Radar RD map labeling method based on track table

Abstract

The invention provides a radar RD map marking method based on a track table. The method comprises the following steps: selecting a plurality of pieces of specific information from track table information as judgment conditions; an echo signal of a target is collected and processed to obtain an RD image information trace point of the target, the information trace point comprises data corresponding to the specific information selected in the track table, and a trace point is positioned in a corresponding row of the track table according to one piece of specific information; setting two decisions, judging whether the difference value between the second specific information of the positioned row and the information trace point data is within a threshold value or not, and meeting the judgment I; judging whether the difference value between the specific information of the point and the corresponding information of the line in the track table is within a threshold value or not; and if the two trace points meet the judgment I and the judgment II at the same time and the Doppler value difference of the two trace points is within the domain, judging the target type through the trace points.

Description

Technical Field

[0001] The present invention relates to a radar image annotation technology, in particular to a radar RD image annotation method based on a track table. Background Art

[0002] In the field of deep learning-based radar detection and recognition, accurate target labeling is crucial for subsequent target detection and recognition tasks. Compared with optical images, radar echo images are highly abstract and specialized. They are constructed based on the principles of electromagnetic wave reflection and reception and cannot directly generate intuitive visual images like optical imaging. Common radar detection data construction methods include generating pulse-range diagrams through pulse compression, obtaining range-Doppler diagrams (RD diagrams) through moving target indication (MTI) and moving target detection (MTD), or generating time-frequency diagrams using short-time Fourier transform (STFT). These images have significant visual features different from optical images and lack intuitive visual cues. Therefore, labelers not only need to process complex radar echo data, but also have a solid knowledge base of radar signal processing, covering professional fields such as radar echo characteristics, signal processing algorithms, and target feature extraction. This places extremely high demands and challenges on labeling work. The traditional method for radar image target annotation is manual labeling. However, because different types of targets (such as drones, birds, and other flying objects) in radar images often have highly similar visual characteristics to clutter, they are sometimes difficult to distinguish with the naked eye, making it difficult to obtain large amounts of field-measured annotated data. Furthermore, the amount of image data generated from unfiltered raw radar echoes is extremely large, while the proportion of echo images containing the desired targets is very small. Labelers often have to expend considerable effort to screen and analyze this massive amount of echo data to obtain a small number of echo images containing the desired targets. This makes manual labeling not only a huge workload but also time-consuming and labor-intensive. This severely restricts the efficiency and scale of radar image target annotation, making it difficult to meet the urgent demand for large amounts of high-quality annotated data in practical applications. Summary of the Invention

[0003] The present invention aims to provide a radar RD map annotation method based on a track table, comprising: selecting a plurality of specific information in the track table information as judgment conditions; collecting the echo signal of the target and processing it to obtain a RD map signal point track of the target, the signal point track containing data corresponding to the specific information selected in the track table, and locating the point track in a corresponding row of the track table according to one of the specific information; collecting the echo signal of the target and processing it to obtain a signal point track of the target, the signal point track containing data corresponding to the specific information selected in the track table, and locating the point track in a corresponding row of the track table according to one of the specific information; setting two judgments, wherein judgment one is whether the difference between the second specific information of the located row and the signal point track data is within a threshold, and judgment one is satisfied; judgment two is whether the difference between the specific information of the point and the corresponding information of the row in the track table is within a threshold; for two point tracks, if both judgment one and judgment two are satisfied and the Doppler value difference of the two point tracks is within a domain, then the target type is determined by the point tracks.

[0004] Furthermore, the specific information includes frame count, elevation position, and distance.

[0005] Furthermore, according to the frame count of the signal point track, a corresponding row in the track table that can be used as a reference is searched as the row for positioning.

[0006] Furthermore, in judgment one, if the difference between the distance information recorded in the located row in the track table and the distance data of the signal point track is within 20m, judgment one is satisfied.

[0007] Furthermore, if the distance difference between the information recorded in the located row in the track table in judgment 2 and the signal point track data is less than or equal to 20 meters, the frame count difference is less than 6, and the pitch wave position is the same as the record in the track table, and is not in the 0 Doppler zone, then judgment 2 is satisfied.

[0008] Furthermore, for two adjacent tracks, if both judgment 1 and judgment 2 are satisfied, and the difference between the Doppler value of the next track and the Doppler value of the previous track is not greater than 3, the target type is determined by the track.

[0009] Compared with the existing technology, the present invention has the following advantages: (1) It adopts a two-stage judgment mechanism. The first judgment preliminarily screens potential targets through track table information (such as frame count and distance). The second judgment dynamically verifies the target movement law in combination with Doppler value and eliminates clutter interference. This mechanism can accurately identify UAV targets, reduce mislabeling, and significantly improve the accuracy of labeling results; (2) The present invention automatically completes target classification and labeling through track table information and multi-channel judgment mechanism (such as frame count, distance, pitch wave position and Doppler value judgment). There is no need for labelers to have complex radar signal processing knowledge. The systematic judgment logic simulates the manual experience judgment process, effectively reducing the dependence on the professional ability of the operator.

[0010] The present invention will be further described below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Schematic diagram of the method of the present invention.

[0012] Figure 2 This is a schematic diagram of an excerpt from the track table. DETAILED DESCRIPTION

[0013] Combine Figure 1 A radar RD chart annotation method based on a track table includes the following steps: Step S100, selecting some specific information in the track table as a judgment condition; Step S200 , obtaining a range-Doppler map (RD map) of the target based on moving target detection, and subjecting the target's echo signal to constant false alarm rate (CFAR) detection and trace aggregation processing to obtain a target trace (signal trace). The trace contains data corresponding to specific information selected in the track table; Step S300: Locate a track point in the corresponding row of the track table according to one of the specific information, and compare the difference between the second specific information in the row and the track point data at that location. If the difference is within the threshold, record the Doppler value of the track point and go to step S400; otherwise, delete the track point. Step S400: If the difference between the specific information of the track point and the corresponding information of the row in the track table is within the threshold, the track point is retained and the process goes to step S300 to select the next track point; otherwise, the process goes to step S300 to reselect the track point. Step S500: If there are two points that meet the conditions of step S300 and step S400, the target type is determined.

[0014] This embodiment uses the frame count, pitch position, distance and Doppler value information in the track table as an example to describe the RD map annotation method. If the echo data currently being parsed is the flight data collected by the jingling4 UAV, then at the beginning of the program, set the target type label to "uav-jingling4". Figure 2 In the track representation, select the target distance, pitch position, and frame count in columns 8, 20, and 26, extract the data in these key columns and store them in new variables. For example, Range_xls stores target distance information, bw_xls stores pitch position, and framecnt_xls is used to record frame count.

[0015] After the collected raw echoes are unpacked and processed through the radar signal processing, each track contains various characteristic information, including frame count, pitch position, range, and Doppler value. Based on the frame count of the signal track, a corresponding row in the track table is searched for as a reference. After successfully locating the reference row, the slant range information recorded in that row is compared with the distance information of the signal track. If the difference between the two is within the threshold range (20 meters), the track is considered to be a highly likely drone target. This is the first step in the judgment.

[0016] In actual scenarios, some points that may be formed by clutter may also pass CFAR detection, and the distance information of the point is exactly within the threshold range with the slant distance information recorded in the track table. However, based on the Doppler value information of the previous and next frames of the point, it can be found that there are obvious anomalies, which do not conform to the motion law of a uniformly moving drone. Therefore, it is considered that the point is not a drone target. Figure 2 It can be seen that the frame counts in adjacent rows differ greatly. Figure 2 Taking the first row as an example, all tracklets with frame counts of [315952, 316282) can be located in this row. However, the Doppler values ​​determined by pitch position, distance, and other factors are relatively unique, so there may be anomalies in the Doppler values ​​of the frames before and after the tracklet. To address this situation, an additional discrimination mechanism is required. The approach employed is to select the Doppler value of a CFAR tracklet whose frame count, pitch position, and distance information closely match the tracklet table and set this value as the reference value. Only when the distance, frame count, and Doppler value of the next tracklet are consistent with the tracklet table information and the Doppler reference value can the target type be determined to be a drone. This is the second pass judgment. If the second pass judgment is not met, a new tracklet is selected for the second pass judgment.

[0017] Altitude alignment is defined as follows: the distance difference is less than or equal to 20 meters, the frame count difference is less than 6 compared to the frame count in a row in the track table, the pitch position is the same as the record in a row in the track table, and the point is not in the zero Doppler region (Doppler value is 0 and the maximum value (i.e., pulse number)), and a Doppler reference value is present (indicating that the point is not the starting point). Doppler reference value determination means that the difference between the Doppler value of the next trace point and the Doppler reference value of the previous trace point is less than or equal to 3.

[0018] Through the above two judgments, the first judgment performs preliminary screening based on the frame count and distance information of the track table to lock potential targets; the second judgment dynamically verifies the movement pattern of the target Doppler value (such as whether it meets the characteristics of a uniform-speed drone) to eliminate clutter interference and improve classification accuracy.

[0019] Combine Figure 1In the judgment process, the traces that pass the first judgment but fail the second judgment are marked as "unknown", which corresponds to the traces where the other information mentioned above meets the drone type judgment but the Doppler value is abnormal; and the traces that fail both judgments are marked as "not".

[0020] Once the classification of each point is determined, a targeted RD map containing the drone target can be output, significantly reducing the output of useless images. Furthermore, the annotation file can be output simultaneously with the RD map. The annotation file is output in PascalVOC format, primarily containing the target category of the annotation box and the absolute coordinates of the point's annotation box within the image. The target category is set by referring to the JUDGE function, and the position of the annotation box is a rectangular box drawn centered on the point. The annotation box size is based on empirical values ​​and is set as a certain proportion of the output image size.

Claims

1. A radar RD map annotation method based on a track table, characterized in that: include: Selecting some specific information in the track table information as the judgment condition; The target's echo signal is collected and processed to obtain the target's RD map signal point trace. The signal point trace contains data corresponding to the specific information selected in the track table. According to one of the specific information, a point trace is located in the corresponding row of the track table. Two judgments are set, where the first judgment is whether the difference between the second specific information of the positioned row and the track data of the signal point is within the threshold, and the second judgment is whether the difference between the specific information of the point and the corresponding information of the row in the track table is within the threshold; For two points, if they satisfy both judgment 1 and judgment 2, and the Doppler value difference between the two points is within the domain, then The method according to claim 1, wherein the specific information includes frame count, elevation position, and distance.

2. The method according to claim 2, characterized in that According to the frame count of the signal point track, the corresponding row in the track table that can be used as a reference is searched as the row for positioning.

3. The method according to claim 3, characterized in that In judgment one, if the difference between the distance information recorded in the located row in the track table and the distance data of the signal point track is within 20m, judgment one is satisfied.

4. The method according to claim 4, characterized in that If the distance difference between the information recorded in the located row in the track table in judgment 2 and the signal point track data is less than or equal to 20 meters, the frame count difference is less than 6, and the pitch wave position is the same as the record in the track table, and is not in the 0 Doppler zone, then judgment 2 is satisfied.

5. The method according to claim 5, characterized in that For two adjacent tracks, if both judgment 1 and judgment 2 are satisfied, and the difference between the Doppler value of the next track and the Doppler value of the previous track is not greater than 3, the target type is determined by the track.

Images