Precise target center positioning method and device of unmanned aerial vehicle, terminal and storage medium

Abstract

The invention discloses an accurate target center positioning method and device for an unmanned aerial vehicle, a terminal and a storage medium, and the method comprises the steps: obtaining environment distance information through cooperation of a laser radar and vertical laser ranging, and carrying out the time synchronization of the environment distance information and obtained inertial measurement unit data, establishing multi-sensor sensing data under a unified coordinate system; constructing an environment map in real time, performing visual identification on the target target through a camera, and calculating a three-dimensional position of the target target in a unified coordinate system; according to the environment map and the three-dimensional position, a preset clustering algorithm is adopted to detect obstacles in the surrounding environment of the target target, and an obstacle avoidance path leading to the target target is generated; and in the process of approaching the target target along the obstacle avoidance path, when the target target meets a preset spatial position condition and a posture stability condition, laser shooting is triggered, and a target positioning and shooting verification result is output. According to the method, unmanned aerial vehicle positioning is more stable and accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicles, in particular to a method and device for accurately positioning a target center by an unmanned aerial vehicle, a terminal and a storage medium. BACKGROUND

[0002] In today's era of rapid technological development, unmanned aerial vehicle technology has made significant progress and plays an important role in many fields such as agriculture, logistics, surveying and mapping, and security. With the continuous expansion of the application scenarios of unmanned aerial vehicles, the requirements for their positioning and obstacle avoidance systems are also becoming higher and higher to ensure that unmanned aerial vehicles can operate safely and accurately in various complex environments.

[0003] Currently, the positioning and obstacle avoidance system of unmanned aerial vehicles mainly relies on visual sensors for environment perception, such as RGB cameras, depth cameras or binocular vision. At the same time, the height determination of existing unmanned aerial vehicles mainly relies on barometers or ultrasonic sensors.

[0004] However, the existing unmanned aerial vehicle system has obvious limitations. Visual sensors perform significantly worse in complex optical conditions such as low light, direct sunlight or smoke, which can easily lead to positioning drift or obstacle avoidance failure. SLAM based on vision requires real-time processing of a large amount of image data, which relies on high-performance processors, increasing the power consumption and hardware cost of unmanned aerial vehicles. In addition, visual algorithms need to analyze the motion trend of obstacles through multiple images, which may cause collision risks in high-speed flight scenarios due to processing delays. Some visual sensors also have distortion or ranging errors at close range, making it difficult to achieve precise obstacle avoidance. In addition, barometers used for height determination are easily affected by weather, leading to data drift, and ultrasonic sensors have limited effective range and are sensitive to the material of the reflecting surface. SUMMARY

[0005] The purpose of the present application is to overcome the above technical problems, and a method and device for accurately positioning a target center by an unmanned aerial vehicle, a terminal and a storage medium are provided.

[0006] In a first aspect, the present application provides a method for accurately positioning a target center by an unmanned aerial vehicle, which adopts the following technical solution: A method for accurately positioning a target center by an unmanned aerial vehicle, comprising the following steps: Obtain environment distance information by laser radar and vertical direction laser ranging in cooperation, time synchronize the environment distance information with the obtained inertial measurement unit data, and establish multi-sensor perception data in a unified coordinate system; Real-time construct an environment map based on the multi-sensor perception data, visually recognize a target target by a camera and calculate the three-dimensional position of the target target in the unified coordinate system; According to the environment map and the three-dimensional position, obstacles in the surrounding environment of the target target are detected by using a preset clustering algorithm, and an obstacle avoidance path to the target target is generated; In the process of approaching the target target along the obstacle avoidance path, laser shooting is triggered when the target target meets preset spatial position conditions and attitude stability conditions, and a target positioning and shooting verification result is output.

[0007] By using the above technical solutions, the environment distance information is obtained by using the laser radar and the vertical direction laser ranging, and the inertial measurement unit data is time-synchronized, the multi-sensor perception data in the unified coordinate system is established, the all-weather, low-delay and high-robust positioning and obstacle avoidance are realized, the drift of the barometer and the limitation of the ultrasonic ranging are solved, and the centimeter-level accurate height control is realized; the environment map is constructed in real time based on the multi-sensor perception data, the environment information can be accurately presented; the target target is visually recognized by the camera and the three-dimensional position of the target target in the unified coordinate system is calculated, the target target can be accurately positioned; the obstacles in the surrounding environment of the target target are detected by using a preset clustering algorithm, and an obstacle avoidance path is generated, the target target can be effectively reached by avoiding obstacles; when the target target meets the preset conditions, laser shooting is triggered and a verification result is output, the accurate positioning of the target center and the shooting verification can be realized.

[0008] Preferably, the environment distance information is obtained by using the laser radar and the vertical direction laser ranging, the environment distance information is time-synchronized with the obtained inertial measurement unit data, and the multi-sensor perception data in the unified coordinate system is established, specifically including the following steps: The horizontal scanning is performed by the two-dimensional laser radar according to a preset first scanning frequency, the plane point cloud data is obtained, each point carries out the vertical direction laser ranging according to a preset second scanning frequency by the TOF principle, the vertical distance data of the unmanned aerial vehicle is obtained, and the inertial measurement unit data in the running process of the unmanned aerial vehicle is obtained in real time by the IMU installed on the unmanned aerial vehicle; The plane point cloud data, the vertical distance data and the inertial measurement unit data are time-synchronized by the ROS message filtering mechanism; The pre-calibrated transformation matrix is loaded, the transformation matrix describes the fixed transformation relationship between the multi-sensor perception data including the plane point cloud data, the vertical distance data and the inertial measurement unit data, and the multi-sensor perception data is converted to the body coordinate system of the unmanned aerial vehicle in real time.

[0009] By adopting the technical scheme, the distance information of the environment is obtained by the two-dimensional laser radar and the vertical direction laser ranging in cooperation, the IMU data is combined, time synchronization is realized through a ROS message filtering mechanism, and the multi-sensor data is converted to the body coordinate system through a pre-calibration transformation matrix, so that the multi-sensor fusion perception data in a unified coordinate system can be established, the accuracy and stability of the environment perception are improved, the defects of the existing unmanned aerial vehicle in the multi-sensor data fusion efficiency are solved, the radar is not affected by light, dust, rain and fog, the inherent defects of the visual sensor can be made up, the distance / azimuth information is directly output, complex image processing is not needed, and the algorithm requirement is reduced.

[0010] Preferably, the environment map is constructed in real time based on the multi-sensor perception data, the target target is visually recognized by the camera, and the three-dimensional position of the target target in the unified coordinate system is calculated, specifically including the following steps: the motion distortion during the laser radar is compensated according to the angular velocity in the inertial measurement unit data to obtain de-distortion point cloud data, and the environment map of the 2.5-dimensional grid is obtained according to the de-distortion point cloud data and the vertical distance data. An image frame captured by the camera is obtained, visual feature extraction is performed on the image frame by using a preset visual recognition model, target heart pixel coordinates of a target heart of the target target in the image frame are obtained, and target target three-dimensional coordinates of the target heart of the target target in the body coordinate system are obtained in combination with the vertical distance data and camera intrinsic data of the camera.

[0011] By adopting the technical scheme, the motion distortion of the laser radar is compensated according to the angular velocity in the inertial measurement unit data, accurate de-distortion point cloud data can be obtained, and an accurate 2.5-dimensional grid environment map can be constructed in combination with the vertical distance data; the visual feature extraction is performed on the image frame captured by the camera by using the preset visual recognition model, and the target target three-dimensional coordinates of the target heart of the target target in the camera coordinate system can be accurately calculated in combination with the vertical distance data and the camera intrinsic data, so that the accurate positioning of the target target is realized.

[0012] Preferably, according to the environment map and the three-dimensional position, a preset clustering algorithm is adopted to detect the obstacles in the surrounding environment of the target target and generate an obstacle avoidance path leading to the target target, specifically including the following steps: the environment map and the target target three-dimensional coordinates are received, a region of interest is defined with the target target as the center, and all obstacle grids in the region of interest are extracted. An improved DBSCAN algorithm is adopted to perform density clustering to obtain a plurality of effective obstacle clusters. Expanding each obstacle cluster to obtain a no-fly zone, marking the remaining space excluding the no-fly zone as a three-dimensional flyable space, generating multiple candidate obstacle avoidance paths that meet preset constraints in velocity space using a dynamic window method, and performing a multi-objective cost evaluation on the candidate obstacle avoidance paths, where the evaluation metrics for each candidate obstacle avoidance path include distance cost, smoothness cost, and safety cost; The candidate obstacle avoidance path with the lowest comprehensive cost is selected to output a control instruction, and when it is detected that the obstacle distance is less than a safety threshold, an emergency stop protocol is triggered.

[0013] By adopting the above technical solution, the region of interest is defined with the target as the center and the obstacle grid is extracted, which can accurately focus on the obstacles around the target; the improved DBSCAN algorithm is used for density clustering to obtain effective obstacle clusters, which can accurately identify the distribution of obstacles; the obstacle clusters are expanded to obtain no-fly zones and mark the three-dimensional flyable space, clarifying the range within which the UAV can fly; the dynamic window method is used to generate candidate obstacle avoidance paths and perform multi-objective cost evaluation, which can comprehensively consider factors such as distance, smoothness and safety to screen paths; the candidate obstacle avoidance path with the lowest comprehensive cost is selected to output control instructions, which can plan the optimal obstacle avoidance path for the UAV; when the obstacle distance is detected to be less than the safety threshold, the emergency stop protocol is triggered to ensure the flight safety of the UAV.

[0014] Preferably, the step of obtaining a plurality of valid obstacle clusters and their boundary polygons includes classifying and identifying static obstacles and dynamic obstacles, specifically including the following steps: Obtain N consecutive frames of planar point cloud data from LiDAR scanning, and calculate the pose consistency of each grid in adjacent submaps through the submap matching mechanism of the Cartographer algorithm. When the pose meets the preset first offset condition, it is determined to be a static obstacle; Calculating the instantaneous velocity vector of the obstacle cluster, determining it as a dynamic obstacle when the velocity vector meets a preset second offset condition, and predicting its motion trajectory through linear regression of M consecutive frames of historical data; When the obstacle meets the preset semi-static obstacle determination condition, the TOF vertical scanning verification is triggered. If the height change rate of the obstacle exceeds the preset height change threshold, the obstacle is determined to be the dynamic obstacle.

[0015] By adopting the technical scheme, the continuous N frames of planar point cloud data scanned by the laser radar and the submap matching mechanism of the Cartographer algorithm can be used to accurately determine the static obstacles and improve the accuracy of environment map construction; the instantaneous velocity vector of the obstacle cluster can be calculated to determine the dynamic obstacles and predict the motion trajectory of the dynamic obstacles, thereby improving the response capability to the dynamic obstacles; the TOF vertical scanning verification is performed when the semi-static obstacle determination condition is triggered, thereby further refining the obstacle classification and improving the accuracy of the obstacle avoidance path planning.

[0016] Preferably, the laser shooting is triggered when the target target meets the preset spatial position condition and attitude stability condition, and a positioning result fused with the multi-sensor fusion perception data is output, and the method specifically comprises the following steps: The target three-dimensional coordinates of the target target are acquired, the rotation matrix of the current pose of the unmanned aerial vehicle is determined through the inertial measurement unit data, and the target three-dimensional coordinates in the body coordinate system are converted into target world coordinates according to the rotation matrix; when the Euclidean distance between the target target and the unmanned aerial vehicle reaches a preset shooting threshold and the offset of the target target in the image coordinate system is within a preset spatial range, it is determined whether the target target meets the preset spatial position condition; if the target target meets the spatial position condition, the displacement standard deviation of the target center pixel coordinates in continuous K frames is calculated according to the image frames acquired by the camera, and if the obtained displacement velocity meets a preset stable velocity value, the simulated laser shooting of the unmanned aerial vehicle on the target target is triggered.

[0017] By adopting the technical scheme, the rotation matrix is determined through the inertial measurement unit data to convert the target three-dimensional coordinates into target world coordinates, the coordinate system is unified, it is determined whether the target target meets the spatial position condition according to the Euclidean distance and the offset of the image coordinate system, and it is determined whether the displacement velocity meets the stable velocity value according to the displacement standard deviation of the target center pixel coordinates, so as to accurately determine whether the target target meets the shooting condition and trigger the simulated laser shooting, precise positioning and shooting can be realized, and the positioning result fused with the multi-sensor fusion perception data is output, thereby improving the reliability and accuracy of the positioning result.

[0018] Preferably, the method further comprises the following steps: The laser spot position in the image frame obtained after the simulated laser is triggered is detected, the Euclidean distance deviation of the laser spot position and the target center pixel coordinates is calculated, the current shooting deviation threshold is determined according to the shooting distance, the hit confidence is calculated according to the Euclidean distance deviation and the shooting deviation threshold, and if the hit confidence exceeds a preset confidence threshold, the grid where the target target is located is marked as a traversable area. The target image region feature template at the hit moment is extracted, and the visual recognition model is updated according to the target image region feature template. If the hit confidence does not exceed a preset confidence threshold, the laser spot position is recorded and a temporary obstacle avoidance point is generated, and the obstacle avoidance path is regenerated.

[0019] By adopting the technical solutions, the laser spot position in the image frame after the simulated laser trigger is detected, the Euclidean distance deviation of the laser spot position from the target center pixel coordinate is calculated, the deviation threshold is determined in combination with the shooting distance, and the hit confidence is calculated, so that the shooting hit condition can be evaluated; when the hit confidence exceeds the threshold, the grid where the target is located is marked as a traversable area, which is helpful for subsequent path planning; the target image region feature template at the hit moment is extracted to update the visual recognition model, so that the visual recognition accuracy can be improved; when the hit confidence does not exceed the threshold, the laser spot position is recorded and a temporary obstacle avoidance point is generated, and the obstacle avoidance path is regenerated, so that the obstacle avoidance strategy and path planning can be optimized.

[0020] In a second aspect, the present application provides a precise positioning device for a target center by a UAV, which adopts the following technical solutions: A precise positioning device for a target center by a UAV, comprising the following modules: A data synchronization module is configured to acquire environmental distance information by laser radar and vertical direction laser ranging in cooperation, time-synchronize the environmental distance information with acquired inertial measurement unit data, and establish multi-sensor perception data in a unified coordinate system; an environmental map construction module is configured to construct an environmental map in real time based on the multi-sensor perception data, perform visual recognition on a target object by a camera, and calculate a three-dimensional position of the target object in the unified coordinate system; An obstacle avoidance planning module is configured to detect obstacles in the surrounding environment of the target object and generate an obstacle avoidance path to the target object by adopting a preset clustering algorithm according to the environmental map and the three-dimensional position; A simulated shooting module is configured to trigger laser shooting when the target object meets preset spatial position conditions and attitude stability conditions during approaching the target object along the obstacle avoidance path, and output target positioning and shooting verification results.

[0021] By adopting the above technical solutions, the data synchronization module can acquire environmental distance information by laser radar and vertical direction laser ranging in cooperation, and time-synchronize the environmental distance information with inertial measurement unit data, so as to establish multi-sensor perception data in a unified coordinate system, and improve data accuracy and stability; the environmental map construction module constructs an environmental map in real time based on multi-sensor perception data, and performs visual recognition on a target object by a camera and calculates a three-dimensional position, so as to facilitate comprehensive understanding of environmental and target information; the obstacle avoidance planning module detects obstacles by adopting a clustering algorithm according to an environmental map and a three-dimensional position of a target object, and generates an obstacle avoidance path, so as to ensure flight safety of a UAV; the simulated shooting module triggers laser shooting when a target object meets conditions, and outputs verification results, so as to realize precise positioning and shooting verification of a target center.

[0022] In a third aspect, the present application provides an intelligent terminal, adopting the technical scheme as follows: An intelligent terminal, comprising a memory and a processor, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to realize the unmanned aerial vehicle precision positioning method for a target center as described above.

[0023] In a fourth aspect, the present application provides a computer-readable storage medium, adopting the technical scheme as follows: A computer-readable storage medium, the readable storage medium storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to realize the unmanned aerial vehicle precision positioning method for a target center as described above.

[0024] In summary, the present application at least contains the following beneficial effects: (1) The present application uses multi-sensor positioning fusion, uses laser radar and vertical direction laser ranging and inertial measurement unit data to establish multi-sensor perception data in a unified coordinate system, so that the unmanned aerial vehicle positioning is more stable and accurate, solves the problems of performance decline, positioning drift or obstacle avoidance failure of visual sensors under complex optical conditions, and the problems of existing height sensors affected by weather, range and reflection surface material; (2) The present application constructs a 2.5-dimensional environment map based on multi-sensor perception data, plans an obstacle avoidance path and updates it in real time, can respond to dynamic obstacles in time in a high-speed flight scene, avoid collision risk, and solve the problem of dynamic obstacle response delay in the prior art.

[0025] (3) In the present application, the radar directly outputs distance / azimuth information, without complex image processing, reducing the dependence on high-performance processors, reducing the power consumption and hardware cost of the unmanned aerial vehicle, and solving the problem of strong dependence on computing power in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of the embodiment unmanned aerial vehicle precision positioning method for a target center; Figure 2 is a static coordinate conversion schematic diagram of the embodiment unmanned aerial vehicle precision positioning method for a target center; Figure 3 is a flowchart of step S3 of the embodiment unmanned aerial vehicle precision positioning method for a target center; Figure 4 is an architectural diagram of the embodiment unmanned aerial vehicle precision positioning device for a target center. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be further described in detail below with reference to the drawings. The described embodiments are only possible technical implementations of the present application, but the present application is not limited to this. Other embodiments obtained by those skilled in the art without creative work based on the embodiments of the present application are also within the protection scope of the present application.

[0028] The present application mainly adopts laser radar and multi-sensor fusion to realize accurate positioning and shooting of a target center by a UAV, so as to improve the positioning accuracy, obstacle avoidance ability and shooting accuracy of the UAV, and achieve the effect of operation in a complex environment. The present application will be described in further detail as follows.

[0029] The steps include acquiring environmental distance information, constructing an environmental map, generating an obstacle avoidance path, triggering laser shooting and outputting a verification result. The multi-sensor fusion can construct an accurate environmental map based on the information acquired by the multi-sensor cooperation and fusion, generate a reasonable obstacle avoidance path, trigger laser shooting and output a verification result when the conditions are met, so as to improve the positioning accuracy and obstacle avoidance ability of the UAV and realize accurate shooting of a target center. This is because the multi-sensor fusion can comprehensively utilize the advantages of each sensor to provide more comprehensive and accurate environmental information, so that the UAV can better cope with a complex environment and make correct decisions.

[0030] A method for accurate positioning of a target center by a UAV, as shown in FIG. 1, includes the following steps: Figure 1 S1, acquiring environmental distance information by laser radar and vertical direction laser ranging cooperation, time synchronizing the environmental distance information and acquired inertial measurement unit data, and establishing multi-sensor perception data in a unified coordinate system, which specifically includes the following steps: S11, performing horizontal scanning by a two-dimensional laser radar according to a preset first scanning frequency to obtain planar point cloud data. In this embodiment, the two-dimensional laser radar performs horizontal mechanical rotation scanning at a frequency of 10 Hz, emits a laser pulse and receives a reflected signal, measures the distance value corresponding to each scanning angle, acquires 360 ranging points each time, the measurement range is 0.1-20 meters, and planar point cloud data in polar coordinate form is generated, each data point containing a distance value and a scanning angle.

[0031] Each point carries vertical direction laser ranging implemented according to a preset second scanning frequency by a TOF principle to acquire UAV vertical distance data. In this embodiment, a laser ranging unit based on the time of flight principle emits modulated infrared light to the front direction at a frequency of 100 Hz, calculates the vertical height of the UAV from the ground by measuring the time difference between the emitted light and the reflected light, outputs a single point distance value, and the unit is meter, the accuracy is ±1 centimeter, and the accuracy is ±1 centimeter.

[0032] The inertial measurement unit data during the operation of the unmanned aerial vehicle is acquired in real time by the IMU installed on the unmanned aerial vehicle; the IMU is installed on the unmanned aerial vehicle, and can acquire data such as acceleration and angular velocity during the operation of the unmanned aerial vehicle in real time; the IMU can be a MEMS (micro-electro-mechanical system) inertial measurement unit, and has the advantages of small size and low cost.

[0033] In this embodiment, the IMU collects three-axis acceleration (range ± 16g) and angular velocity data (range ± 2000° / s) at a frequency of 200 times per second, eliminates zero drift by a temperature compensation algorithm, and obtains the real-time attitude angle rate and linear acceleration of the unmanned aerial vehicle.

[0034] S12, time synchronization is performed through a ROS message filtering mechanism, a coordinate system conversion chain is established through a TF2 library of ROS, and a transformation between any two coordinate systems is dynamically calculated; specifically, in a TF2 framework of ROS, a relative relationship of a laser radar coordinate system (X axis forward, Y axis left, and Z axis upward) of planar point cloud data, a TOF coordinate system (Z axis downward) of vertical distance data, and an IMU coordinate system of inertial measurement unit data is predefined, and a static coordinate conversion tree is formed, as shown in the figure. Figure 2 The planar point cloud data, the vertical distance data, and the inertial measurement unit data are fused to form multi-sensor perception data in a unified coordinate system.

[0035] Before data fusion, the planar point cloud data, the vertical distance data, and the inertial measurement unit data are processed respectively; the processing of the planar point cloud data includes removing invalid points and ground segmentation; in an implementable specific mode, the invalid points are defined as points with a distance greater than 20 m or an intensity less than 10, and ground segmentation is realized based on RANSAC plane fitting.

[0036] The time synchronization in this embodiment includes hardware trigger synchronization and software synchronization; the hardware trigger synchronization is that a hardware trigger signal is generated by the FPGA, and the laser radar and the TOF ranging module in the unmanned aerial vehicle are awakened at the same time; the software synchronization is that a time synchronization strategy is created by ROS, and time alignment of the planar point cloud data, the vertical distance data, and the inertial measurement unit data is realized.

[0037] S13, a pre-calibrated transformation matrix is loaded; the transformation matrix describes a fixed transformation relationship between multi-sensor data including the planar point cloud data, the vertical distance data, and the inertial measurement unit data, including a rotation matrix and a translation vector of the laser radar to the body coordinate system, and a vertical offset of the installation position of the TOF sensor.

[0038] S14, the multi-sensor data is converted to the body coordinate system of the unmanned aerial vehicle in real time.

[0039] In a specific implementation method, after the polar coordinate point cloud of the lidar is converted into a Cartesian coordinate system, it is mapped to the body coordinate system through rotation and translation transformation, and the vertical distance data is superimposed with the installation offset and converted into the vertical height in the body coordinate system, so that all data are finally expressed in a unified manner in the body coordinate system.

[0040] S2. Build an environment map in real time based on multi-sensor perception data, use the camera to visually identify the target and calculate the three-dimensional position of the target in a unified coordinate system. This specifically includes the following steps: S21. Calculate the change in the attitude of the UAV at each scanning point collected by the two-dimensional laser radar based on the angular velocity in the inertial measurement unit data.

[0041] S22. Compensate for motion distortion during the laser radar scan to obtain dedistorted point cloud data. In a specific feasible method, apply reverse rotation compensation to the original point cloud to eliminate the point cloud distortion caused by the drone's motion during scanning.

[0042] S23. Obtain a 2.5-dimensional grid environment map based on the dedistorted point cloud data and the vertical distance data.

[0043] In one specific implementation, the compensated, dedistorted horizontal point cloud is discretized into a 5-centimeter resolution grid map, with each grid recording the probability of being occupied by an obstacle. Vertical distance data is then integrated to assign a vertical height attribute to each horizontal grid, forming a 2D enhanced map with a height field.

[0044] S24, obtaining an image frame captured by the camera, performing visual feature extraction on the image frame using a preset visual recognition model, and obtaining the pixel coordinates of the bull's eye of the target in the image frame.

[0045] The RGB image captured by the camera is preprocessed. In one implementation, contrast-limited adaptive histogram equalization is used. The ORB algorithm is used to detect the target corner features of the target. These features are matched with the pre-stored target template. The random sampling consensus algorithm (RANSAC) is used to eliminate mismatched point pairs, and the pixel coordinates (u, v) of the center of the target in the image are finally determined.

[0046] S25. Combining the vertical distance data and the camera intrinsic parameter data of the camera, the three-dimensional coordinates of the target center in the body coordinate system are obtained.

[0047] Combining the pixel coordinates of the bull's eye, the camera intrinsic parameters, and the vertical distance data provided by TOF, the camera 3D coordinates of the bull's eye in the camera coordinate system are calculated through perspective projection inverse transformation. The camera intrinsic parameter data includes the focal length and optical center coordinates.

[0048] The projection model is specifically: X1 = (u - cx) x Z_tof / fx; Y1 = (v - cy) x Z_tof / fy; Z1 = Z_tof; fx, fy are focal length, cx, cy are optical center, Z_tof is vertical distance data, here used to represent the distance along the optical axis of the camera, i.e. the camera depth value.

[0049] P1(X1, Y1, Z1) is the camera three-dimensional coordinates of the target target in the camera coordinate system.

[0050] The three-dimensional coordinates of the target target relative to the unmanned aerial vehicle body coordinate system are converted by the camera to the body calibration external parameter, and the calibration external parameter includes a first rotation matrix and a first translation vector.

[0051] Specifically, Wherein, R1 is the first rotation matrix, used to represent the rotation relationship of the camera coordinate system relative to the body coordinate system; T1 is the first translation vector, used to represent the position offset of the camera coordinate system origin in the body coordinate system.

[0052] The calibration external parameter is calibrated and obtained using a special calibration tool and a calibration board. In this embodiment, the calibration board is placed in the camera field of view, and the unmanned aerial vehicle is kept stationary. By shooting multiple sets of calibration board images at different angles, the first rotation matrix and the first translation vector between the camera coordinate system and the body coordinate system are calculated by using image processing algorithm and optimization method.

[0053] The above calibration process is performed before the unmanned aerial vehicle executes the task, and the calibration result is relatively fixed, unless the installation position of the camera or the unmanned aerial vehicle changes.

[0054] S3, according to the environment map and the three-dimensional position, using a preset clustering algorithm to detect the obstacles in the surrounding environment of the target target and generate an obstacle avoidance path to the target target, as shown in Figure 3 As shown, the method comprises the following steps: S31, receiving the environment map and the three-dimensional coordinates of the target target, defining a region of interest centered on the target target, and extracting all obstacle grids in the region of interest.

[0055] In this embodiment, the region of interest is a cylindrical region with the target target as the center and a radius of 5m. All grids with an occupancy probability greater than 70% in the region of interest are taken as obstacle grids.

[0056] Each grid stores an occupancy probability P1, ranging from [0, 1]. When a new frame of point cloud is observed, the planar point cloud data of the laser radar needs to be projected into the grid map, and the occupancy probability of each grid is updated, specifically, where P p is the historical occupancy probability of the grid, i.e., the value after the last update, P n is the new occupancy probability of the grid determined by the planar point cloud data.

[0057] S32, using the improved DBSCAN algorithm to perform density clustering on the obstacle grid of the region of interest, merging spatially adjacent grids, obtaining a plurality of effective obstacle clusters and the centroid coordinates of each obstacle cluster according to the point cloud of the environment map, as basic data for subsequent obstacle recognition.

[0058] The parameter configuration of the improved DBSCAN algorithm includes: neighborhood radius ε = 0.3m, which can be adjusted in value.

[0059] The minimum cluster point threshold Pmin = 5, that is, each obstacle cluster contains at least 5 or more laser ranging points to meet the minimum cluster sample number and also to ensure noise filtering capability.

[0060] For the improved DBSCAN algorithm, a height dimension constraint is introduced: only when the horizontal distance between two points is less than the neighborhood radius and the vertical distance is less than 0.5m, the two points are determined as the same cluster obstacle cluster.

[0061] According to the convex hull algorithm, a 3D boundary polygon with height information is generated and output.

[0062] S33, obtaining a plurality of effective obstacle clusters and their boundary polygons includes classification and recognition of static obstacles and dynamic obstacles, specifically including the following steps: S331, acquiring continuous N frames of planar point cloud data obtained by laser radar scanning, where N is 10 in this embodiment; The pose consistency of each grid in adjacent sub-maps is calculated through the sub-map matching mechanism of the Cartographer algorithm, that is, the pose change of the obstacle grid is calculated by matching adjacent sub-maps through the ICP algorithm.

[0063] When the pose satisfies a preset first offset condition, it is determined as a static obstacle, and the static obstacle is mapped to a persistent layer of the environment map, and morphological closing operation is used to eliminate measurement noise. In this embodiment, the first offset condition is that the pose offset is ≤0.2 meters and the contour similarity is >90%. For example, in a specific implementation, the grid coordinates of a building change by <0.1 meters in three consecutive sub-maps, which is determined as a static obstacle and mapped to the persistent layer of the environment map. The threshold of the pose offset can be determined according to 10 times the tolerance of the laser ranging accuracy.

[0064] S332, calculate the instantaneous velocity vector of the obstacle cluster, when the velocity vector meets the preset second offset condition, determine it as a dynamic obstacle, and predict the motion trajectory through linear regression of the historical data of the continuous M frames; In this embodiment, the centroid coordinate change of each obstacle cluster is tracked, and the instantaneous velocity v of the obstacle cluster is calculated. Δx is the displacement between adjacent frames, with the unit of m, and Δt is the change period, with the unit of s.

[0065] If the velocity v is greater than or equal to 0.5 m / s, it is marked as a dynamic obstacle. For the identified dynamic obstacle, based on the position change data of the continuous 5 frames, the linear regression method is used to predict the motion trajectory in the future 0.3-1.0 seconds.

[0066] S333, when the obstacle meets the preset semi-static obstacle determination condition, trigger the TOF vertical scanning verification, if the height change rate of the obstacle exceeds the preset height change threshold, the obstacle is determined as a dynamic obstacle.

[0067] When the speed of the detected obstacle is in the preset change range, it is determined that the semi-static obstacle determination condition is met, and the obstacle is a semi-static obstacle. The speed change range in this embodiment is 0.05-0.2 m / s; For semi-static obstacles, trigger TOF vertical scanning verification. In this embodiment, if the height change rate is greater than 0.1 m / s, the semi-static obstacle is reclassified as a dynamic obstacle, otherwise it is a static obstacle. The semi-static obstacle reduces the misjudgment rate through the double verification mechanism.

[0068] The purpose of identifying static obstacles is to obtain the impassable area of the global path, and the purpose of calculating the predicted trajectory of dynamic obstacles is to perform real-time obstacle avoidance constraints for the DWA algorithm.

[0069] S34, perform inflation processing on each obstacle cluster to obtain a no-fly zone, and perform differential processing on different types of obstacles.

[0070] For static obstacles, the inflation radius is the radius of the unmanned aerial vehicle plus a 0.2 m margin, which is adjustable; For dynamic obstacles, calculate the time-varying inflation radius, that is, r=r0+k×v×t, where r0 is the reference radius, k is the safety factor, and t is the prediction time.

[0071] In this embodiment, r0=0.3 m, k=1.2, and t=1 s.

[0072] Mark the remaining space excluding the no-fly zone as a three-dimensional flyable space, and output the visualization mapping of the three-dimensional flyable space and the no-fly zone, such as marking the three-dimensional flyable space as a white area and the no-fly zone as a red area.

[0073] S35, adopt a dynamic window method to generate a plurality of candidate obstacle avoidance paths in the speed space that meet the preset constraint conditions, and the speed space in the embodiment is defined as a linear speed range ∈ [0, 1.2] m / s and an angular velocity range ∈ [-1.5, 1.5] rad / s.

[0074] Multi-objective cost evaluation is performed on the candidate obstacle avoidance paths, and the evaluation indexes of each candidate obstacle avoidance path include distance cost, smoothness cost and safety cost.

[0075] The distance cost is the proximity to the target position, that is, the Euclidean distance between the trajectory endpoint and the target; The safety cost is the minimum interval distance between the trajectory point and the obstacle; The smoothness cost is the trajectory curvature smoothness, that is, the trajectory curvature integral.

[0076] The distance cost is: wherein, wherein P end is the trajectory endpoint, P t is the target coordinate, and σ1 controls the decay speed, and in the embodiment, σ1 = 1 m.

[0077] The smoothness cost is the curvature integral, The trajectory curvature k is calculated through three adjacent points on the trajectory, and the specific calculation method is not described here. wherein, M is the number of points on the trajectory; if the smoothness cost is greater than the preset maximum curvature, which is 5 rad / m in the embodiment, the smoothness cost is normalized to [0, 1].

[0078] For each point on the trajectory, the distance to the nearest obstacle is calculated and the minimum value d min is taken. If dmin is less than the preset safety threshold, the safety cost C3 = 1, and the highest safety cost is directly assigned to trigger the emergency stop; otherwise, wherein, σ2 controls the decay, and in the embodiment, σ2 = 0.5 m; S36, select the candidate obstacle avoidance path with the lowest comprehensive cost to output the control instruction and the corresponding speed instruction, and when it is detected that the distance to the obstacle is less than the safety threshold, the emergency stop protocol is triggered.

[0079] In the embodiment, the comprehensive cost of each candidate path is calculated as follows: C = w1 x C1 + w2 x C2 + w3 x C3; Wherein, C1 is distance cost, C2 is smoothness cost, C3 is safety cost, w1, w2, w3 are corresponding weight coefficients respectively. In a specific implementation, w1=0.4; w2=0.3; w3=0.3.

[0080] Calculate the comprehensive cost of all candidate trajectories, and sort them in ascending order.

[0081] Select the linear velocity and angular velocity corresponding to the minimum comprehensive cost, and output.

[0082] When the distance of the detected obstacle is less than the safety threshold, the linear velocity and angular velocity are set to zero, and an alarm is triggered.

[0083] S4, during approaching the target target along the obstacle avoidance path, trigger the laser shooting when the target target meets the preset spatial position condition and attitude stability condition, and output the target target positioning and shooting verification result, specifically including the following steps.

[0084] S41, obtain the target target three-dimensional coordinates, determine the rotation matrix of the current pose of the unmanned aerial vehicle through the inertial measurement unit data, and convert the target target three-dimensional coordinates in the body coordinate system to the target target world coordinates according to the rotation matrix; The inertial measurement unit data includes the roll angle α, the pitch angle β and the yaw angle γ of the unmanned aerial vehicle, and a second rotation matrix is constructed according to the inertial measurement unit data, Suppose the coordinates of the target target in the body coordinate system are P2=(X2, Y2, Z2), and the position of the unmanned aerial vehicle in the world coordinate system is T2=(x_world, y_world, z_world), then the coordinates P3 of the target target in the world coordinate system are: P3=RXP2+T2; The judgment of the shooting condition needs to be based on the absolute position relationship between the unmanned aerial vehicle and the target, rather than the relative position in the body coordinate system. The body coordinate system changes with the attitude of the unmanned aerial vehicle. By converting the target three-dimensional coordinates to the world coordinate system through the second rotation matrix, the Euclidean distance and the image offset can be stably calculated, and false judgment caused by the change of the attitude of the unmanned aerial vehicle can be avoided.

[0085] S42, calculate the Euclidean distance between the target target and the unmanned aerial vehicle in the world coordinate system; Convert the world coordinates to the image coordinate system through the camera calibration parameters, calculate the coordinate difference of the target target in the image coordinate system between the current frame and the previous frame, and the coordinate difference is the offset of the target target in the image coordinate system.

[0086] When the Euclidean distance between the target target and the unmanned aerial vehicle reaches the preset shooting threshold and the offset is within the preset spatial range, it is determined whether the target target meets the preset spatial position condition.

[0087] S43, if the target target meets the spatial position condition, the displacement standard deviation of the target center pixel coordinates in the continuous K frames is calculated according to the image frame obtained by the camera; In this embodiment, the average value of the target center pixel coordinates in the K frames in the x and y directions is calculated, and the displacement standard deviation σ u and the displacement standard deviation σ v in the y direction is calculated according to the average value and the standard deviation formula. S44, the displacement velocity is calculated according to the displacement standard deviation, and the displacement velocity is If the obtained displacement velocity meets the preset stable velocity value, that is, less than or equal to the stable velocity value, the simulation laser shooting of the unmanned aerial vehicle on the target target is triggered.

[0088] The simulation laser shooting can use a hardware trigger type laser pen, which is automatically activated when the condition is met.

[0089] S45, the position of the laser spot in the image frame obtained after the simulation laser trigger is detected, the Euclidean distance deviation e of the laser spot position and the target center pixel coordinates is calculated, and the current shooting deviation threshold is determined according to the shooting distance; The shooting deviation threshold e_threshold = e x k; Wherein, k is a proportional coefficient, in this embodiment, k = 1.2; in actual implementation, it can be adjusted.

[0090] S46, the hit confidence is calculated according to the Euclidean distance deviation and the shooting deviation threshold; If e ≤ e_threshold, the hit confidence Otherwise, F = 0.

[0091] If the hit confidence exceeds the preset confidence threshold, the grid where the target target is located is marked as a traversable area, and the subsequent flight efficiency is improved.

[0092] S47, the target image region feature template at the hit moment is extracted, and the visual recognition model is updated according to the target image region feature template.

[0093] S48, if the hit confidence does not exceed the preset confidence threshold, the position of the laser spot is recorded, which is converted into the coordinates in the world coordinate system through camera inverse projection, and a temporary obstacle avoidance point is generated with a safety distance set as the center of the world coordinates of the laser spot position, and the obstacle avoidance path is regenerated to improve the accuracy of the next shooting.

[0094] The embodiment obtains environment information through multi-sensor fusion, constructs an accurate environment map, generates a reasonable obstacle avoidance path, realizes accurate shooting of the target center in combination with visual recognition and attitude judgment, and performs feedback and optimization through shooting verification results. Compared with the prior art, the multi-sensor fusion solves the limitations of a single sensor in a complex environment, improves the positioning accuracy and obstacle avoidance capability of the unmanned aerial vehicle, and the visual recognition and shooting verification mechanism ensures accurate shooting of the target center, has stronger environmental adaptability and higher reliability, and can be widely applied to precise operations in indoor and outdoor environments without GPS.

[0095] Based on the same inventive concept, the embodiment of the present application also discloses an accurate positioning device for a target center of an unmanned aerial vehicle, which is constructed as shown in Figure 4 The device comprises the following modules: A data synchronization module is configured to obtain environment distance information in cooperation with a vertical direction laser ranging through a laser radar, time-synchronize the environment distance information and obtained inertial measurement unit data, and establish multi-sensor perception data in a unified coordinate system; an environment map construction module is configured to construct an environment map in real time based on the multi-sensor perception data, perform visual recognition on a target target through a camera, and calculate a three-dimensional position of the target target in the unified coordinate system; An obstacle avoidance planning module is configured to detect obstacles in the surrounding environment of the target target and generate an obstacle avoidance path to the target target by using a preset clustering algorithm according to the environment map and the three-dimensional position; A simulated shooting module is configured to trigger laser shooting when the target target meets preset spatial position conditions and attitude stability conditions in the process of approaching the target target along the obstacle avoidance path, and output target positioning and shooting verification results.

[0096] In a specific implementable scheme, the data synchronization module comprises the following units: A first data synchronization unit is configured to perform horizontal scanning through a two-dimensional laser radar according to a preset first scanning frequency, obtain plane point cloud data, wherein each point carries vertical direction laser ranging according to a preset second scanning frequency through a TOF principle, obtain unmanned aerial vehicle vertical distance data, and obtain inertial measurement unit data in the running process of the unmanned aerial vehicle through an IMU installed on the unmanned aerial vehicle; A second data synchronization unit is configured to time-synchronize the plane point cloud data, the vertical distance data and the inertial measurement unit data through a ROS message filtering mechanism; A third data synchronization unit is configured to load a pre-calibrated transformation matrix, the transformation matrix describes a fixed transformation relationship between multi-sensor perception data including the plane point cloud data, the vertical distance data and the inertial measurement unit data, and converts the multi-sensor perception data to a body coordinate system of the unmanned aerial vehicle in real time.

[0097] In one specific implementation, the environment map construction module comprises the following units: A first environment map construction unit is configured to compensate for motion distortion during laser radar scanning according to angular velocity in the inertial measurement unit data to obtain de-distorted point cloud data, and obtain a 2.5-dimensional grid environment map according to the de-distorted point cloud data and the vertical distance data; A second environment map construction unit is configured to obtain an image frame captured by a camera, extract visual features of the image frame using a preset visual recognition model to obtain a target center pixel coordinate of a target center of a target in the image frame, and obtain a three-dimensional coordinate of the target center of the target in a body coordinate system in combination with the vertical distance data and camera intrinsic data of the camera.

[0098] In one specific implementation, the obstacle avoidance planning module comprises the following units: A first obstacle avoidance planning unit is configured to receive the environment map and the three-dimensional coordinate of the target, define a region of interest centered on the target, and extract all obstacle grids in the region of interest; A second obstacle avoidance planning unit is configured to perform density clustering using an improved DBSCAN algorithm to obtain a plurality of effective obstacle clusters; A third obstacle avoidance planning unit is configured to perform inflation processing on each obstacle cluster to obtain a no-fly zone, mark the remaining space excluding the no-fly zone as a three-dimensional flyable space, generate a plurality of candidate obstacle avoidance paths in a velocity space that satisfy preset constraint conditions using a dynamic window method, and perform multi-objective cost evaluation on the candidate obstacle avoidance paths, wherein the evaluation indexes of each candidate obstacle avoidance path include distance cost, smoothness cost, and safety cost; A fourth obstacle avoidance planning unit is configured to select a candidate obstacle avoidance path with the lowest comprehensive cost to output a control instruction, and trigger an emergency stop protocol when the distance to an obstacle is detected to be less than a safety threshold.

[0099] In one specific implementation, the second obstacle avoidance planning unit comprises the following sub-units: A first obstacle avoidance planning sub-unit is configured to obtain continuous N frames of planar point cloud data scanned by the laser radar, calculate the pose consistency of each grid in adjacent sub-maps through a sub-map matching mechanism of the Cartographer algorithm, and determine a static obstacle when the pose satisfies a preset first offset condition, and map the static obstacle to a persistent layer of the environment map; and a second obstacle avoidance planning sub-unit is configured to calculate an instantaneous velocity vector of the obstacle cluster, determine a dynamic obstacle when the velocity vector satisfies a preset second offset condition, and predict a motion trajectory through linear regression of continuous M frames of historical data. The third obstacle avoidance planning subunit is configured to trigger TOF vertical scanning verification when the obstacle satisfies a preset semi-static obstacle determination condition, and determine the obstacle as a dynamic obstacle if a height variation rate of the obstacle exceeds a preset height variation threshold.

[0100] In one specific implementation, the simulation shooting module comprises the following units: The first simulation shooting unit is configured to obtain target three-dimensional coordinates of a target, determine a rotation matrix of a current pose of the UAV through inertial measurement unit data, and convert the target three-dimensional coordinates in a body coordinate system into target world coordinates according to the rotation matrix. The second simulation shooting unit is configured to determine whether the target satisfies a preset spatial position condition when a Euclidean distance between the target and the UAV reaches a preset shooting threshold and an offset of the target in an image coordinate system is within a preset spatial range. The third simulation shooting unit is configured to calculate a displacement standard deviation of target center pixel coordinates in K consecutive image frames according to the image frames obtained by the camera if the target satisfies the spatial position condition, and trigger simulation laser shooting of the target by the UAV if a displacement velocity obtained by the calculation satisfies a preset stable velocity value.

[0101] The fourth simulation shooting unit is configured to detect a laser spot position in an image frame obtained after the simulation laser is triggered, calculate a Euclidean distance deviation of the laser spot position from the target center pixel coordinates, determine a current shooting deviation threshold according to a shooting distance, calculate a hit confidence according to the Euclidean distance deviation and the shooting deviation threshold, and mark a grid where the target is located as a traversable area if the hit confidence exceeds a preset confidence threshold. The fifth simulation shooting unit is configured to extract a target image region feature template at a hit moment, and update a visual recognition model according to the target image region feature template. The sixth simulation shooting unit is configured to record the laser spot position and generate a temporary obstacle avoidance point if the hit confidence does not exceed the preset confidence threshold, and regenerate an obstacle avoidance path.

[0102] Based on the same inventive concept, the embodiments of the present application further disclose a computer readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set can be loaded and executed by a processor to implement the UAV target center accurate positioning method provided by the above method embodiments.

[0103] Similarly, based on the same inventive concept, the embodiments of the present application further disclose a computer readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set can be loaded and executed by a processor to implement the UAV target center accurate positioning method provided by the above method embodiments.

[0104] Those skilled in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or can be instructed by programs to complete the related hardware, and the programs can be stored in the computer readable storage medium, such as U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and various storage medium capable of storing program codes.

[0105] The above only describes optional embodiments of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for accurately locating a target using a drone, characterized in that: The steps include: Acquire environmental distance information through the collaboration of LiDAR and vertical laser ranging, synchronize the environmental distance information with the acquired inertial measurement unit data, and establish multi-sensor perception data in a unified coordinate system; Building an environmental map in real time based on the multi-sensor perception data, visually identifying a target through a camera and calculating the three-dimensional position of the target in the unified coordinate system; According to the environmental map and the three-dimensional position, a preset clustering algorithm is used to detect obstacles in the surrounding environment of the target and generate an obstacle avoidance path to the target; In the process of approaching the target along the obstacle avoidance path, when the target meets the preset spatial position conditions and posture stability conditions, laser shooting is triggered, and target positioning and shooting verification results are output.

2. The method for accurately locating a target using a drone according to claim 1, wherein: The method of collaboratively acquiring environmental distance information by using a laser radar and vertical laser ranging, time-synchronizing the environmental distance information with the acquired inertial measurement unit data, and establishing multi-sensor perception data in a unified coordinate system specifically includes the following steps: The two-dimensional laser radar performs horizontal scanning at a preset first scanning frequency to obtain planar point cloud data, where each point carries vertical laser ranging using the TOF principle at a preset second scanning frequency to obtain the vertical distance data of the drone. The inertial measurement unit data of the drone during operation is obtained in real time through the IMU installed on the drone. Time synchronization of the planar point cloud data, the vertical distance data, and the inertial measurement unit data is performed through a ROS message filtering mechanism; Load a pre-calibrated transformation matrix, which describes a fixed transformation relationship between multi-sensor perception data including the planar point cloud data, the vertical distance data, and the inertial measurement unit data, and convert the multi-sensor perception data into the drone's body coordinate system in real time.

3. The method for accurately locating a target's center using a drone according to claim 2, wherein: The real-time construction of an environment map based on the multi-sensor perception data, visual recognition of a target through a camera, and calculation of the three-dimensional position of the target in the unified coordinate system specifically include the following steps: Compensating for motion distortion during laser radar measurement based on the angular velocity in the inertial measurement unit data to obtain dedistorted point cloud data, and obtaining a 2.5-dimensional grid environment map based on the dedistorted point cloud data and the vertical distance data; Acquire an image frame captured by a camera, extract visual features from the image frame using a preset visual recognition model, obtain the pixel coordinates of the bull's eye of the target in the image frame, and combine the vertical distance data and the camera intrinsic parameter data of the camera to obtain the three-dimensional coordinates of the bull's eye of the target in the body coordinate system.

4. The method for accurately locating a target's center using a drone according to claim 1, wherein: The method of detecting obstacles in the surrounding environment of the target and generating an obstacle avoidance path to the target using a preset clustering algorithm based on the environment map and the three-dimensional position specifically includes the following steps: Receiving the environment map and the three-dimensional coordinates of the target, defining a region of interest with the target as the center, and extracting all obstacle grids within the region of interest; The improved DBSCAN algorithm is used to perform density clustering and obtain several valid obstacle clusters; Expanding each obstacle cluster to obtain a no-fly zone, marking the remaining space excluding the no-fly zone as a three-dimensional flyable space, generating multiple candidate obstacle avoidance paths that meet preset constraints in velocity space using a dynamic window method, and performing a multi-objective cost evaluation on the candidate obstacle avoidance paths, where the evaluation metrics for each candidate obstacle avoidance path include distance cost, smoothness cost, and safety cost; The candidate obstacle avoidance path with the lowest comprehensive cost is selected to output a control instruction, and when it is detected that the obstacle distance is less than a safety threshold, an emergency stop protocol is triggered.

5. The method for accurately locating a target's center using a drone according to claim 4, wherein: The step of obtaining a plurality of valid obstacle clusters and their boundary polygons includes classification and identification of static obstacles and dynamic obstacles, which specifically includes the following steps: Obtain N consecutive frames of planar point cloud data from LiDAR scanning, and calculate the pose consistency of each grid in adjacent submaps through the submap matching mechanism of the Cartographer algorithm. When the pose meets the preset first offset condition, it is determined to be a static obstacle; Calculating the instantaneous velocity vector of the obstacle cluster, determining it as a dynamic obstacle when the velocity vector meets a preset second offset condition, and predicting its motion trajectory through linear regression of M consecutive frames of historical data; When the obstacle meets the preset semi-static obstacle determination condition, the TOF vertical scanning verification is triggered. If the height change rate of the obstacle exceeds the preset height change threshold, the obstacle is determined to be the dynamic obstacle.

6. The method for accurately locating a target's center using a drone according to claim 3, wherein: When the target meets the preset spatial position condition and posture stability condition, triggering laser shooting specifically includes the following steps: Obtaining the three-dimensional coordinates of the target, determining the rotation matrix of the current posture of the drone using the inertial measurement unit data, and converting the three-dimensional coordinates of the target in the body coordinate system into target world coordinates based on the rotation matrix; When the Euclidean distance between the target and the UAV reaches a preset shooting threshold and the offset of the target in the image coordinate system is within a preset spatial range, determining whether the target meets a preset spatial position condition; If the target meets the spatial position condition, the displacement standard deviation of the pixel coordinates of the bull's eye in consecutive K frames is calculated based on the image frames acquired by the camera. If the obtained displacement speed meets the preset stable speed value, the drone is triggered to simulate laser shooting at the target.

7. The method for accurately locating a target's center using a drone according to claim 6, wherein: The following steps are also included: Detecting the laser spot position in the image frame obtained after the simulated laser is triggered, calculating the Euclidean distance deviation between the laser spot position and the pixel coordinates of the bull's eye, determining a current shooting deviation threshold according to the shooting distance, calculating a hit confidence level according to the Euclidean distance deviation and the shooting deviation threshold, and if the hit confidence level exceeds a preset confidence threshold, marking the grid where the target is located as a traversable area; Extracting a target image region feature template at the time of hitting, and updating the visual recognition model based on the target image region feature template; If the hit confidence does not exceed a preset confidence threshold, the laser spot position is recorded and a temporary obstacle avoidance point is generated, and the obstacle avoidance path is regenerated.

8. A device for accurately positioning a target on a target by a drone, characterized in that: Includes the following modules: A data synchronization module is used to acquire environmental distance information through the collaboration of the laser radar and the vertical laser ranging, synchronize the environmental distance information with the acquired inertial measurement unit data, and establish multi-sensor perception data in a unified coordinate system; An environment map construction module is used to construct an environment map in real time based on the multi-sensor perception data, perform visual recognition of a target through a camera, and calculate the three-dimensional position of the target in the unified coordinate system; an obstacle avoidance planning module, configured to detect obstacles in the surrounding environment of the target and generate an obstacle avoidance path to the target using a preset clustering algorithm based on the environment map and the three-dimensional position; The simulation shooting module is used to trigger laser shooting when the target meets preset spatial position conditions and posture stability conditions during the process of approaching the target along the obstacle avoidance path, and output target positioning and shooting verification results.

9. An intelligent terminal, characterized in that: The method comprises a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the method for accurately locating the center of a target by a drone as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The readable storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the method for accurately positioning the center of a target by a drone as described in any one of claims 1 to 7.

Images