A multi-radar omnidirectional detection method for unmanned aerial vehicle inertial navigation system

Abstract

The invention discloses a multi-radar omnidirectional detection method for an unmanned aerial vehicle inertial navigation system, and relates to the field of radar detection technology. The multi-radar omnidirectional detection method for the unmanned aerial vehicle inertial navigation system comprises the following steps: S1, obtaining point cloud detection results; S2, obtaining target data; S3, target tracking judgment; S4, autonomous obstacle avoidance judgment. The invention obtains two-dimensional frequency domain data by compressing radar raw data, and obtains point cloud detection results based on the two-dimensional frequency domain data, and then determines whether to track the target based on the point cloud detection results and the track starting conditions, and determines whether to update the target data and maintain the track based on the correlation results between the point cloud and the track, and finally obtains multi-dimensional omnidirectional track point data based on the radar data, and determines whether to perform autonomous obstacle avoidance based on the obtained radar output distance and the preset safety distance, so as to achieve a more accurate multi-radar omnidirectional detection effect, and solves the problem that the multi-radar omnidirectional detection results in the prior art are greatly affected by the environment.

Description

Technical Field

[0001] The present invention relates to the field of radar detection technology, and in particular to a multi-radar omnidirectional detection method of an unmanned aerial vehicle (UAV) inertial navigation system. Background Art

[0002] With the rapid development of drones, industrial application-level drones have begun to be widely used in various fields, among which drones play a very important role in plant protection. However, when working continuously over a large area in complex environments such as steep mountainous areas, drones are prone to accidents due to obstacles such as power lines, hillsides and trees. The method proposed in this application integrates radar data from six directions: front, back, left, right, up and down, and then obtains two-dimensional frequency domain data through compression. Constant false alarm rate detection is then performed to obtain point cloud detection results, forming multi-dimensional omnidirectional detection to achieve comprehensive perception of the surrounding environment without being affected by ambient light and dust. It feeds back information such as the relative distance, azimuth and altitude between obstacles and radar, guiding the drone system to autonomously avoid obstacles and achieve true omnidirectional safety protection.

[0003] The prior art invention patent application with publication number CN115033026A discloses a millimeter-wave radar obstacle avoidance and altitude determination method for a side-mounted drone equipped with a gimbal. The method includes pre-processing the radar raw data; outputting three-dimensional target point cloud data after processing with a high-precision radar angle resolution algorithm; segmenting the entire area point cloud, with the point cloud near -45 degrees as the obstacle avoidance processing area and the point cloud near +45 degrees as the altitude determination processing area. The three-dimensional point cloud data is detected and tracked within the obstacle avoidance processing area. If the point cloud of several consecutive frames matches its speed, an obstacle avoidance distance track is formed, otherwise the track is cleared; the drone altitude data is calculated in the altitude determination source point cloud cluster; and the data results are fused to realize the omnidirectional detection function of a single radar.

[0004] For example, the invention patent with publication number CN115980742B discloses a radar detection method and device for unmanned aerial vehicles. The method obtains the current flight information of the UAV through a first radar, and regularly uploads the current flight information data to a central controller for monitoring and early warning. When the current flight information is analyzed by the central controller and it is determined that the UAV has deviated from its flight course, a yaw alarm instruction of the UAV is sent to the second radar. The second radar obtains the simulated yaw planning information of the UAV based on the yaw position of the UAV. The second radar sends the simulated yaw planning information of the UAV to the UAV, and the UAV performs flight navigation based on the simulated yaw planning information.

[0005] For example, the invention patent with publication number CN113820709B discloses a wall-penetrating radar detection system and detection method based on a drone, which includes: a drone, an environmental perception unit, a wall-penetrating radar, a gimbal and a control unit; the drone, the environmental perception unit, the wall-penetrating radar and the gimbal are all connected to the control unit; the wall-penetrating radar is located on the drone and is used to detect the external structure, internal structure and internal personnel of the building to be tested; the environmental perception unit is located on the drone and is used to sense the flight environment information of the drone and send the flight environment information to the control unit; the control unit controls the flight of the drone according to the flight environment information so that the distance between the drone and the outer wall is kept within a certain range; the gimbal is located on the drone, and the transceiver antenna of the wall-penetrating radar is located on the gimbal, and the control unit controls the movement of the gimbal according to the flight environment information so that the array surface of the transceiver antenna remains parallel to the outer wall.

[0006] However, in the process of implementing the technical solutions of the embodiments of the present application, the inventors of the present application discovered that the above technology has at least the following technical problems:

[0007] Existing technologies typically rely on a single radar (such as a side-mounted millimeter-wave radar) mounted on the drone's gimbal and its gimbal adjustment capabilities to accurately detect and measure the height of obstacles ahead. This improves the drone's altitude determination capabilities and is suitable for drone systems with high requirements for obstacle avoidance and altitude determination. However, these systems typically rely on a relatively single data source, primarily focusing on information about obstacles ahead of the drone, and offer limited comprehensive environmental awareness. In complex mountainous environments, power lines, hillsides, and trees can obstruct radar signals, resulting in blind spots. Furthermore, meteorological conditions such as dust, haze, rain, and snow can attenuate radar signals, leading to significant environmental impact on the omnidirectional detection results of multiple radars. Summary of the Invention

[0008] The embodiments of the present application provide a multi-radar omnidirectional detection method for an unmanned aerial vehicle inertial navigation system, thereby solving the problem in the prior art that the multi-radar omnidirectional detection results are greatly affected by the environment, and achieving more accurate multi-radar omnidirectional detection.

[0009] An embodiment of the present application provides a multi-radar omnidirectional detection method for an unmanned aerial vehicle (UAV) inertial navigation system, comprising the following steps: S1, compressing collected radar raw data to obtain two-dimensional frequency domain data, accumulating the two-dimensional frequency domain data and performing constant false alarm rate detection to obtain a point cloud detection result, wherein the compression includes one-dimensional compression and two-dimensional compression; S2, classifying the point cloud to obtain a point cloud clustering result, obtaining target data based on the point cloud clustering result, and classifying an obstacle target to be detected based on the target data to obtain a target classification result, wherein the obstacle target to be detected represents an obstacle in the flight direction of the UAV; S3, determining whether to track the obstacle target to be detected based on the point cloud detection result and a track starting condition, and determining whether to update the target data and maintain the track based on a result of associating the point cloud with the track; if the point cloud and the track are not successfully associated, only saving historical data; S4, splicing the radar data to obtain multi-dimensional omnidirectional track point data, obtaining a radar output distance based on the flight direction of the UAV, and determining whether to perform autonomous obstacle avoidance based on the radar output distance and a preset safety distance.

[0010] Furthermore, the specific process of the constant false alarm rate detection is as follows: performing a one-dimensional Fourier transform on the radar raw data to obtain one-dimensional range image data of the radar; performing a two-dimensional Fourier transform on the one-dimensional range image data to obtain two-dimensional frequency domain data; dividing the target signal into a preset number of detection windows according to the result of accumulating the two-dimensional frequency domain data, obtaining a background noise signal through the neighborhood area of ​​the detection window, and obtaining a detection threshold according to the background noise signal; judging whether the target signal to be detected in the detection window exceeds the detection threshold, if the target signal to be detected exceeds the detection threshold, it is determined to be a target signal, otherwise it is determined to be noise.

[0011] Furthermore, the specific method of obtaining the point cloud detection result is as follows: input the radar data into the radar coordinate system, and convert the radar coordinate system into the drone coordinate system, and input the target point cloud coordinates into the drone coordinate system to obtain the point cloud detection result.

[0012] Furthermore, the specific method for obtaining the association result of the point cloud and the track is as follows: compare the point cloud coordinates and radial velocity with the coordinates and radial velocity of each track, and determine whether the first coordinate difference, the second coordinate difference, the third coordinate difference and the speed difference are not greater than the preset difference threshold; if the first coordinate difference, the second coordinate difference, the third coordinate difference and the speed difference are not greater than the preset difference threshold, it indicates that the point cloud is successfully associated with the track, otherwise it indicates that the association fails; the first coordinate difference represents the difference between the horizontal coordinate of the point cloud and the horizontal coordinate of the track; the second coordinate difference represents the difference between the vertical coordinate of the point cloud and the vertical coordinate of the track; the third coordinate difference represents the difference between the vertical coordinate of the point cloud and the vertical coordinate of the track; the speed difference represents the difference between the radial velocity and the radial velocity of the track.

[0013] Furthermore, the specific process of judging whether to update the target data and maintain the track based on the association result of the point cloud and the track is as follows: judging whether the point cloud and the track are successfully associated; if the point cloud and the track are successfully associated, updating the target data and maintaining the track according to the system state vector; otherwise judging whether the cumulative number of frames reaches the preset frame number requirement; if the cumulative number of frames reaches the preset frame number requirement, updating the target data and maintaining the track according to the system state vector; otherwise only saving the historical data; combining each tracked track to obtain the mean of the associated target data, and obtaining the system state vector, the update error covariance and the update prediction error covariance according to the measurement error between the tracked track state estimate and the target data mean.

[0014] Furthermore, the updating of target data and maintaining the track also includes: B1, associating the point cloud and the target bounding box, judging whether the target point cloud coordinates and radial velocity are within the association threshold, if the target point cloud coordinates and radial velocity are not within the association threshold, performing a deletion operation, otherwise retaining the tracking track; B2, judging whether there is data associated with the radar ID and measurement ID associated with the previous frame in the point cloud data of the current frame, if so, updating the target data according to the point cloud data of the current frame, otherwise executing B3; B3, searching for obstacle targets to be detected within the association threshold, judging whether there is an associated measurement, if there is no associated measurement, executing B4, otherwise judging whether there is a unique associated measurement, if there is a unique associated measurement, updating the target data according to the associated measurement, otherwise comparing the point cloud data of the current frame with the associated measurement to obtain a measurement score, and updating the target data according to the measurement score corresponding to the minimum measurement score obtained by arranging the measurement scores in descending order; B4, adjusting the radial velocity association threshold step by step, re-searching for obstacle targets to be detected within the association threshold, and judging whether Whether there is an associated measurement. If no associated measurement exists, execute B5. Otherwise, determine whether there is a unique associated measurement. If there is a unique associated measurement, update the target data according to the associated measurement. Otherwise, compare the point cloud data of the current frame with the associated measurement to obtain a measurement score. Update the target data according to the measurement score corresponding to the minimum measurement score obtained by arranging the measurement scores in descending order. B5, adjust the CFAR energy threshold step by step and search for targets on the range Doppler heat map to determine whether there is an associated measurement. If there is no associated measurement, update the fusion target through the system state vector and accumulate the measurement loss time. If the measurement loss time exceeds the preset time, delete the obstacle target to be detected. Otherwise, determine whether there is a unique associated measurement. If there is a unique associated measurement, update the target data according to the associated measurement. Otherwise, compare the point cloud data of the current frame with the associated measurement to obtain the degree of association. Arrange the degree of association in ascending order to obtain the maximum degree of association. Update the target data according to the measurement corresponding to the maximum degree of association. The measurement loss time indicates the number of frames in which the radar data is not successfully associated.

[0015] Furthermore, the specific method for obtaining the multi-dimensional omnidirectional track point data is as follows: E1, receiving and storing radar data within a preset time period, judging whether the stored radar data exceeds the queue length, if the stored radar data exceeds the queue length, deleting the queue head data and storing the unstored radar data at the queue tail, otherwise storing the radar data and executing E2; E2, splicing the time-continuous track point data to obtain omnidirectional detection data, the time-continuous track point data represents the track point data of continuous frames in the time dimension; E3, splicing the spatially continuous track point data to obtain overlapping parts, transforming the track point data into the track coordinate system according to the radar illumination angle and position to obtain multi-dimensional detection data, the spatially continuous track point data represents the track point data of continuous frames in the spatial dimension; E4, integrating the track point data according to the omnidirectional detection data and the multi-dimensional detection data to obtain multi-dimensional omnidirectional track point data.

[0016] Furthermore, the specific method for obtaining the radar output distance is as follows: obtain the current flight direction of the drone, and at the same time obtain the corresponding unique identifier based on the track point in the track point coordinate system, the unique identifier including the ID of the track point and the corresponding radar number; determine whether there is a track point in the flight direction of the drone, if there is no track in the flight direction of the drone, output the track point as the first identifier, otherwise output the track point as the second identifier and output the radar output distance and speed information.

[0017] Furthermore, the method of obtaining the current flight direction of the drone also includes: obtaining the inclination data of the drone in various directions at the current moment by reading the posture information of the drone when flying in the air; compiling the inclination data into an electronic signal by a compiler in the drone flight control system, and transmitting it to the single-chip microcomputer in the drone flight control system in real time; obtaining a compensation angle according to the inclination data in various directions of the drone at the current moment, and performing track point coordinate system compensation processing on the drone according to the obtained compensation angle.

[0018] Furthermore, the specific process of the autonomous obstacle avoidance is as follows: the driving motor speed of each blade of the drone is controlled according to the radar output distance, and it is determined whether the radar output distance is less than the preset safety distance. If the radar output distance is less than the preset safety distance, the flight action is interrupted and the brakes are applied. At the same time, the radar output distance and relative radar movement speed are output in real time through the drone ground station display, otherwise no processing is performed.

[0019] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0020] 1. By compressing the original radar data to obtain two-dimensional frequency domain data, the point cloud detection results are obtained based on this data. Then, based on the point cloud detection results and the track starting conditions, whether to track the target is determined. Finally, the radar data is spliced ​​to obtain multi-dimensional omnidirectional track point data. Based on this, the radar output distance is obtained. Based on the radar output distance and the preset safety distance, whether to perform autonomous obstacle avoidance is determined. This allows for more accurate target identification and detection, thereby achieving more accurate multi-radar omnidirectional detection, effectively solving the problem in existing technologies where multi-radar omnidirectional detection results are significantly affected by the environment.

[0021] 2. By receiving and storing radar data within a preset time period, the time-continuous track point data is then spliced ​​to obtain omnidirectional detection data. The spatially continuous track point data is then spliced ​​to obtain overlapping parts, and the data is transformed into a track coordinate system according to the radar illumination angle and position to obtain multi-dimensional detection data. Finally, the track point data is compared with the omnidirectional detection data and the multi-dimensional detection data to obtain multi-dimensional omnidirectional track point data, thereby more comprehensively describing the target's motion state and thus achieving multi-dimensional omnidirectional detection.

[0022] 3. By reading the attitude information of the UAV when it is flying in the air, the inclination data of the UAV in all directions at the current moment is obtained, the inclination data is compiled into an electronic signal, and transmitted to the single-chip microcomputer in the UAV flight control system in real time. The compensation angle is obtained according to the inclination data of the UAV in all directions at the current moment, and the track point coordinate system of the UAV is compensated according to the obtained compensation angle, so as to accurately obtain the inclination data of the UAV, thereby achieving more accurate radar angle measurement and ranging. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a multi-radar omnidirectional detection method for a UAV inertial navigation system provided in an embodiment of the present application;

[0024] Figure 2 A schematic diagram of coordinate changes for converting a radar coordinate system into a drone coordinate system provided in an embodiment of the present application;

[0025] Figure 3 A flowchart of the specific implementation steps of a multi-radar omnidirectional detection method for an unmanned aerial vehicle inertial navigation system provided in an embodiment of the present application;

[0026] Figure 4 This is a flowchart of the autonomous obstacle avoidance operation of the drone system provided in the embodiment of the present application. DETAILED DESCRIPTION

[0027] The embodiment of the present application solves the problem in the prior art that the results of multi-radar omnidirectional detection are greatly affected by the environment by providing a multi-radar omnidirectional detection method for an unmanned aerial vehicle inertial navigation system. By compressing the original radar data to obtain two-dimensional frequency domain data, a point cloud detection result is obtained based on the data. Then, based on the point cloud detection result and the track starting condition, it is determined whether to track the target. Finally, the radar data is spliced ​​to obtain multi-dimensional omnidirectional track point data, based on which the radar output distance is obtained. Based on the data, it is determined whether to perform autonomous obstacle avoidance based on the radar output distance and the preset safety distance, thereby achieving more accurate multi-radar omnidirectional detection.

[0028] The technical solution in the embodiment of the present application is to solve the problem that the multi-radar omnidirectional detection results are greatly affected by the environment. The overall idea is as follows:

[0029] The point cloud detection results and the track starting conditions are used to determine whether to track the target. At the same time, the radar output distance and the preset safety distance are used to determine whether to perform autonomous obstacle avoidance, achieving more accurate multi-radar omnidirectional detection.

[0030] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0031] like Figure 1 As shown, it is a flow chart of a multi-radar omnidirectional detection method of a UAV inertial navigation system provided by an embodiment of the present application, and the method includes the following steps: S1, compressing the collected radar raw data to obtain two-dimensional frequency domain data, and accumulating the two-dimensional frequency domain data and performing constant false alarm rate detection to obtain a point cloud detection result, the compression includes one-dimensional compression and two-dimensional compression, and the obstacle target to be detected represents the obstacle in the flight direction of the UAV; S2, classifying the point cloud to obtain a point cloud clustering result, obtaining target data according to the point cloud clustering result, and classifying the obstacle target to be detected according to the target data to obtain a target classification result, the point cloud represents the spatial position of the millimeter wave radar target, and the target data includes the target position, the number of point clouds, the length, the width, Height, volume, projected area, radar cross section and signal-to-noise ratio. The obstacle target to be detected represents the obstacle in the flight direction of the UAV; S3, based on the point cloud detection results and the track starting conditions, it is determined whether to track the obstacle target to be detected, and based on the association result of the point cloud and the track, it is determined whether to update the target data and maintain the track. If the point cloud and the track are not successfully associated, only the historical data is saved; S4, the radar data is spliced ​​to obtain multi-dimensional omnidirectional track point data, and the radar output distance is obtained in combination with the flight direction of the UAV. Based on the radar output distance and the preset safety distance, it is determined whether to perform autonomous obstacle avoidance. The radar data includes data from the front radar, rear radar, left radar, right radar, upper radar and lower radar. The radar output distance represents the distance to the nearest track point of the UAV.

[0032] In this embodiment, in radar signal processing, compression refers to converting the original echo signal into a frequency domain signal through Fourier transform technology, reducing the amount of data and improving signal processing efficiency. One-dimensional compression refers to extracting the distance information of the obstacle target to be detected through one-dimensional Fourier transform. Two-dimensional compression refers to converting the two-dimensional signal from the spatial domain to the frequency domain through two-dimensional Fourier transform to extract two-dimensional frequency domain data. The direction of arrival estimation algorithm is used to obtain the target point cloud coordinates based on the phase difference between the obstacle target to be detected and the two antennas. The target point cloud coordinates are mapped to the drone coordinate system to obtain the point cloud detection result. Constant False Alarm Rate detection (CFAR) refers to a radar signal processing technology that uses adaptive threshold technology to maintain a constant false alarm rate, thereby effectively detecting targets in complex environments. Radar Cross Section (RCS) is used to reflect the target's ability to reflect radar signals. The larger the radar cross section, the easier it is for the target to be detected by the radar. Signal-to-Noise Ratio (SNR) SNR) represents the signal-to-noise ratio. The higher the SNR, the clearer the signal. The track represents a trajectory representing the target's movement path formed by the radar through continuous tracking of the target's position. The target classification result represents the type of obstacle target to be detected, such as trees, rocks, etc. The track starting conditions include the current track cumulative tracking times being less than the preset tracking times, the point cloud detection results having at least a preset number of millimeter-wave point clouds, the point cloud detection results having a height greater than the preset height, and the point cloud detection results having a maximum radar cross-section greater than the preset radar cross-section. The preset tracking times, the preset number of millimeter-wave point clouds, and the preset height are all required. The preset radar cross section is obtained from the preset database, the preset tracking times are set to 10 times, the preset number of millimeter-wave point clouds are set to 3 point clouds, the preset height is set to 1.5m, and the preset radar cross section is set to -5dB; through signal compression, accumulation and CFAR detection, more accurate target identification and detection are achieved, through point cloud classification and track tracking, effective identification of different targets and continuous tracking of motion trajectories are achieved, and through the fusion of multi-radar data and safe distance judgment, more accurate multi-radar omnidirectional detection is ensured to avoid collision risks, thereby achieving more accurate multi-radar omnidirectional detection.

[0033] Specifically, the Direction of Arrival (DOA) estimation algorithm is typically used to estimate the direction of a signal source relative to a receiver. To perform angle estimation, at least two receiving antennas are required. The distance between the antennas is usually known and is called the array spacing. The spacing between the two antennas is generally a fraction of the signal wavelength to ensure appropriate spatial resolution. Then, the signal paths from the obstacle to be detected to the two antennas are different, resulting in a phase difference between the signals received by the two antennas. The phase difference and the signal wavelength are then used to calculate the differential distance between the two antennas. Finally, based on the geometric relationship, the differential distance and the antenna spacing are used to calculate the angle of arrival of the signal. The differential distance is calculated using the following formula:

[0034] ;

[0035] Where d is the differential distance between the two antennas, is the phase difference between the signals received by the two antennas, is the wavelength of the signal received by the antenna; the angle of arrival is calculated using the following formula:

[0036] ;

[0037] Where, is the angle of arrival of the signal, and D is the distance between the two antennas.

[0038] Specifically, classification is performed using a clustering method (DBSCAN), starting with a starting data point. The neighborhood of the starting data point is extracted using a distance d. Neighboring points represent points at a distance of d. If the number of points in the neighborhood exceeds the standard, which is set to 3 points, clustering begins, and the current data point becomes the first point in the new cluster and is marked as a core point. If the number of points in the neighborhood is less than the threshold, the point is marked as a noise point (which can be covered by other clusters later). For the first point in the new cluster, its points at a distance of d will also become part of the same cluster. This process makes all points in the neighborhood of d belong to the same cluster, and then repeats for all new points that have just been added to the cluster group. Repeat the above steps until all points near the cluster have been visited and marked.

[0039] like Figure 3As shown in FIG, a flowchart of the specific implementation steps of a multi-radar omnidirectional detection method for an unmanned aerial vehicle inertial navigation system provided by an embodiment of the present application, step 1: configuring the millimeter wave radar waveform, and collecting radar raw data by transmitting and receiving electromagnetic waves through the antenna; step 2: 1dfft (one-dimensional Fourier transform), performing 1D-FFT on the radar raw data to complete one-dimensional compression; step 3: 2dfft (two-dimensional Fourier transform), performing 2D-FFT on the radar raw data to complete two-dimensional compression; step 4: cfar (constant false alarm rate detection), performing CFAR processing on the accumulated amplitude matrix; step Step 5: DOA angle solution (direction of arrival estimation algorithm), estimate the angle based on the phase difference between the target and the two antennas, and output the target point cloud coordinates; Step 6: Coordinate conversion, convert the radar coordinate system with its own coordinate system; Step 7: Target classification, classify the target and obtain the target classification result; Step 8: Track initiation; Step 9: Track maintenance, if the track is successfully associated with the point cloud, update the target attributes and maintain the track; Step 10: Multi-radar fusion processing, splice the data of multiple radars to obtain the multi-radar fusion track points; Step 11: Realize multi-target omnidirectional detection.

[0040] Furthermore, the specific process of constant false alarm rate detection is as follows: the radar original data is subjected to one-dimensional Fourier transform to obtain the radar's one-dimensional range image data. The radar original data is A matrix of , where N is the number of sampling points under a single chirp, M is the number of chirps corresponding to the channel in a single frame, and K is the number of antenna channels. The one-dimensional Fourier transform represents the conversion along the time axis, and the one-dimensional range image data represents the distance of the target from the radar in the frequency domain. The one-dimensional range image data is subjected to a two-dimensional Fourier transform to obtain two-dimensional frequency domain data. The two-dimensional Fourier transform represents the conversion along the distance and speed directions. The two-dimensional frequency domain data is used to analyze the distribution of the target in the distance and speed directions. According to the result of the accumulation of the two-dimensional frequency domain data, the target signal is divided into a preset number of detection windows. The background noise signal is obtained by the neighborhood area of ​​the detection window. The detection threshold is obtained according to the background noise signal. The neighborhood area includes an inner window and an outer window. The inner window includes the target to be detected, and the outer window is used to evaluate the background noise signal. It is judged whether the target signal to be detected in the detection window exceeds the detection threshold. If the target signal to be detected exceeds the detection threshold, it is determined to be a target signal, otherwise it is determined to be noise.

[0041] In this embodiment, the radar raw data represents the echo signal received by the radar, which is usually represented in matrix form, where N is the number of sampling points under a single frequency modulation signal (chirp), that is, the number of times the echo signal is sampled during an FM signal period, M is the number of FM signals per frame, and a frame contains multiple FM signals, and K is the number of antenna channels, indicating the number of antennas; a one-dimensional Fourier transform represents a Fourier transform along the time axis (i.e., the chirp sampling points) to obtain a one-dimensional range profile data; a one-dimensional Fourier transform converts the radar's raw time domain signal into the frequency domain, reflecting the distance of the target relative to the radar; the data after the one-dimensional Fourier transform represents the target distance information, because there is a linear relationship between the target distance and the frequency; the one-dimensional range profile data represents the distance of different targets from the radar, which is obtained through the relationship between frequency and distance; a two-dimensional Fourier transform is performed on the one-dimensional range profile data (transformed along the two dimensions of distance and speed); this step converts the data from the distance-time domain to the distance-velocity domain, with the purpose of simultaneously analyzing the distance and speed of the target ; The obtained two-dimensional frequency domain data contains the distance and relative speed information of the target, which is used to locate the motion state of the target; the obtained two-dimensional frequency domain data is accumulated and processed to enhance the signal strength, making the target more obvious and facilitating subsequent detection; the accumulated data is divided into a preset number of detection windows, each window containing a target signal area; based on the background noise data in the outer window, the threshold value for target detection is calculated, and the CFAR algorithm sets the threshold value by dynamically estimating the background noise so that the false alarm rate remains constant; the preset number is obtained from a preset database, usually set according to known radar resolution and data characteristics; the inner window includes the obstacle target signal to be detected, which is the area that the radar hopes to detect; the outer window is used to evaluate the background noise signal and does not contain the target signal; the data of the outer window is used to adaptively estimate the background noise in the detection window, which is suitable for processing the fluctuation of background noise in different scenarios; the false alarm rate represents the probability of a false alarm, that is, the probability of mistaking noise for a target; through the above steps, the target detection performance of the radar is improved.

[0042] Furthermore, the specific method for obtaining the point cloud detection results is as follows: input the radar data into the radar coordinate system, and then convert the radar coordinate system into the drone coordinate system. Then, input the target point cloud coordinates into the drone coordinate system to obtain the point cloud detection results. The point cloud detection results are calculated using the following formula: ;

[0043] Where, is the point cloud detection result, is the x-axis of the drone coordinate system, is the y-axis of the drone coordinate system, is the z-axis of the drone coordinate system, is the x-axis of the radar coordinate system, is the y-axis of the radar coordinate system, is the z-axis of the radar coordinate system, is the counterclockwise rotation angle of the radar coordinate system around the z-axis, is the horizontal coordinate of the radar relative to the UAV coordinate position, is the vertical coordinate of the radar relative to the UAV coordinate position, is the height coordinate of the radar relative to the drone's coordinate position.

[0044] In this embodiment, if Figure 2 As shown in FIG, a schematic diagram of coordinate changes for converting a radar coordinate system into a drone coordinate system is provided in an embodiment of the present application. is the x-axis of the drone coordinate system; is the y-axis of the drone coordinate system; is the z-axis of the drone coordinate system; is the x-axis of the radar coordinate system; is the y-axis of the radar coordinate system; is the z-axis of the radar coordinate system.

[0045] The radar coordinate system is usually centered on the radar itself, and the data (point cloud) acquired by the radar is relative to the radar coordinate system; the drone coordinate system represents the coordinate system of the drone itself, usually with the center of the drone as the origin; the relative position of the drone and the radar requires consideration of coordinate transformation to convert the radar data into the drone coordinate system; the point cloud represents a set of discrete points in three-dimensional space collected by the radar, usually used to describe the surface shape and structure of an object. Each discrete point has its own coordinate value, indicating its position in space; the rotation angle represents the rotation angle of the radar around the z-axis relative to the drone coordinate system, usually indicating the relative attitude of the radar and the drone, for example, the direction of the radar relative to the drone; through the above steps, it is ensured that the drone can understand the surrounding environment based on its own perspective.

[0046] Specifically, the rotation matrix for counterclockwise rotation around the Z axis is defined as:

[0047] ;

[0048] Where Re is the rotation matrix of counterclockwise rotation around the Z axis, is the counterclockwise rotation angle around the z-axis; Re is used to represent the rotation of the radar relative to the drone coordinate system;

[0049] The radar coordinate system is converted to the drone coordinate system using the following formula:

[0050] ;

[0051] Where, is the drone coordinate system, is the x-axis of the drone coordinate system, is the y-axis of the drone coordinate system, is the z-axis of the drone coordinate system, is the radar coordinate system, is the x-axis of the radar coordinate system, is the y-axis of the radar coordinate system, is the z-axis of the radar coordinate system.

[0052] Furthermore, the specific method for obtaining the association result of the point cloud and the track is as follows: compare the point cloud coordinates and radial velocity with the coordinates and radial velocity of each track, and determine whether the first coordinate difference, the second coordinate difference, the third coordinate difference and the speed difference are not greater than the preset difference threshold. If the first coordinate difference, the second coordinate difference, the third coordinate difference and the speed difference are not greater than the preset difference threshold, it indicates that the point cloud is successfully associated with the track, otherwise it indicates that the association fails; the first coordinate difference represents the difference between the horizontal coordinate of the point cloud and the horizontal coordinate of the track; the second coordinate difference represents the difference between the vertical coordinate of the point cloud and the vertical coordinate of the track; the third coordinate difference represents the difference between the vertical coordinate of the point cloud and the vertical coordinate of the track; the speed difference represents the difference between the radial velocity and the radial velocity of the track.

[0053] In this embodiment, radial velocity refers to the velocity component of an object in the radial direction relative to the radar, that is, the velocity of the object toward or away from the observation point; the preset difference threshold is obtained from a preset database and is represented by the threshold of the track coordinates and radial velocity that meet the association conditions. For example, the preset difference threshold of the first coordinate difference is represented by the threshold on the horizontal coordinate of the track that meets the association conditions; if all differences are within the preset difference threshold, the point cloud is successfully associated with the track, otherwise the point cloud and the track are not associated; through the above steps, real-time tracking of multiple targets is achieved.

[0054] It should be understood that the first coordinate difference, the second coordinate difference, the third coordinate difference and the speed difference are not greater than the preset difference threshold, which can be expressed as:

[0055] ;

[0056] ;

[0057] ;

[0058] ;

[0059] Where, Indicates the horizontal coordinate corresponding to the currently traversed point cloud, Indicates the horizontal coordinate of the track associated with the currently traversed point cloud. Indicates the threshold value on the x-coordinate of the track that meets the association condition, Indicates the vertical coordinate corresponding to the currently traversed point cloud, Indicates the vertical coordinate corresponding to the track associated with the currently traversed point cloud. Indicates the threshold value on the vertical coordinate of the track that meets the association condition, Indicates the height coordinate corresponding to the currently traversed point cloud, Indicates the height coordinate corresponding to the track associated with the currently traversed point cloud. Indicates the threshold value on the track height coordinate that meets the association condition, Indicates the radial velocity corresponding to the currently traversed point cloud, Indicates the radial velocity corresponding to the track associated with the currently traversed point cloud. Indicates the threshold value of the track radial velocity that meets the association condition.

[0060] Furthermore, the specific process of judging whether to update the target data and maintain the track based on the association result of the point cloud and the track is as follows: judging whether the point cloud and the track are successfully associated; if so, updating the target data and maintaining the track according to the system state vector; otherwise, judging whether the cumulative number of frames reaches the preset frame number requirement; if so, updating the target data and maintaining the track according to the system state vector; otherwise, only saving the historical data, where the number of frames indicates that the radar completes one transmission and reception of electromagnetic waves; obtaining the mean of the associated target data in combination with each tracking track; obtaining the system state vector, updating the error covariance, and updating the posterior prediction error covariance according to the measurement error between the tracking track state estimate and the target data mean; and calculating the system state vector using the following formula:

[0061] ;

[0062] Where n is the frame number, , N is the total number of frames, is the target data mean of the nth frame, is the state estimation vector, is the prior state corresponding to the tracking target of the nth frame, is the system state vector of the nth frame, is the Kalman gain of the nth frame; the updated error covariance is calculated by the following formula:

[0063] ;

[0064] Where, is the update error covariance of the nth frame, is the observation variance of the nth frame, is the prediction error covariance of the nth frame, The observation matrix of Kalman tracking, is the transpose of the observation matrix of Kalman tracking; the updated posterior prediction error covariance is calculated by the following formula:

[0065] ;

[0066] ;

[0067] Where, is a matrix The inverse matrix of is the updated posterior prediction error covariance of the nth frame.

[0068] In this embodiment, in the Kalman filter, the system state vector represents the predicted state of the target, such as position, speed, etc.; the error covariance represents the size of the system state estimation error, which measures the uncertainty of the current system state estimate; the Kalman gain is used to combine the predicted state and the measured state to update the system state in an optimal way; successful association indicates that a point in the point cloud matches the target trajectory that the system has tracked; the radar system updates these attributes through the data of each frame to make the trajectory and state of the target more accurate; it is determined whether the cumulative number of frames meets the preset frame number requirement. If the cumulative number of frames does not meet the requirement, only historical data is saved; historical data includes unassociated point cloud data and track information; through the above steps, it can be ensured that the radar system can accurately track targets in multi-target scenarios and reduce the occurrence of false alarms and missed detections.

[0069] Furthermore, the target data is updated and the track is maintained, which also includes: B1, associating the point cloud with the target bounding box, judging whether the target point cloud coordinates and radial velocity are within the associated threshold, if the target point cloud coordinates and radial velocity are not within the associated threshold, then performing a deletion operation, otherwise retaining the tracking track, the associated threshold includes the target point cloud horizontal coordinate threshold, the target point cloud vertical coordinate threshold, the target point cloud vertical coordinate threshold and the radial velocity associated threshold; B2, judging whether there is data associated with the radar ID and measurement ID associated with the previous frame in the point cloud data of the current frame, if so, then according to the current frame Update the target data with the point cloud data of the frame, otherwise execute B3; B3, search for the obstacle target to be detected within the association threshold, determine whether there is an associated measurement, if there is no associated measurement, execute B4, otherwise determine whether there is a unique associated measurement, if there is a unique associated measurement, update the target data according to the associated measurement, otherwise compare the point cloud data of the current frame with the associated measurement to obtain a measurement score, and update the target data according to the measurement score corresponding to the minimum measurement score obtained by arranging the measurement scores in descending order; B4, adjust the radial velocity association threshold step by step, and re-search for the obstacle to be detected within the association threshold Target, determine whether there is an associated measurement. If there is no associated measurement, execute B5. Otherwise, determine whether there is a unique associated measurement. If there is a unique associated measurement, update the target data according to the associated measurement. Otherwise, compare the point cloud data of the current frame with the associated measurement to obtain a measurement score. Update the target data according to the measurement score corresponding to the minimum measurement score obtained by arranging the measurement scores in descending order. The measurement score represents the coordinate difference and radial velocity difference between the point cloud data of the current frame and the associated measurement. B5, gradually adjust the CFAR energy threshold and perform target search on the range Doppler heat map to determine whether it is Whether there is an associated measurement. If there is no associated measurement, the fusion target is updated through the system state vector and the measurement loss time is accumulated. If the measurement loss time exceeds the preset time, the obstacle target to be detected is deleted. Otherwise, it is determined whether there is a unique associated measurement. If there is a unique associated measurement, the target data is updated according to the associated measurement. Otherwise, the point cloud data of the current frame is compared with the associated measurement to obtain the degree of association. The degree of association is arranged in ascending order to obtain the maximum degree of association. The target data is updated according to the measurement corresponding to the maximum degree of association. The measurement loss time indicates the number of frames in which the radar data has not been successfully associated.

[0070] In this embodiment, each tracking track has its own life cycle. During the maintenance process, we may change the state or delete the tracking track that is no longer in use; life cycle management mainly handles the generation and extinction of targets; when the track is generated, it is associated with the target bounding box four times according to the coordinates and radial velocity of the point cloud; if the coordinates and velocity of the point cloud data are within the set association threshold, the tracking track of the target will continue to be retained, if it exceeds the association threshold, the target will be deleted; the measurement ID represents the unique identifier of a single measurement data obtained from the point cloud, which is usually used to identify different measurement results when tracking multiple targets; associated measurement means that by comparing the data of the current frame with the previous frame, it is found that some measurement values ​​or targets are associated with the previously detected targets; in the target In target detection, CFAR is an adaptive threshold algorithm used to control the false alarm rate. It dynamically adjusts the threshold to adapt to changes in background noise, thereby more reliably detecting targets. The CFAR energy threshold is used to determine whether certain areas contain target values. The range Doppler heat map represents a two-dimensional image that combines distance and speed information, and is usually used for radar signal processing and target detection. On the range Doppler heat map, the distance and speed information of the target are both displayed to help determine the target's position and motion status. The loss time indicates the number of frames in which the radar data is not successfully associated. If the radar data is not successfully associated within multiple frames, it is usually considered that the target has been lost and deleted. Through the above steps, the drone can accurately track and update the motion trajectory and attributes of the target in different environments.

[0071] Furthermore, the specific method for obtaining multi-dimensional omnidirectional track point data is as follows: E1, receiving and storing radar data within a preset time period, judging whether the stored radar data exceeds the queue length, if the stored radar data exceeds the queue length, deleting the queue head data and storing the unstored radar data to the queue tail, otherwise storing the radar data and executing E2, the queue head data represents the earliest received and stored data; E2, splicing the time-continuous track point data to obtain the omnidirectional detection data, the time-continuous track point data represents the track point data of continuous frames in the time dimension, splicing The connection includes data alignment, merging and deduplication. The omnidirectional detection data includes track information within a continuous time period; E3, splicing the spatially continuous track point data to obtain the overlapping part, and transforming the track point data into the track coordinate system according to the radar illumination angle and position to obtain multi-dimensional detection data. The spatially continuous track point data represents the track point data of continuous frames in the spatial dimension; E4, integrating the track point data according to the omnidirectional detection data and the multi-dimensional detection data to obtain multi-dimensional omnidirectional track point data. The multi-dimensional omnidirectional track point data is used to describe the track point data under different radar observations.

[0072] In this embodiment, the queue length represents a pre-set upper limit for storing radar data. When the received radar data exceeds this capacity, in order to ensure the real-time and up-to-dateness of the data, the system needs to delete the earliest received and stored data (i.e., the queue head data) to make room for new data. If there is no latest data, the fusion target is not updated. The radar data includes the data of the front radar, rear radar, left radar, right radar, upper radar, and lower radar. The latest data represents the radar echo signal data currently received. The omnidirectional detection data represents a data set obtained by splicing the time-continuous track point data. The omnidirectional detection data covers the target's track information in a continuous time period and can fully reflect the target's motion trajectory in the time dimension. The multidimensional detection data represents the data set obtained by splicing the time-continuous track point data. The data set is obtained by converting the continuous track point data; the multi-dimensional detection data not only contains the target's position information in space, but also takes into account the radar's illumination angle and position, so that it can more accurately describe the target's multi-dimensional characteristics under different radar observations; the track coordinate system represents the reference coordinate system used to describe the target's motion trajectory; in the track coordinate system, the target's position, speed, acceleration and other parameters can be more conveniently described and analyzed; if some track points do not meet the fusion index requirements, the track points can be filtered according to the status of movement, distance, azimuth and other states, for example, setting moving target detection (filtering out static targets), setting the distance range (such as 1m-20m) and setting the azimuth range (such as -10m to 10m); through the above steps, a comprehensive description of the target's motion state is achieved.

[0073] Specifically, the specific process of splicing time-continuous track point data is as follows: F1, synchronize the radar data to ensure that all data are in the same time frame; F2, convert the track point data from different data sources into a unified coordinate system for spatial alignment; F3, segment the time-continuous track point data in chronological order, and for adjacent track segments, use the nearest neighbor matching algorithm to find the best matching point between adjacent track segments; after finding the matching point, merge the adjacent track segments; during the splicing process, there may be overlapping parts between track segments to eliminate duplicate parts and ensure seamless splicing.

[0074] Specifically, the specific process of splicing spatially continuous track point data is as follows: H1, for each pair of tracks, the Euclidean distance calculation method can be used to obtain the minimum distance between their respective track points; H2, if the distance between two track points is less than a preset distance threshold, they are considered to be overlapping; H3, based on the distance calculation result, all possible overlapping track segments are marked; H4, for the identified overlapping track segments, a splicing strategy (such as direct connection, interpolation connection, smooth connection, etc.) is selected to merge the track points. In this embodiment, smooth connection is selected to merge the track points, which can effectively filter noise and generate a smooth and coherent trajectory. The effect is particularly significant when there is noise in the track points. The trajectory after smooth connection has good stability and accuracy.

[0075] Furthermore, the specific method of obtaining the radar output distance is as follows: obtain the current flight direction of the UAV, and at the same time obtain the corresponding unique identifier based on the track point in the track point coordinate system, the unique identifier includes the ID of the track point and the corresponding radar number; determine whether there is a track point in the flight direction of the UAV, if there is no track in the flight direction of the UAV, the track point is output as the first identifier, otherwise the track point is output as the second identifier and the radar output distance and speed information is output, the first identifier is blue and the second identifier is red.

[0076] In this embodiment, the drone flight direction represents the direction in which the drone is currently moving, which is a vector relative to the track point coordinate system; the drone flight direction is obtained through the drone flight control system inertial measurement unit (IMU); the unique identifier represents a special mark used to identify a track point; the unique identifier includes the track point ID (unique number) and the radar number (indicating which radar detected the point) to ensure that each track point can be uniquely identified in the system; the radar output distance represents the distance from the drone to the obstacle target to be monitored measured by the radar system; the speed information represents the speed of the drone measured by the radar system; through the above steps, accurate tracking of the drone is achieved.

[0077] Furthermore, obtaining the current flight direction of the drone also includes: reading the attitude information of the drone when flying in the air through the drone flight control system to obtain the inclination data of the drone in various directions at the current moment; compiling the inclination data into an electronic signal through the compiler in the drone flight control system, and transmitting it to the single-chip microcomputer in the drone flight control system in real time, and the single-chip microcomputer is used for electronic signal calculation; obtaining the compensation angle according to the inclination data of the drone in various directions at the current moment, and performing track point coordinate system compensation processing on the drone according to the obtained compensation angle, and the track point coordinate system compensation processing means that the drone system dynamically compensates the tilt angle of the drone with an internal loop of 20HZ.

[0078] In this embodiment, the tilt angle is the tilt angle of the drone relative to the horizontal plane; the tilt angle data includes the pitch angle (Pitch), roll angle (Roll) and yaw angle (Yaw); the tilt angle data reflects the attitude changes of the drone in various directions during flight; the pitch angle (Pitch) indicates the tilt of the drone in the front and back directions; the roll angle (Roll) indicates the tilt of the drone in the left and right directions; the yaw angle (Yaw) indicates the rotation angle of the drone around the vertical axis, which affects the direction of the drone; the electronic signal means that after the drone flight control system obtains the tilt angle data through the sensor, it converts this data into electronic signals, which are calculated and processed by the single-chip microcomputer; in the drone, the single-chip microcomputer The computer processes the attitude information received from the sensor and adjusts the flight attitude based on this data. During flight, the attitude of the drone may deviate due to external influences (wind, airflow, etc.). The compensation angle is the angle calculated to correct the deviation, which is used to ensure that the drone can fly in the set attitude and direction. The track point coordinate system compensation corrects the flight attitude of the drone to make its position and direction in three-dimensional space more accurate. The inner loop represents the update frequency of the drone flight control system. The 20Hz inner loop means that the system will update and compensate the attitude 20 times per second. Through the above steps, the drone can cope with various flight environments and maintain a stable flight attitude.

[0079] Furthermore, the specific process of autonomous obstacle avoidance is as follows: the speed of the driving motor of each blade of the drone is controlled according to the radar output distance, and it is judged whether the radar output distance is less than the preset safety distance. If the radar output distance is less than the preset safety distance, the flight action is interrupted and the brakes are applied. At the same time, the radar output distance and relative radar movement speed are output in real time through the drone ground station display, otherwise no processing is performed.

[0080] In this embodiment, the radar represents a sensor installed on the drone for detecting the surrounding environment; the drive motor represents the power source of each blade on the drone, and the flight attitude and speed of the drone are adjusted by controlling the speed of the motor; the preset safety distance represents a minimum safety distance value set to ensure that the drone does not collide with obstacles during flight; the preset safety distance is obtained from a preset database, and the safety distance is dynamically adjusted according to the real-time environmental information and system status. For example, when trees are detected in front of the flight path, the safety distance is automatically shortened and the flight trajectory is adjusted; during the flight of the drone, braking means slowing down or stopping the flight of the drone by reducing or stopping the speed of the drive motor to avoid collision with obstacles; the drone ground station display represents the equipment used by ground operators to monitor and control the drone, which can display various status information of the drone in real time, such as the distance output by the radar, relative movement speed, etc.; through the above steps, the flight safety and stability of the drone are improved.

[0081] like Figure 4 As shown, this is a flowchart of the autonomous obstacle avoidance operation of the drone system provided in an embodiment of the present application, step 1: obtaining multi-radar data, the system receives data from the front radar, rear radar, left radar, right radar, upper radar and lower radar; step 2: data caching, the cache stores multi-radar data; step 3: data splicing, the data splicing is divided into two dimensions, one is the time dimension and the other is the space dimension; step 4: dynamic compensation, the obtained compensation angle is transmitted to the radar to complete the spatial coordinate system compensation processing; step 5: determine whether there is a track ID in the flight direction of the drone. If the track point is in the flight direction, the track point is output in red, and the distance and speed information of the nearest track point to the drone are output; otherwise, when the track point is not in the flight direction, the track point is output in blue; step 6: the drone system performs autonomous obstacle avoidance, interrupts the flight action when the detected distance is less than the preset safety distance, and brakes.

[0082] To summarize, the embodiment of the present application compresses the radar raw data to obtain two-dimensional frequency domain data, and obtains the point cloud detection results based on this. Then, based on the point cloud detection results and the track starting conditions, it is determined whether to track the target. Finally, the radar data is spliced ​​to obtain multi-dimensional omnidirectional track point data, and the radar output distance is obtained based on this. Based on the radar output distance and the preset safety distance, it is determined whether to perform autonomous obstacle avoidance, thereby more accurately identifying and detecting targets, and further achieving more accurate multi-radar omnidirectional detection, effectively solving the problem in the prior art that the multi-radar omnidirectional detection results are greatly affected by the environment.

[0083] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0085] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0087] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0088] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A multi-radar omnidirectional detection method for an unmanned aerial vehicle inertial navigation system, characterized in that: The following steps are involved: S1, compressing the collected radar raw data to obtain two-dimensional frequency domain data, accumulating the two-dimensional frequency domain data and performing constant false alarm rate detection to obtain a point cloud detection result, wherein the compression includes one-dimensional compression and two-dimensional compression; S2, classifying the point cloud to obtain a point cloud clustering result, obtaining target data according to the point cloud clustering result, and classifying the obstacle target to be detected according to the target data to obtain a target classification result, wherein the obstacle target to be detected represents an obstacle in the flight direction of the UAV; S3, based on the point cloud detection results and the track starting conditions, determine whether to track the obstacle target to be detected, and based on the association result of the point cloud and the track, determine whether to update the target data and maintain the track. If the point cloud and the track are not successfully associated, only the historical data is saved; S4, splicing the radar data to obtain multi-dimensional omnidirectional track point data, and combining it with the flight direction of the UAV to obtain the radar output distance, and judging whether to perform autonomous obstacle avoidance based on the radar output distance and the preset safety distance; The specific method of obtaining the multi-dimensional omnidirectional track point data is as follows: E1, receiving and storing radar data within a preset time period, determining whether the stored radar data exceeds the queue length, if the stored radar data exceeds the queue length, deleting the queue head data and storing the unstored radar data at the queue tail, otherwise storing the radar data and executing E2; E2, splicing the time-continuous track point data to obtain omnidirectional detection data, wherein the time-continuous track point data represents the track point data of continuous frames in the time dimension; E3, splicing the spatially continuous track point data to obtain overlapping parts, transforming the track point data into a track coordinate system according to the radar illumination angle and position to obtain multi-dimensional detection data, wherein the spatially continuous track point data represents track point data of continuous frames in the spatial dimension; E4, integrating the track point data according to the omnidirectional detection data and the multi-dimensional detection data to obtain multi-dimensional omnidirectional track point data.

2. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 1, characterized in that: The specific process of the constant false alarm rate detection is as follows: Perform one-dimensional Fourier transform on the radar raw data to obtain the radar's one-dimensional range image data; Performing a two-dimensional Fourier transform on the one-dimensional range image data to obtain two-dimensional frequency domain data; The target signal is divided into a preset number of detection windows according to the result of the accumulation of the two-dimensional frequency domain data, the background noise signal is obtained through the neighborhood area of ​​the detection window, and the detection threshold is obtained according to the background noise signal; It is determined whether the target signal to be detected in the detection window exceeds the detection threshold. If the target signal to be detected exceeds the detection threshold, it is determined to be a target signal, otherwise it is determined to be noise.

3. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 1, characterized in that: The specific method of obtaining the point cloud detection result is as follows: The radar data is input into the radar coordinate system, and the radar coordinate system is converted into the UAV coordinate system. The target point cloud coordinates are input into the UAV coordinate system to obtain the point cloud detection results.

4. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 1, characterized in that: The specific method of obtaining the association result of the point cloud and the track is as follows: Compare the point cloud coordinates and radial velocity with the coordinates and radial velocity of each track to determine whether the first coordinate difference, the second coordinate difference, the third coordinate difference and the velocity difference are not greater than the preset difference threshold. If the first coordinate difference, the second coordinate difference, the third coordinate difference and the velocity difference are not greater than the preset difference threshold, it indicates that the point cloud is successfully associated with the track, otherwise it indicates that the association fails. The first coordinate difference represents the difference between the point cloud abscissa and the track abscissa; The second coordinate difference represents the difference between the point cloud ordinate and the track ordinate; The third coordinate difference represents the difference between the vertical coordinate of the point cloud and the vertical coordinate of the track; The velocity difference represents the difference between the radial velocity and the track radial velocity.

5. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 1, characterized in that: The specific process of determining whether to update the target data and maintain the track based on the association result of the point cloud and the track is as follows: Determine whether the point cloud is successfully associated with the track. If so, update the target data according to the system state vector and maintain the track. Otherwise, determine whether the cumulative number of frames reaches the preset frame number requirement. If so, update the target data according to the system state vector and maintain the track. Otherwise, only save the historical data. The mean of the associated target data is obtained by combining each tracking track, and the system state vector, updated error covariance and updated prediction error covariance are obtained according to the measurement error between the tracking track state estimate and the target data mean.

6. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 5, characterized in that: The updating of target data and maintaining the track further includes: B1, associate the point cloud with the target bounding box, and determine whether the target point cloud coordinates and radial velocity are within the association threshold. If the target point cloud coordinates and radial velocity are not within the association threshold, a deletion operation is performed, otherwise the tracking track is retained; B2, determine whether there is data associated with the radar ID and measurement ID associated with the previous frame in the point cloud data of the current frame. If so, update the target data according to the point cloud data of the current frame, otherwise execute B3; B3, search for the obstacle target to be detected within the associated threshold, determine whether there is an associated measurement, if there is no associated measurement, execute B4, otherwise determine whether there is a unique associated measurement, if there is a unique associated measurement, update the target data according to the associated measurement, otherwise compare the point cloud data of the current frame with the associated measurement to obtain a measurement score, and update the target data according to the measurement corresponding to the minimum measurement score obtained by arranging the measurement scores in descending order; B4, adjust the radial velocity correlation threshold step by step, re-search the obstacle target to be detected within the correlation threshold, and determine whether there is a correlation measurement. If there is no correlation measurement, execute B5, otherwise determine whether there is a unique correlation measurement. If there is a unique correlation measurement, update the target data according to the correlation measurement, otherwise compare the point cloud data of the current frame with the correlation measurement to obtain a measurement score, and update the target data according to the measurement score corresponding to the minimum measurement score obtained by arranging the measurement scores in descending order; B5, adjust the CFAR energy threshold step by step and perform target search on the range Doppler heat map to determine whether there is an associated measurement. If there is no associated measurement, update the fusion target through the system state vector and accumulate the measurement loss time. If the measurement loss time exceeds the preset time, delete the obstacle target to be detected. Otherwise, determine whether there is a unique associated measurement. If there is a unique associated measurement, update the target data according to the associated measurement. Otherwise, compare the point cloud data of the current frame with the associated measurement to obtain the degree of association, arrange the degree of association in ascending order to obtain the maximum degree of association, and update the target data according to the measurement corresponding to the maximum degree of association. The measurement loss time indicates the number of frames where the radar data is not successfully associated.

7. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 1, characterized in that: The specific method for obtaining the radar output distance is as follows: Obtain the current flight direction of the drone, and at the same time obtain the corresponding unique identifier according to the track point in the track point coordinate system, the unique identifier includes the track point ID and the corresponding radar number; Determine whether there is a track point in the flight direction of the drone. If there is no track point in the flight direction of the drone, output the track point as the first identifier; otherwise, output the track point as the second identifier and output the radar output distance and speed information.

8. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 7, characterized in that: The method of obtaining the current UAV flight direction also includes: By reading the attitude information of the drone when it is flying in the air, the inclination data of the drone in each direction at the current moment is obtained; The tilt angle data is compiled into an electronic signal by a compiler in the UAV flight control system and transmitted to the single chip microcomputer in the UAV flight control system in real time; The compensation angle is obtained according to the inclination data of the UAV in various directions at the current moment, and the track point coordinate system compensation processing is performed on the UAV according to the obtained compensation angle.

9. The multi-radar omnidirectional detection method of an unmanned aerial vehicle inertial navigation system as claimed in claim 1, characterized in that: The specific process of autonomous obstacle avoidance is as follows: The speed of the driving motor of each blade of the drone is controlled according to the radar output distance, and it is determined whether the radar output distance is less than the preset safety distance. If the radar output distance is less than the preset safety distance, the flight action is interrupted and the brakes are applied. At the same time, the radar output distance and relative radar movement speed are output in real time through the drone ground station display, otherwise no processing is performed.

Images