Unmanned aerial vehicle laser obstacle removing system and obstacle removing method for improving laser obstacle removing efficiency

Abstract

The invention provides an unmanned aerial vehicle laser obstacle clearance system and method for improving laser obstacle clearance efficiency, the obstacle clearance system comprises an unmanned aerial vehicle, an airborne computer carried on the unmanned aerial vehicle, a stability augmentation holder, a holder camera and a laser transmitter, and the unmanned aerial vehicle, the holder camera and the laser transmitter are all fixed on the stability augmentation holder; the pan-tilt camera is used for collecting a trunk swing video stream in real time and transmitting the video stream to the visual module of the airborne computer; the visual module extracts tree trunk skeleton tracking points in the video stream; a control module of the airborne computer generates a holder direction control instruction and a laser trigger signal according to the tracking point coordinate position; the stability augmentation holder carries out real-time attitude adjustment according to the holder direction control instruction; and the laser emitter performs laser cutting on the swing tree trunk after receiving the laser trigger signal. According to the invention, real-time tracking of the target local area of the swing tree trunk is realized by using the unmanned aerial vehicle and the tracking algorithm, and the laser obstacle removing precision and the temperature rising efficiency are improved.

Description

Technical Field

[0001] The present application relates to the technical field of laser obstacle removal, and in particular, to a drone laser obstacle removal system and an obstacle removal method for improving laser obstacle removal efficiency. Background Art

[0002] With the rapid development of China's economy and the advancement of the urbanization process, the scale of China's power grid has been continuously expanding, widely covering urban and rural areas and various terrains. As a key channel for power transmission, the safe and stable operation of transmission lines is the cornerstone of the orderly operation of the social economy. Transmission lines often pass through complex terrains such as mountains, forests, and farmlands. The problem of tree obstacles has become a prominent hidden danger threatening the normal operation of power lines, and there are also many limitations in the current tree obstacle cleaning and maintenance technologies, including:

[0003] (1) Traditional obstacle removal methods: Manual climbing towers, throwing ropes for pulling, and mechanical cutting require power outage operations, with risks of high-altitude falling and electric shock, resulting in large economic losses.

[0004] (2) Ground laser obstacle removers: Installed on the ground, restricted by terrain (such as mountains and hills), difficult to deploy, with cleaning dead angles; relying on manual aiming, unable to track the swinging of trees in real time, with low cutting accuracy, resulting in impaired continuity of obstacle removal operations and low overall efficiency.

[0005] (3) Robot obstacle removal technology: Ground robots (manipulators), tree-climbing robots, cable robots, knife-equipped drones, etc. These intelligent devices can operate autonomously or remotely, and can handle tree obstacles without power outage. However, they have high requirements for technology and funds, and there are also problems of low reliability and efficiency. Regarding target tracking algorithms: Currently, most existing target tracking algorithms focus on tracking the overall motion trajectory of moving objects. For the scenario of tree obstacle laser cutting, which focuses on tracking local tree branch areas, there is little relevant research; and existing tracking algorithms (such as those based on feature matching or target detection) have insufficient robustness in complex backgrounds (such as leaf occlusion and light changes), and are prone to losing the target; while deep learning-based target tracking algorithms (such as SiamFC) have high computational costs and are difficult to meet the real-time requirements of drones.

[0006] In view of this, the present application provides a drone laser obstacle removal system and an obstacle removal method for improving laser obstacle removal efficiency, so as to achieve real-time tracking and precise laser cutting of dynamic swinging tree obstacles by drones in complex terrains. Summary of the Invention

[0007] In order to achieve the above object, the technical solutions adopted in the present application are as follows:

[0008] To achieve the above object, the technical solutions adopted in the present application are as follows: A drone laser obstacle clearing system for improving laser obstacle clearing efficiency, comprising: a drone, an on-board computer carried on the drone, a stabilization gimbal, a gimbal camera, and a laser emitter. The drone, the gimbal camera, and the laser emitter are all fixed on the stabilization gimbal. The laser emitter is fixedly installed directly below the gimbal camera. The on-board computer includes a vision module, a control module, and a communication module; The gimbal camera is used to collect the trunk swing video stream in real time and transmit it to the vision module of the on-board computer. The vision module of the on-board computer uses the LK pyramid optical flow algorithm and the target feature screening mechanism to extract the trunk skeleton tracking points in the video stream. The control module of the on-board computer generates a gimbal direction control command and a laser trigger command according to the tracking point coordinate position. The communication module sends the gimbal direction control command to the stabilization gimbal. The stabilization gimbal makes real-time attitude adjustments according to the gimbal direction control command to adjust the irradiation angle of the laser emitter. The laser emitter performs laser cutting on the swinging trunk after receiving the laser trigger command.

[0009] Further, the drone is a quadcopter drone, and the stabilization gimbal is a three-axis stabilization gimbal.

[0010] Further, the stabilization gimbal communicates with the on-board computer through a CAN bus.

[0011] Further, the on-board computer further includes an automatic search module for automatically entering the search mode when the tracking points are lost.

[0012] A method for a drone laser obstacle clearing system for improving laser obstacle clearing efficiency, applied to the above-mentioned drone laser obstacle clearing system for improving laser obstacle clearing efficiency. The method includes: S1: The gimbal camera collects the trunk swing video stream in real time and transmits it to the vision module of the on-board computer; S2: The vision module uses the LK pyramid optical flow algorithm and the target feature screening mechanism to extract the trunk skeleton tracking points in the video stream. The target feature screening mechanism includes the Shi-Tomasi feature point (corner point) detection algorithm and the RANSAC algorithm; S3: The control module of the on-board computer generates a gimbal direction control command and a laser trigger command according to the tracking point coordinate position; S4: The communication module sends the gimbal direction control command to the stabilization gimbal; S5: The stabilization gimbal makes real-time attitude adjustments according to the gimbal direction control command to adjust the irradiation angle of the laser emitter; S6: After receiving the laser trigger command, the laser emitter with the adjusted angle performs laser cutting on the swinging trunk.

[0013] Further, the specific steps of S2 include: S2.1: Read the current image frame and convert it into a grayscale image; S2.2: Calculate the R value of all pixel points on the grayscale image through the quality evaluation function of the Shi-Tomasi feature point detection algorithm. The quality evaluation function of the Shi-Tomasi feature point detection algorithm is:

[0014] where represents the second-order derivative of the image in the x direction (the element of the Hessian matrix), reflecting the edge intensity; represents the second-order derivative of the image in the y direction, reflecting the edge intensity; S2.3: Screen the feature points through non-maximum suppression. The feature points refer to the edge points of the tree trunk or the points at the texture change. Set the threshold of the R value to 0.01 and the maximum number of feature points to 100; Screen out the pixel points that meet the following spatial constraint conditions as feature points: pixel where and represent pixel points, and S represents the set of all pixel points; S2.4: Execute the LK pyramid optical flow algorithm to construct an image pyramid: Perform Gaussian blur and downsampling on the grayscale image converted in step S2.1 to generate multi-scale image layers; S2.5: Iteratively calculate from top to bottom: Starting from the top layer, calculate the sparse optical flow, that is, estimate the rough displacement of the feature points, transfer the displacement result to the next layer, and amplify the displacement amount for more refined displacement correction; Repeat the above process until the original resolution layer to complete the total number of iterations; The optical flow calculation equation is: Let the motion vector of the feature point P=(x,y) between consecutive frames be (u,v), then the optical flow equation can be expressed as: (x + uΔt, y + vΔt, t + Δt) = (x, y, t) The iterative equation is: The optical flow equation is expanded by Taylor and discretized to obtain the iterative equation:

[0015] where represents the image gradient, represents the gradient of the feature point P in the x direction, represents the gradient of the feature point P in the y direction, represents the motion time interval, is the time series change amount, and k represents the number of iterations; S2.6: Final displacement synthesis: After the correction of each layer, the displacements are accumulated to obtain the final optical flow vector of the feature point P.

[0016] S2.7: Execute the RANSAC algorithm to eliminate the mismatched points; S2.8: Centroid calculation: After successful optical flow tracking, collect all valid feature points, and calculate the average of the coordinates of the valid feature points to obtain the centroid position, that is, the trunk skeleton tracking point.

[0017] Further, the S3 specifically includes: S3.1: Calculate the offset according to the centroid position and the image center and normalize the offset; S3.2: Control mapping: Convert the normalized offset into the pan-tilt angular velocity, and generate the pan-tilt direction control instruction based on the pan-tilt angular velocity.

[0018] Further, the laser trigger signal includes the laser emission power and the emission frequency.

[0019] Compared with the prior art, the beneficial effects of the present application are: 1. The present application uses a drone and a tracking algorithm to achieve real-time tracking of the target local area of the swinging tree trunk, solves the problem of laser spot offset caused by tree swinging, and improves the accuracy of laser obstacle clearance.

[0020] 2. The target tracking algorithm (such as the LK pyramid optical flow algorithm) of the present application has a small computational cost and can meet the real-time requirements of the drone.

[0021] 3. Under the wind speed of level 5 (10.5 m / s), the present application can achieve a spot tracking error ≤ 5 cm and an optical power loss rate < 5%, which can greatly improve the laser focus heating efficiency.

[0022] 4. The efficiency of the present application is improved compared with the traditional manual operation, and the number of tree obstacles that can be cleared in a single task is increased.

[0023] 5. The present application supports deployment in complex terrains and expands the operation radius coverage. Description of the Drawings

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative efforts.

[0025] Figure 1 It is a structural diagram of a drone laser obstacle clearance system for improving the laser obstacle clearance efficiency of the present application.

[0026] Figure 2 This is the structural block diagram of the on-board computer of this application.

[0027] Figure 3 This is the flowchart of the UAV laser obstacle clearing method for improving laser obstacle clearing efficiency in this application. Specific implementation manners

[0028] To make the objectives, technical solutions and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Apparently, the described embodiments are some but not all of the embodiments of this application.

[0029] As Figure 1 and Figure 2 shown, a UAV laser obstacle clearing system for improving laser obstacle clearing efficiency includes: a UAV, an on-board computer carried on the UAV, a stabilization gimbal, a gimbal camera and a laser emitter. The UAV, the gimbal camera and the laser emitter (fiber optic probe) are all fixed on the stabilization gimbal. The laser emitter is fixedly installed directly below the gimbal camera. The on-board computer includes a vision module, a control module and a communication module; The gimbal camera is used to collect the trunk swing video stream in real time and transmit it to the vision module of the on-board computer. The vision module of the on-board computer uses the LK pyramid optical flow algorithm and the target feature screening mechanism to extract the trunk skeleton tracking points in the video stream. The control module of the on-board computer generates a gimbal direction control instruction and a laser trigger instruction according to the tracking point coordinate position. The communication module sends the gimbal direction control instruction to the stabilization gimbal. The stabilization gimbal makes real-time attitude adjustments according to the gimbal direction control instruction to adjust the irradiation angle of the laser emitter. The laser emitter performs laser cutting on the swinging trunk after receiving the laser trigger instruction.

[0030] Specifically, the gimbal camera can be an OV OS12D40 CMOS image sensor with a resolution of 2.7K@60fps, a wide-angle lens (FOV 143°), and transmits images through RTSP streaming media. Communicates with the on-board computer through the CAN bus, and the instruction transmission delay < 8ms.

[0031] When collecting images, the UAV can autonomously fly to a position 5m away from the trunk. The gimbal camera collects the trunk swing video in real time, and transmits the picture to the vision module through the on-board computer.

[0032] When the laser emitter is installed, it is rigidly coupled with the output shaft of the gimbal camera to realize synchronous control of the beam pointing and the gimbal rotation.

[0033] The model of the on-board computer can be Allspark2 X86; the appearance size is selected as 109.5*72.5*31.67mm; the weight is 204g; the CPU model is Intel Core i7-1165G7, with 4 cores and 8 threads, and the maximum turbo frequency is 4.7GHZ; the maximum dynamic frequency of the GPU is 1.3GHz, and it is configured with an Intel Iris X graphics card.

[0034] Further, the drone is a quadcopter drone, and the stabilization gimbal is a three-axis stabilization gimbal.

[0035] In specific implementation, the drone is a quadcopter with a load capacity of ≥5kg; the gimbal is a three-axis stabilization gimbal, the yaw angle adjustment range is ±30°, and the three axes are independently controlled (pitch axis +90°~-30°, roll axis ±45°, yaw axis ±60°), the maximum rotation speed is 180° / second, and the angle jitter is ≤±0.005°. The three-axis stabilization gimbal independently controls pitch (Pitch) and yaw (Yaw) to adapt to different cutting angle requirements.

[0036] When installing the laser emitter, the three-axis stabilization gimbal realizes the quick loading and unloading of the gimbal and the laser emitter through a quick-release interface.

[0037] Further, the stabilization gimbal communicates with the on-board computer through the CAN bus.

[0038] Specifically, the communication baud rate of the CAN bus can be 115200bps.

[0039] Further, the on-board computer further includes an automatic search module for automatically entering the search mode when the tracking point is lost.

[0040] The automatic search module belongs to the exception handling mechanism function module. When the tracking point is lost, it automatically enters the search mode (the needToInit flag is set). In addition, the exception handling mechanism can also trigger a security protocol (gimbalNoTrack()) when there are no valid tracking points for 5 consecutive frames.

[0041] The gimbal attitude reset of this application can adopt an adaptive zeroing algorithm (setHome()), and the termination condition of the target tracking process can be that the tracking point is cleared manually through a button or the gimbal is reset.

[0042] As Figure 3 shown, a method for laser obstacle clearing of a drone to improve laser obstacle clearing efficiency, which is applied to the system for laser obstacle clearing of a drone to improve laser obstacle clearing efficiency, the method includes: S1: The gimbal camera continuously collects the trunk swing video stream and transmits it to the vision module of the on-board computer; Specifically, a 720P@30fps video stream can be obtained through the OpenCV image acquisition interface.

[0043] S2: The vision module uses the LK pyramid optical flow algorithm and the target feature screening mechanism to extract the trunk skeleton tracking points in the video stream. The target feature screening mechanism includes the Shi-Tomasi feature point (corner point) detection algorithm and the RANSAC algorithm; S3: The control module of the airborne computer generates a pan-tilt direction control command and a laser trigger command according to the tracking point coordinate position; S4: The communication module sends the pan-tilt direction control command to the stabilized pan-tilt; S5: The stabilized pan-tilt makes real-time attitude adjustments according to the pan-tilt direction control command to adjust the irradiation angle of the laser emitter; S6: After receiving the laser trigger command, the laser emitter with the adjusted angle performs laser cutting on the swinging trunk.

[0044] Furthermore, the specific steps of S2 include: S2.1: Read the current image frame and convert it into a grayscale image; S2.2: Calculate the R value of all pixel points on the grayscale image through the quality evaluation function of the Shi-Tomasi feature point detection algorithm. The quality evaluation function of the Shi-Tomasi feature point detection algorithm is:

[0045] where represents the second-order derivative of the image in the x direction (the element of the Hessian matrix), reflecting the edge strength; represents the second-order derivative of the image in the y direction, reflecting the edge strength.

[0046] The physical meaning of the quality evaluation function is: calculate the minimum eigenvalue of the structure tensor to measure the response strength of the pixel point. A high R value indicates that there are obvious texture changes (such as edge intersection points) around the point; a low R value indicates that the point is in a flat area or a weak texture area.

[0047] S2.3: Screen the feature points through non-maximum suppression. The feature points refer to the edge points of the trunk or the points at the texture change. Set the threshold of the R value to 0.01 and the maximum number of feature points to 100; screen out the pixel points that meet the following spatial constraint conditions as feature points: pixel where and Denote a pixel point, and \(S\) represents the set of all pixel points; the threshold is a relative threshold rather than an absolute threshold, which is defaulted to 0.01 in the quality level of this algorithm, indicating that all pixel points with \(R\) value greater than 1% of the global maximum eigenvalue (\(R_{max}\)) of the image are retained. In simple scenarios in practical applications (such as artificial textures), generally take 0.01 - 0.05, and in complex scenarios (such as natural images), generally take 0.001 - 0.01.

[0048] S2.4: Execute the LK pyramid optical flow algorithm to construct an image pyramid: perform Gaussian blur and downsampling on the grayscale image (such as prevGray) transformed in step S2.1 to generate multi-scale image layers; Specifically, the multi-scale image layers (the number of pyramid layers) are set to 4 layers, and the scaling factor is 0.5; the scaling factor represents the resolution scaling ratio between adjacent layers when constructing the image pyramid. For example: the 0th layer is the original image (1280×720), the 1st layer is 640×360, and so on to the 4th layer.

[0049] S2.5: Iteratively calculate from top to bottom: start from the topmost layer (the smallest resolution), calculate the sparse optical flow, that is, estimate the rough displacement of the feature points, transfer the displacement result to the next layer, and amplify the displacement amount (multiply by the reciprocal of the scaling factor 0.5) for more refined displacement correction; repeat the above process until the original resolution layer (the 0th layer), and complete 10 total iterations; The optical flow calculation equation is: Let the motion vector of the feature point \(P=(x,y)\) between consecutive frames be \((u,v)\), then the optical flow equation can be expressed as: (x + uΔt, y + vΔt, t + Δt) = (x,y,t) The iterative equation is: The optical flow equation is expanded by Taylor series and discretized to obtain the iterative equation:

[0050] where represents the image gradient, represents the gradient of the feature point \(P\) in the \(x\) direction, represents the gradient of the feature point \(P\) in the \(y\) direction, represents the motion time interval, is the time sequence change amount, and \(k\) represents the number of iterations.

[0051] S2.6: Final displacement synthesis: After the correction of each layer, perform displacement accumulation to obtain the final optical flow vector of the feature point \(P\).

[0052] In specific implementation, the sampling frequency of the LK pyramid optical flow algorithm can be 30 Hz. The problem of large displacement tracking is solved by constructing an image pyramid (the number of pyramid layers is set to 4, and the scaling factor is 0.5). The size of the window (referring to the neighborhood search window size for each feature point in optical flow calculation) is set to 21×21 pixels, and the number of pyramid iterations is 10 times.

[0053] The Shi-Tomasi algorithm detects strong feature points (such as edge intersection points) in the image as the initial tracking target to ensure that the feature points are located in significant regions (such as the edge of the tree trunk). The LK optical flow algorithm relies on these initial points and predicts their positions in subsequent frames through spatio-temporal gradient information.

[0054] Supplementary note: In the code, goodFeaturesToTrack initializes feature points when the needToInit flag is triggered (such as after the first run or reset), and the optical flow continuously tracks these points in subsequent frames.

[0055] S2.7: Execute the RANSAC algorithm to eliminate mismatched points; During the feature point tracking process, the RANSAC algorithm is used to eliminate mismatched points. The number of iterations is set to 50 times, and the inlier ratio threshold is set to 0.8.

[0056] For example: Based on the current matching point pairs, a group of points is randomly sampled to estimate the optical flow model (such as affine transformation). The reliability of the model is judged according to the inlier ratio (>80%), and the inliers are retained as valid matching points.

[0057] Shi-Tomasi provides initial reliable feature points, and RANSAC further optimizes the tracking robustness. The two form a collaborative process of "feature extraction - mismatched point elimination".

[0058] S2.8: Centroid calculation: After successful optical flow tracking, all valid feature points are collected, and the centroid position is obtained by averaging the coordinates of the valid feature points, that is, the tracking point of the tree trunk skeleton.

[0059] The centroid, as the central position of the target area, replaces single-point tracking to suppress noise.

[0060] In the code, the centroid is used to calculate its offset (xOffset, yOffset) relative to the center of the image, and then control the rotation of the pan-tilt. The centroid is the coordinate position of the tracking point in the image (when the pan-tilt direction is not aligned with the tracking point, the tracking point is in a non-central position in the picture, and when the centroid position is at the center of the picture, it represents that the pan-tilt direction is facing the tracking point position). By calculating the offset of the centroid relative to the center of the image, the pan-tilt rotation is controlled to achieve the tracking function of the pan-tilt.

[0061] Furthermore, the specific content of S3 includes: S3.1: Calculate the offset based on the centroid position and the image center, and normalize the offset. Input: Centroid position (center) and image center (frame.cols / 2, frame.rows / 2).

[0062] Offset calculation: xOffset = (center.x / frame.cols) - 0.5; / / Normalize to [-0.5, 0.5] yOffset = (center.y / frame.rows) - 0.5; S3.2: Control mapping: Convert the normalized offset to the gimbal angular velocity (such as xOffset * 180° / s), and generate a gimbal direction control command based on the gimbal angular velocity. The three-axis gimbal realizes attitude adjustment through PID control and angular velocity feedback, and performs real-time calibration in combination with the dynamic target position.

[0063] Gimbal control logic: Tracking mode: Continuous tracking: When valid feature points exist, call gimbalTrack(xOffset, yOffset).

[0064] Control frequency: Synchronize with the frame rate (30Hz) to ensure real-time performance.

[0065] Motor response: Send the angular velocity command through the serial port ( / dev / ttyUSB0) to drive the three-axis gimbal to rotate.

[0066] Non-tracking mode: Stop tracking: When the feature points are lost, call gimbalNoTrack(), and set the angular velocity to 0.

[0067] Return mechanism: Trigger gimbal->setHome() to make the gimbal return to the preset safe position.

[0068] Furthermore, the laser trigger signal includes the laser emission power and the emission frequency.

[0069] Specifically, the laser emitter can cut at a pulse interval of 50ms until the tree trunk breaks.

[0070] ​This application combines a target feature screening mechanism with a pyramid optical flow algorithm to improve the tracking robustness in complex backgrounds; and uses centroid calculation to replace single-point tracking to reduce local noise interference. In addition, a dynamic tracking algorithm based on optical flow method is provided to achieve real-time alignment of the laser direction and the cutting point, improving the obstacle clearing efficiency; and solves the problems of tracking delay and insufficient accuracy of traditional laser obstacle clearing systems in dynamic tree obstacle scenarios.

[0071] The above are only specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims.

Claims

1. An unmanned aerial vehicle laser obstacle clearing system for improving laser obstacle clearing efficiency, characterized in that, Including: A drone, an on-board computer mounted on the drone, a stabilization gimbal, a gimbal camera, and a laser emitter. The drone, the gimbal camera, and the laser emitter are all fixed on the stabilization gimbal. The laser emitter is fixedly installed directly below the gimbal camera. The on-board computer includes a vision module, a control module, and a communication module. The gimbal camera is used to collect the trunk swing video stream in real time and transmit it to the vision module of the on-board computer. The vision module of the on-board computer uses the LK pyramid optical flow algorithm and the target feature screening mechanism to extract the trunk skeleton tracking points in the video stream. The control module of the on-board computer generates a gimbal direction control instruction and a laser trigger instruction according to the tracking point coordinate position. The communication module sends the gimbal direction control instruction to the stabilization gimbal. The stabilization gimbal makes real-time attitude adjustments according to the gimbal direction control instruction to adjust the irradiation angle of the laser emitter. The laser emitter performs laser cutting on the swinging trunk after receiving the laser trigger instruction.

2. The drone laser obstacle clearing system for improving the laser obstacle clearing efficiency as described in claim 1, wherein The drone is a quadcopter drone, and the stabilization gimbal is a three-axis stabilization gimbal.

3. The drone laser obstacle clearing system for improving the laser obstacle clearing efficiency as described in claim 1, characterized in that, The stabilization gimbal communicates with the on-board computer through the CAN bus.

4. The UAV laser obstacle clearing system for improving the laser obstacle clearing efficiency according to claim 1, wherein, The on-board computer further includes an automatic search module for automatically entering the search mode when the tracking point is lost.

5. A method for laser obstacle removal by an unmanned aerial vehicle to improve the efficiency of laser obstacle removal, which is applied to the system for laser obstacle removal by an unmanned aerial vehicle to improve the efficiency of laser obstacle removal, and is characterized in that The method includes: S1: The gimbal camera collects the trunk swing video stream in real time and transmits it to the vision module of the on-board computer. S2: The vision module uses the LK pyramid optical flow algorithm and the target feature screening mechanism to extract the trunk skeleton tracking points in the video stream. The target feature screening mechanism includes the Shi-Tomasi feature point detection algorithm and the RANSAC algorithm. S3: The control module of the on-board computer generates a gimbal direction control instruction and a laser trigger instruction according to the tracking point coordinate position. S4: The communication module sends the gimbal direction control instruction to the stabilization gimbal. S5: The stabilization gimbal makes real-time attitude adjustments according to the gimbal direction control instruction to adjust the irradiation angle of the laser emitter. S6: The laser emitter after adjusting the angle performs laser cutting on the swinging trunk after receiving the laser trigger instruction.

6. The method for clearing obstacles with a drone laser as claimed in claim 5, characterized in that, The specific steps of S2 include: S2.1: Read the current image frame and convert it into a grayscale image. S2.2: Calculate the R value of all pixel points on the grayscale image through the quality evaluation function of the Shi-Tomasi feature point detection algorithm. The quality evaluation function of the Shi-Tomasi feature point detection algorithm is: Among them, represents the second-order derivative of the image in the x direction (the element of the Hessian matrix), reflecting the edge intensity; represents the second-order derivative of the image in the y direction, reflecting the edge intensity; S2.3: Screen the feature points through non-maximum suppression. The feature points refer to the edge points of the trunk or the points at the texture change. Set the threshold of the R value to 0.01 and the maximum number of feature points to 100. Screen out the pixel points whose spatial distribution satisfies the following spatial constraint conditions as feature points: Pixel Among them, and represent pixel points, and S represents the set of all pixel points; S2.4: Execute the LK pyramid optical flow algorithm and construct an image pyramid: Perform Gaussian blur and downsampling on the grayscale image converted in step S2.1 to generate multi-scale image layers. S2.5: Top-down iterative calculation: Starting from the topmost layer, calculate the sparse optical flow, that is, estimate the rough displacement of feature points, transfer the displacement result to the next layer, and amplify the displacement amount for more refined displacement correction; repeat the above process until the original resolution layer to complete the total number of iterations; The optical flow calculation equation is: Assume that the motion vector of feature point P=(x, y) between consecutive frames is (u, v), then the optical flow equation can be expressed as: (x + uΔt, y + vΔt, t + Δt) = (x, y, t) The iterative equation is: The optical flow equation is expanded by Taylor series and discretized to obtain the iterative equation: Among them represents the image gradient represents the gradient of feature point P in the x direction represents the gradient of feature point P in the y direction represents the motion time interval is the time series change amount, and k represents the number of iterations S2.6: Final displacement synthesis: After the correction of each layer, perform displacement accumulation to obtain the final optical flow vector of feature point P; S2.7: Execute the RANSAC algorithm to eliminate mis-matched points; S2.8: Centroid calculation: After successful optical flow tracking, collect all valid feature points, and calculate the average of the coordinates of the valid feature points to obtain the centroid position, that is, the trunk skeleton tracking point.

7. The method for a drone laser obstacle clearing to improve the laser obstacle clearing efficiency according to claim 6, wherein, The specific steps of S3 include: S3.1: Calculate the offset according to the centroid position and the image center and normalize the offset; S3.2: Convert the normalized offset into the pan-tilt angular velocity, and generate a pan-tilt direction control instruction based on the pan-tilt angular velocity.

8. The method for a drone laser obstacle removal to improve the laser obstacle removal efficiency according to claim 7, characterized in that, The laser trigger signal includes the laser emission power and the emission frequency.

Images