meanwhile, random errors can be eliminated by filtering the real offset, and a more accurate system error estimation value is obtained.

Description

Technical Field

[0001] The present invention relates to a multi-radar system error correction method based on DBSCAN clustering in the field of radar technology, in particular to a method for correcting system errors among multi-radar systems based on a DBSCAN (Density-Based Spatial Clustering of Application with Noise) clustering algorithm. Background Art

[0002] In the process of multi-radar point-track fusion tracking, it is usually necessary to convert the measurements of each radar into a unified coordinate system before subsequent fusion processing. Due to differences in installation and performance parameters, different radars will inevitably produce systematic errors. The existence of systematic errors will cause different radars to have certain deviations in the measurement of the same target in a unified coordinate system, resulting in an increase in the probability of point-track correlation errors from multiple radars of the same target, affecting the target tracking effect. In the scenario without missed detection and false alarm interference, the average values ​​of the two radar measurements are calculated and subtracted respectively, which can better obtain the systematic error between the two radars. However, in the case of missed detection or false alarm, missed detection will cause the systematic error between radars to be unable to be calculated, and false alarm will cause a large deviation in the calculation of the systematic error. However, existing methods are difficult to overcome the problem that the systematic error between multiple radars cannot be accurately obtained in the scenario of missed detection or false alarm. Summary of the invention

[0003] In view of this, the present invention provides a multi-radar system error correction method based on DBSCAN clustering, which can effectively overcome the problem that the multi-radar system error is difficult to accurately estimate under missed detection and false alarm interference.

[0004] In order to meet the above requirements, the technical solution adopted by the present invention is:

[0005] A multi-radar system error correction method based on DBSCAN clustering includes the following steps:

[0006] Step 1: Configure system parameters, including radar ID, radar data transmission cycle, sensor latitude and longitude, measurement error and detection range;

[0007] Step 2: After the radar sends the point data, the coordinate system of the radar with the highest data rate is regarded as the reference coordinate system, and the corresponding radar is called the reference radar; then, based on the sensor latitude and longitude, the measurement values ​​and measurement errors of all radars are converted to Cartesian coordinates, and the measurement errors are compensated to the measurement values ​​to form a measurement data set in the Cartesian coordinate system, including real target measurement and clutter; the measurement values ​​include radial distance, azimuth and pitch angle;

[0008] Step 3: Within N consecutive time stamps, the reference radar scans, then performs correlation and filtering. If the associated measurement number is greater than the threshold value M, the target is deemed to have successfully started the track. After the track is successfully started, the correlation and filtering operations are continued until the error remains stable. Wherein, M and N are both set values.

[0009] Step 4: Set the clustering time window w, record the current time as t, and calculate (t-w+1) ～ The vector difference between the extrapolated values ​​of the non-reference radar and its nearest reference radar measurement at timestamp t is saved in the offset matrix;

[0010] Step 5: Classify the offset matrices obtained in w time windows based on the DBSCAN clustering method, count the offset classes of each cluster after classification, and extract the cluster with the highest data density as the real offset cluster;

[0011] Step 6: Calculate the mean of all offsets in the real offset cluster at time t and perform filtering to obtain the systematic error between the reference radar and non-reference radar measurements.

[0012] The beneficial effects of the present invention are:

[0013] Different radars will inevitably produce systematic errors due to differences in installation and performance parameters, resulting in uncertainty in target measurement; at the same time, in missed detection and false alarm scenarios, the systematic error is difficult to accurately estimate. The present invention uses DBSCAN clustering to classify the offsets between radars within w consecutive timestamps, extracts the offset clusters with the highest density and takes the mean as the true offset, and eliminates the false offsets caused by missed detection and false alarm; at the same time, filtering the true offset can eliminate random errors and obtain a more accurate estimate of the systematic error. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of a multi-radar system error correction method according to an embodiment of the present invention.

[0015] Figure 2 It is the simulation experiment result of the multi-radar (radar 1 and radar 2) system error correction method according to the embodiment of the present invention.

[0016] Figure 3 It is the simulation experiment result of the multi-radar (radar 1 and radar 3) system error correction method according to the embodiment of the present invention. DETAILED DESCRIPTION

[0017] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0018] like Figure 1, a multi-radar system error correction method based on DBSCAN clustering, comprising the following steps:

[0019] Step 1: Parameter configuration: configure system parameters, including radar ID, radar data sending cycle, sensor latitude and longitude, measurement error, detection range, etc.

[0020] Step 2: Data preprocessing: After the radar sends the point data, all radar measurements (radial distance, azimuth, and elevation) need to be converted into a unified Cartesian coordinate system. The coordinate system of the radar with the highest data rate is usually regarded as the reference coordinate system, and its radar is called the reference radar. Then, based on the sensor latitude and longitude, all radar measurements and measurement errors are converted into Cartesian coordinates, and the measurement errors are compensated to the measurement values ​​to form a measurement data set in the Cartesian coordinate system (including real target measurements and clutter).

[0021] Step 3, track initiation and maintenance, uses the measurement of the reference radar to initiate the track of multiple targets, and uses the M / N logic method to initiate and terminate the track. The process is: within the N consecutive timestamps scanned by the reference radar, if the number of associated measurements is greater than the threshold value M, the target track is considered to have been successfully initiated; after the track is successfully initiated, the association and filtering operations are continued until the error remains stable; where M and N are both set values.

[0022] Step 4: Multi-window offset calculation. Set the clustering time window w, record the current time as t, and calculate (t-w+1) ～ The vector difference between the extrapolated values ​​of the non-reference radar and its nearest temporal neighbor reference radar measurements at timestamp t is saved in the offset matrix.

[0023] Step 5, DBSCAN clustering, automatically classifies the offset matrices obtained in w time windows based on the DBSCAN clustering method, and counts the offset classes of each cluster after classification, and extracts the one with the highest data density as the real offset cluster.

[0024] Step 6, system error calculation, calculate the mean of all offsets in the real offset cluster at time t, and perform a first-order extended Kalman filter on the average offset to obtain the system error between the reference radar and non-reference radar measurements.

[0025] Method simulation experiment verification:

[0026] Plot simulation: 3 targets, 3 radars, the scanning period of radar 1 is 1.0s, the latitude and longitude are (28.4, 119.2, 100), the radar installation angle error is (0.1°, 0.2°, -0.1°), and the distance error is (10m, 5m, 5m). The scanning period of radar 2 is 1.2s, the latitude and longitude are (28.4, 120.1, 706.6), the radar installation angle error is (-0.1°, 0.2°, -0.1°), and the distance error is (-10m, -5m, 5m). The scanning period of radar 3 is 1.3s, the latitude and longitude are (28.8, 119.7, 404.1), the radar installation angle error is (-0.1°, -0.2°, -0.1°), and the distance error is (-10m, 5m, -5m). Radar 1 is used as the reference radar, and radars 2 and 3 are used as non-reference radars. In addition, a clutter is randomly generated around the true measurement of each target in each timestamp, and the radar measurement miss detection rate is set to 0.1.

[0027] The calculation results of the system error between the reference radar and the non-reference radar are as follows: Figure 2 and Figure 3 As shown in the figure, it can be seen intuitively that due to differences in installation errors, the system error results between different radars are also different. However, the specific value of the system error in each timestamp can be quantitatively calculated using this method.

[0028] In order to evaluate the accuracy of the system error calculation, the measured values ​​processed by the method of the present invention are respectively subjected to point trace fusion (i.e., correlation and filtering), and the filtering results are compared. The specific filtering errors are shown in the following table:

[0029] Uniform linear motion Objective 1 Filter RMSE Objective 2 Filter RMSE Objective 3 Filter RMSE Correction of systematic errors 23.18 21.62 23.10 Uncorrected systematic error 229.86 222.13 221.53

[0030] It can be seen from the above table that after the measurement values ​​processed by the method of the present invention are fused with points, the target tracking effect is significantly improved, and it can be intuitively seen that its filtering error is much smaller than the filtering result of the uncorrected system error, indicating that the present invention can more accurately calculate the system error value between radars.

[0031] In summary, the present invention can better solve the problem of difficult estimation of system errors due to missed detection and false alarm by clustering the system error offsets in multiple timestamps through DBSCAN, correctly correct the system error, and significantly improve the real-time tracking accuracy of the target, which has good application prospects.

Claims

1. A multi-radar system error correction method based on DBSCAN clustering, characterized in that: The following steps are involved: Step 1: Configure system parameters, including radar ID, radar data transmission cycle, sensor latitude and longitude, measurement error and detection range; Step 2: After the radar sends the point data, the coordinate system of the radar with the highest data rate is regarded as the reference coordinate system, and the corresponding radar is called the reference radar; Then, based on the sensor latitude and longitude, all radar measurement values ​​and measurement errors are converted to Cartesian coordinates, and the measurement errors are compensated to the measurement values ​​to form a measurement data set in the Cartesian coordinate system, including real target measurement and clutter; wherein the measurement values ​​include radial distance, azimuth and pitch angle; Step 3: Within N consecutive time stamps, the reference radar scans, then performs correlation and filtering. If the associated measurement number is greater than the threshold value M, the target is deemed to have successfully started the track. After the track is successfully started, the correlation and filtering operations are continued until the error remains stable. Wherein, M and N are both set values. Step 4: Set the clustering time window w, record the current time as t, and calculate (t-w+1) ～ The vector difference between the extrapolated values ​​of the non-reference radar and its nearest reference radar measurement at timestamp t is saved in the offset matrix; Step 5: Classify the offset matrices obtained in w time windows based on the DBSCAN clustering method, count the offset classes of each cluster after classification, and extract the cluster with the highest data density as the real offset cluster; Step 6: Calculate the mean of all offsets in the real offset cluster at time t, and filter the average offset to obtain the systematic error between the reference radar and non-reference radar measurements.

Images