Airborne multi-target intelligent identification and positioning method and system based on laser radar

Abstract

The invention belongs to the field of airborne target identification and positioning, and provides an airborne multi-target intelligent identification and positioning method based on a laser radar, and the method comprises the steps: receiving original data in an air refueling scene through the laser radar, extracting continuous single-frame point clouds according to a communication protocol, and adjusting the continuous single-frame point clouds into a position coordinate (x, y, z) form; a frame of complete point cloud is segmented into a plurality of unknown target point cloud clusters by using a segmentation module, a target category of the point cloud clusters is identified by using a target identification network, a target position of the target is positioned by using a target positioning module, finally the types and positions of the plurality of point cloud clusters of the frame of point cloud are identified, and a detection result is sent to a monitoring computer. According to the method, a modularized frame is adopted, the structure is simple, testing and maintenance are easy, high-frequency detection information of the laser radar can be processed in real time, and the method has the advantages of being high in real-time performance, high in detection accuracy and the like.

Description

Technical Field

[0001] This invention belongs to the field of airborne target identification and positioning technology, specifically relating to an airborne multi-target intelligent identification and positioning method and system based on lidar. Background Technology

[0002] LiDAR, an emerging sensor, boasts advantages such as long range, large field of view, and high accuracy. It can accurately measure targets within its field of view and output information such as position and reflection intensity, and has been widely applied in autonomous identification and obstacle avoidance fields such as intelligent vehicles and intelligent robots. In the airborne field, LiDAR-based target localization technology is a research hotspot, while also placing higher demands on the number, type, and positional accuracy of detected targets.

[0003] Aerial refueling technology can effectively improve the endurance of fighter jets and is an important measure to enhance the maneuverability and strike capability of air forces. The key to this technology is the accurate identification and positioning of the receiver aircraft and the drogue, ensuring the accuracy and success rate of docking. Currently, the main method of aerial refueling is visual operation by the pilot, which places extremely high demands on the pilot's operational skills and psychological state. To reduce the pilot's burden, the relative positioning problem of the receiver aircraft and the drogue can be effectively solved by using lidar, cameras, satellites, and other means.

[0004] To address this issue, existing relative positioning solutions utilize vision and satellites. However, vision is insensitive to depth information, and satellite positioning is subject to atmospheric interference, resulting in inaccuracies and a lack of complete autonomy. Drawing on the successful application of vehicle-mounted LiDAR, this solution proposes a LiDAR sensor-assisted approach to achieve target segmentation, identification, and positioning of the receiver aircraft and the boom. Summary of the Invention

[0005] The purpose of this invention is to propose a multi-target intelligent identification and positioning method based on lidar. This method processes and analyzes the data collected by lidar to achieve multi-target segmentation, clustering, identification, and positioning functions from the radar point cloud. It accurately measures the position of the receiving aircraft and the drogue, providing relative positioning information for display and control equipment and flight control system. This method can effectively reduce the pilot's operational burden during refueling and improve accuracy, continuity, and reliability.

[0006] On one hand, this invention provides an airborne multi-target intelligent identification and localization method based on lidar, comprising the following steps: S1, the lidar system acquires point cloud information of continuous frames in the aerial refueling scenario in real time; S2, preprocess the point cloud information obtained in S1 to obtain a complete frame of point cloud information, and extract the three-dimensional coordinate values ​​of the complete frame of point cloud information. S3, construct a point cloud segmentation model, perform multi-target segmentation and noise filtering based on the three-dimensional coordinate values ​​of a complete point cloud information frame, and form multiple point cloud clusters; S4. Construct a point cloud cluster type recognition model to identify multiple point cloud clusters generated in S3 and determine the target type corresponding to each point cloud cluster. S5: Calculate the centroid coordinates of the multiple point cloud clusters generated in S3, and use them as the localization results of the corresponding point cloud clusters; S6 outputs the type and localization results of multiple targets in a complete point cloud frame; S7, repeat S2~S6 to complete the point cloud information recognition and localization calculation for consecutive frames.

[0007] Furthermore, in S2, the three-dimensional coordinate values ​​of a complete point cloud frame are extracted, including: Read a single frame of lidar point cloud data. The information for each point is the measured distance r, vertical angle α, and horizontal angle θ. The method to convert it to three-dimensional coordinates (x, y, z) is as follows: .

[0008] Furthermore, in S3, a point cloud segmentation model is constructed. Based on the 3D coordinate values ​​of a complete frame of point cloud information, multi-target segmentation and noise filtering are performed to form multiple point cloud clusters, including: Construct a point cloud segmentation model with noise removal and point cloud clustering functions; The design sets a distance threshold r and a quantity value m. When the number of points n within a radius r around a point in the point cloud information is less than m, the point is considered a noise point and is filtered out; when the number of points n within a radius r around a point is greater than m, the point is considered a target point and is retained. All target points are judged based on a distance threshold r. If the distance d between target points is less than r, they are considered to be in the same point cloud cluster, thus dividing all target points in the point cloud information into multiple point cloud clusters.

[0009] Furthermore, in S4, a point cloud cluster type recognition model is constructed to identify multiple point cloud clusters generated in S3, and each point cloud cluster is classified into a class, including: By setting all points N of a point cloud cluster as a single input, the distribution pattern of the point cloud does not need to be adjusted. Utilize neural networks to extract feature information from all points in a point cloud cluster; The point cloud cluster is classified using feature information into either "receiving aircraft" or "cone".

[0010] Furthermore, in step S5, the centroid coordinates of the multiple point cloud clusters generated in step S3 are calculated respectively, and used as the localization results of the corresponding point cloud clusters. The method is as follows: .

[0011] On the other hand, the present invention also provides an airborne multi-target intelligent identification and positioning system based on lidar, the system comprising: a point cloud frame preprocessing module, a point cloud segmentation module, a point cloud identification module, and a target positioning module; The point cloud frame preprocessing module is used to acquire point cloud information of continuous frames in the aerial refueling scenario input by the lidar system; and to preprocess the acquired point cloud information to obtain a complete frame of point cloud information and extract the three-dimensional coordinate values ​​of the complete frame of point cloud information. The point cloud segmentation module is used to perform multi-target segmentation and noise filtering based on the three-dimensional coordinate values ​​of a complete point cloud information frame, based on the constructed point cloud segmentation model, to form multiple point cloud clusters; The point cloud recognition module is used to identify multiple point cloud clusters based on the constructed point cloud cluster type recognition model, and determine the target type corresponding to each point cloud cluster. The target localization module is used to calculate the centroid coordinates of multiple point cloud clusters as the localization result of the corresponding point cloud clusters; and outputs the type and localization result of multiple targets in a complete frame of point cloud information.

[0012] Furthermore, the point cloud frame preprocessing module is specifically used for: Read a single frame of lidar point cloud data. The information for each point is the measured distance r, vertical angle α, and horizontal angle θ. The method to convert it to three-dimensional coordinates (x, y, z) is as follows: .

[0013] Furthermore, the point cloud segmentation module is specifically used for: Construct a point cloud segmentation model with noise removal and point cloud clustering functions; The design sets a distance threshold r and a quantity value m. When the number of points n within a radius r around a point in the point cloud information is less than m, the point is considered a noise point and is filtered out; when the number of points n within a radius r around a point is greater than m, the point is considered a target point and is retained. All target points are judged based on a distance threshold r. If the distance d between target points is less than r, they are considered to be in the same point cloud cluster, thus dividing all target points in the point cloud information into multiple point cloud clusters.

[0014] Furthermore, the point cloud recognition module is specifically used for: By setting all points N of a point cloud cluster as a single input, the distribution pattern of the point cloud does not need to be adjusted. Utilize neural networks to extract feature information from all points in a point cloud cluster; The point cloud cluster is classified using feature information into either "receiving aircraft" or "cone".

[0015] The advantages of this invention are: 1. Based on lidar sensors, it can realize multi-target identification and positioning of receiver aircraft and drogue during aerial refueling; 2. For airborne air-to-air scenarios, by utilizing distance threshold constraints, it can effectively distinguish different target obstacles and noise, and can simultaneously segment any number of targets; 3. Employing a neural network-based target recognition strategy, it offers high efficiency and accuracy for recognizing both cone-shaped targets and aircraft. 4. The method is universal and can be applied to refueling scenarios for different types of aircraft. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 For multi-target intelligent recognition and localization functional architecture; Figure 2 The process flow of the multi-object segmentation method; Figure 3 The operation flow of the target recognition method. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0019] Figure 1 This is a functional architecture diagram for multi-target intelligent recognition and positioning. The input is continuous data acquisition from the lidar sensor, and the output is the recognition and positioning results of multiple targets within the lidar's field of view, which will be sent to the host computer, such as the display and control equipment and flight control system. The data processing stage of multi-target intelligent recognition and positioning is mainly divided into four parts, including a point cloud frame preprocessing module, a point cloud segmentation module, a point cloud recognition module, and a target positioning module.

[0020] The point cloud frame preprocessing module receives continuous frame data collected in real time by the lidar and processes the radar point cloud into frames based on information such as frame header, check bit, and frame tail to ensure real-time target recognition and positioning.

[0021] The point cloud segmentation module reads a complete frame of point cloud data, clusters and segments the points in it into different point cloud clusters, and each point cloud cluster represents the same target.

[0022] The point cloud recognition module identifies each point cloud cluster formed after point cloud segmentation and calculates the type of the point cloud cluster.

[0023] The target positioning block performs positioning processing on each point cloud cluster formed after point cloud segmentation and calculates the centroid position of the point cloud cluster.

[0024] Example 1 like Figure 1 As shown, an airborne multi-target intelligent identification and localization method based on lidar includes the following steps: Step S1: Acquire point cloud information in the aerial refueling scenario in real time using a lidar system; Step S2: Preprocess the radar point cloud obtained in S1 to obtain a complete frame of point cloud information and extract the three-dimensional coordinate values; Step S3: Construct a point cloud segmentation model, perform multi-target segmentation and noise filtering based on the point cloud coordinates in S2, and form multiple point cloud clusters; Step S4: Construct a point cloud cluster type recognition model to identify multiple point cloud clusters generated in S3 and classify each point cloud cluster into a class; Step S5: Perform position regression on the point cloud clusters generated in S3, that is, calculate the centroid coordinates once for each point cloud cluster; Step S6: Combine the calculation results of S4 and S5, and output the type and localization results of multiple targets in a single frame of lidar point cloud obtained in S2.

[0025] Step S7: Repeat S2~S6 to complete the point cloud recognition and localization calculation of the continuous frames acquired in real time in S1.

[0026] In one possible embodiment, in step S1, the lidar system collects point cloud data in real time to obtain point cloud information of the receiver aircraft and the drogue during aerial refueling. In one possible embodiment, in step S2, the three-dimensional coordinate values ​​of a radar point cloud frame are extracted based on the information of the frame header, data bits, check bits, and frame tail data bits transmitted by the lidar sensor. The extracted coordinate values ​​are in the form of (N×3), where N is the number of all radar point clouds acquired in the current frame, and each point contains three-dimensional coordinates (x, y, z).

[0027] In one possible embodiment, in step S3, a point cloud segmentation model is constructed, and the point cloud coordinates are used for multi-target segmentation and noise filtering. The calculation process is as follows: Figure 2 The calculation method is as follows: (a) Assume the coordinates of a measurement point in the current frame are (x... i ,y i ,z i), calculate the number of point clouds n within the radius r of the current point. When the number of points n is greater than the set quantity threshold m, the current point and the surrounding N points are considered to be the same target point cloud; when the number of points n is less than the set quantity threshold m, the current point is considered to be a noise point.

[0028] (b) Traverse the remaining points. If a point has already been classified, skip it; otherwise, repeat the previous step until all points have been classified. (c) Depending on the output, multiple point cloud clusters can be generated, and each point cloud cluster is a target point cloud.

[0029] In one possible embodiment, in step S4, a point cloud cluster type recognition model is constructed to identify each point cloud cluster as a class. The calculation process is as follows: Figure 3 The calculation method is as follows: (a) Set all points of the point cloud cluster as a single input, without needing to adjust the distribution pattern of the point cloud; (b) Use neural networks to extract feature information of all point clouds in a point cloud cluster; (c) Classify point cloud clusters using point cloud feature information, such as the current point cloud cluster being "receiving aircraft" or "cone".

[0030] In one possible embodiment, in step S5, a point cloud cluster is subjected to coordinate regression to calculate the centroid coordinates of the point cloud cluster. The calculation method uses the average value method, that is, calculating the centroid coordinates of all points in the point cloud cluster. The result is the target centroid.

[0031]

[0032] In one possible embodiment, in step S6, the number of targets, type, and positioning results of a radar point cloud frame are fused together to form a complete positioning output frame.

[0033] In one possible embodiment, in step S7, the data acquisition of the lidar is real-time, and the result processing of consecutive single frames needs to be repeated to output the target recognition and positioning results in real time.

[0034] This invention proposes applying a lidar system to the aerial refueling process, enabling accurate identification and localization of multiple targets, including the receiver aircraft and the drogue. This method integrates intelligent strategies such as neural networks, achieving high efficiency and accuracy, and enabling real-time positioning in highly dynamic scenarios.

[0035] The above detailed embodiments are a description of the present invention. It should not be considered that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the protection scope of the present invention.

Claims

1. An airborne multi-target intelligent identification and localization method based on lidar, characterized in that: Includes the following steps: S1, the lidar system acquires point cloud information of continuous frames in the aerial refueling scenario in real time; S2, preprocess the point cloud information obtained in S1 to obtain a complete frame of point cloud information, and extract the three-dimensional coordinate values ​​of the complete frame of point cloud information. S3, construct a point cloud segmentation model, perform multi-target segmentation and noise filtering based on the three-dimensional coordinate values ​​of a complete point cloud information frame, and form multiple point cloud clusters; S4. Construct a point cloud cluster type recognition model to identify multiple point cloud clusters generated in S3 and determine the target type corresponding to each point cloud cluster. S5: Calculate the centroid coordinates of the multiple point cloud clusters generated in S3, and use them as the localization results of the corresponding point cloud clusters; S6 outputs the type and localization results of multiple targets in a complete point cloud frame; S7, repeat S2~S6 to complete the point cloud information recognition and localization calculation for consecutive frames.

2. The method according to claim 1, characterized in that, Extract the 3D coordinate values ​​of a complete point cloud from S2, including: Read a single frame of lidar point cloud data. The information for each point is the measured distance r, vertical angle α, and horizontal angle θ. The method to convert it to three-dimensional coordinates (x, y, z) is as follows: 。 3. The method according to claim 1, characterized in that, S3, construct a point cloud segmentation model, perform multi-target segmentation and noise filtering based on the 3D coordinate values ​​of a complete frame of point cloud information, forming multiple point cloud clusters, including: Construct a point cloud segmentation model with noise removal and point cloud clustering functions; The design sets a distance threshold r and a quantity value m. When the number of points n within a radius r around a point in the point cloud information is less than m, the point is considered a noise point and is filtered out; when the number of points n within a radius r around a point is greater than m, the point is considered a target point and is retained. All target points are judged based on a distance threshold r. If the distance d between target points is less than r, they are considered to be in the same point cloud cluster, thus dividing all target points in the point cloud information into multiple point cloud clusters.

4. The method according to claim 1, characterized in that, S4, construct a point cloud cluster type recognition model to identify multiple point cloud clusters generated in S3, and classify each point cloud cluster into a class, including: By setting all points N of a point cloud cluster as a single input, the distribution pattern of the point cloud does not need to be adjusted. Utilize neural networks to extract feature information from all points in a point cloud cluster; The point cloud cluster is classified using feature information into either "receiving aircraft" or "cone".

5. The method according to claim 1, characterized in that, S5. Calculate the centroid coordinates of each of the multiple point cloud clusters generated in S3, and use these coordinates as the localization results for the corresponding point cloud clusters. The method is as follows: 。 6. An airborne multi-target intelligent identification and positioning system based on lidar, characterized in that, The system includes: a point cloud frame preprocessing module, a point cloud segmentation module, a point cloud recognition module, and a target localization module; The point cloud frame preprocessing module is used to acquire point cloud information of continuous frames in the aerial refueling scenario input by the lidar system; and to preprocess the acquired point cloud information to obtain a complete frame of point cloud information and extract the three-dimensional coordinate values ​​of the complete frame of point cloud information. The point cloud segmentation module is used to perform multi-target segmentation and noise filtering based on the three-dimensional coordinate values ​​of a complete point cloud information frame, based on the constructed point cloud segmentation model, to form multiple point cloud clusters; The point cloud recognition module is used to identify multiple point cloud clusters based on the constructed point cloud cluster type recognition model, and determine the target type corresponding to each point cloud cluster. The target localization module is used to calculate the centroid coordinates of multiple point cloud clusters as the localization result of the corresponding point cloud clusters; and outputs the type and localization result of multiple targets in a complete frame of point cloud information.

7. The airborne multi-target intelligent identification and positioning system based on lidar according to claim 6, characterized in that, The point cloud frame preprocessing module is specifically used for: Read a single frame of lidar point cloud data. The information for each point is the measured distance r, vertical angle α, and horizontal angle θ. The method to convert it to three-dimensional coordinates (x, y, z) is as follows: 。 8. The airborne multi-target intelligent identification and positioning system based on lidar according to claim 7, characterized in that, The point cloud segmentation module is specifically used for: Construct a point cloud segmentation model with noise removal and point cloud clustering functions; Design a distance threshold r and a number value m. If the number of points n within a radius r around a point in the point cloud information is less than m, the point is considered a noise point and is filtered out. If the number of point cloud points n within a radius r around a point is greater than m, then the point is considered a target point and is retained. All target points are judged based on a distance threshold r. If the distance d between target points is less than r, they are considered to be in the same point cloud cluster, thus dividing all target points in the point cloud information into multiple point cloud clusters.

9. The airborne multi-target intelligent identification and positioning system based on lidar according to claim 8, characterized in that, The point cloud recognition module is specifically used for: By setting all points N of a point cloud cluster as a single input, the distribution pattern of the point cloud does not need to be adjusted. Utilize neural networks to extract feature information from all points in a point cloud cluster; The point cloud cluster is classified using feature information into either "receiving aircraft" or "cone".

Images