Dual-base station laser correlation information and visual fusion guided unmanned aerial vehicle landing method

Abstract

The application discloses a dual-base station laser correlation information and visual fusion guided unmanned aerial vehicle landing method, which is used for realizing high-precision positioning and autonomous safe landing of an unmanned aerial vehicle in a complex environment. The method is characterized in that at least two laser base stations are arranged in a landing area, a time difference of arrival (TDOA) is extracted by performing cross-correlation operation on the basis of laser quantum correlation signals, and a three-dimensional hyperboloid geometric prior of a position of the unmanned aerial vehicle is constructed. A downward-looking camera on the unmanned aerial vehicle is used for visual odometry positioning and matching with remote sensing images to obtain absolute position information. The TDOA geometric constraint and visual pose information are fused, high-precision and low-drift fused pose estimation is realized through time alignment and sliding window graph optimization, a segmented landing trajectory model is further constructed, online trajectory adjustment is performed on the basis of real-time fused pose and speed information under energy constraint, and stable tracking of the trajectory is realized in combination with PID control, so that the unmanned aerial vehicle completes stable, safe and precise autonomous landing.

Description

Technical Field

[0001] This invention belongs to the field of UAV navigation, guidance and control technology, and mainly relates to a method for guiding UAV landing by dual-base station laser correlation information and visual fusion. Background Technology

[0002] With the widespread application of drones in emergency rescue, logistics transportation, and patrol and reconnaissance, autonomous landing capability has become one of the key technologies for ensuring flight safety and mission reliability. Existing drone landing systems typically rely on Global Navigation Satellite Systems (GNSS) for positioning and guidance. However, in urban canyons, complex terrain, environments with electromagnetic interference, or limited satellite signals, the accuracy and reliability of GNSS positioning significantly decrease, making it difficult to meet the engineering requirements for high-precision and safe landing. Therefore, there is an urgent need to develop autonomous landing guidance technologies with environmental adaptability and high reliability.

[0003] In recent years, visual perception-based drone landing methods have attracted widespread attention. These methods utilize onboard cameras to acquire ground images and estimate the drone's pose using visual odometry or visual SLAM techniques, achieving autonomous positioning with low dependence on external signals. However, visual methods are susceptible to changes in lighting, insufficient ground texture, and environmental variations, and inevitably suffer from cumulative drift during long-term operation, making it difficult to consistently guarantee landing accuracy. To improve system robustness, researchers have attempted to introduce multi-source information fusion techniques, jointly processing visual, inertial, laser, and radio positioning information. However, existing methods still suffer from limited accuracy and insufficient stability in complex environments.

[0004] Laser temporal differential positioning (TDOA) technology offers a novel approach to high-precision UAV positioning due to its advantages such as high measurement accuracy and strong anti-interference capabilities. In particular, the introduction of laser quantum correlation characteristics further enhances time measurement accuracy and system reliability. However, the deep integration of this technology with visual perception systems remains limited, and existing landing methods lack a unified design framework between positioning results, trajectory planning, and control, making it difficult to achieve stable, safe, and autonomous precise landings in complex environments. Therefore, it is necessary to propose a UAV landing guidance method that integrates laser quantum correlation positioning and visual perception information to comprehensively improve the accuracy, stability, and engineering applicability of autonomous UAV landing. Summary of the Invention

[0005] To address the problems in existing technologies, this invention proposes a UAV landing guidance method based on dual-base station laser correlation information and visual fusion. This method deploys at least two laser base stations in the landing area, using cross-correlation of laser signals to extract the time difference of arrival and construct a three-dimensional geometric constraint for the UAV's position. Simultaneously, visual odometry positioning is performed using an onboard downward-view camera on the UAV, and the absolute position information is obtained by matching it with remote sensing imagery to correct the visual trajectory. Laser geometric constraints and visual pose information are fused in a unified coordinate system, and high-precision, low-drift UAV fusion positioning is achieved through time alignment and sliding window graph optimization. Furthermore, a segmented landing trajectory model is constructed, and the nominal landing trajectory is adjusted online based on real-time fused pose and velocity information under energy constraints. Stable trajectory tracking is achieved through PID control, thereby guiding the UAV to complete a stable, safe, and accurate autonomous landing.

[0006] To achieve the above objectives, the technical solution provided by this invention is: a method for guiding UAV landing by fusion of dual-base station laser correlation information and visual information, the method comprising:

[0007] Step 1: Based on the laser quantum signals from the dual base stations, the time difference of arrival (TDOA) is extracted through cross-correlation calculation and transformed into a hyperboloid geometric prior that constrains the position of the UAV in three-dimensional space;

[0008] Step 2: Real-time acquisition of ground images using an airborne downward-view camera; estimation of the UAV's continuous relative pose using visual odometry; and matching with remote sensing images to obtain the absolute position and correct visual trajectory drift.

[0009] Step 3: In a unified coordinate system, fuse TDOA geometric constraints and visual pose information, perform time alignment on asynchronous data, and output high-precision, low-drift fused pose estimates in a sliding window graph optimization framework.

[0010] Step 4: Establish a segmented landing trajectory model. Utilize real-time acquired UAV attitude and velocity information to adjust the trajectory altitude and altitude change rate online, achieving dynamic trajectory planning under energy constraints. PID control is then employed to achieve stable trajectory tracking, thereby guiding the UAV to complete a stable, safe, and precise landing.

[0011] Furthermore, in step 1, two laser quantum base stations are deployed on the ground to send time-stamped quantum signals to the UAV. After receiving the signals from each base station, the UAV performs cross-correlation processing on the different signals to obtain their Time Difference of Arrival (TDOA). Based on the light speed propagation model, this time difference is mapped into a hyperboloid positioning constraint relationship for the UAV in three-dimensional space, thereby providing stable and reliable global spatial prior information for UAV position calculation. This includes the following two steps:

[0012] Step 1.1, assuming the transmitting base station is... and The base station coordinates are and Each base station transmits entangled laser quantum signals modulated by pseudo-random codes; the arrival time of the quantum signals at each base station is... and The signal arrival time (TDOA) is estimated by intensity cross-correlation calculation, and then the TDOA is calculated. The signal arrival time is extracted using the time-energy entanglement property of photon pairs through a cross-correlation function.

[0013]

[0014] in, To receive signal strength time series, This is a copy of the corresponding pseudo-random code stored locally on the drone. Due to the orthogonality of the pseudo-random codes, and The cross-correlation between the signals is close to zero, achieving signal separation. This is based on the arrival time of the signals from the two base stations. and Calculate the TDOA of the signals from the two base stations:

[0015]

[0016] Step 1.2, convert the time difference into a distance difference:

[0017]

[0018] in, It is the speed of light. (This is combined with TDOA and the locations of the two base stations.) and the location of the drone to be estimated A hyperboloid geometric observation model with the UAV's position as the variable is constructed to form geometric constraints on the UAV's position:

[0019]

[0020] This refers to the three-dimensional geometric priors provided by TDOA.

[0021] Furthermore, in step 2, ground images are acquired in real time using an airborne downward-view camera, and the continuous relative pose of the UAV is estimated in real time using visual odometry; this is then matched with remote sensing imagery to obtain the absolute position and correct visual trajectory drift. This includes the following two steps:

[0022] Step 2.1: Using continuous ground images acquired by the airborne downward-view camera as input, the ORB-SLAM framework based on features is employed to perform high-frequency estimation of the UAV's relative motion. Assume the airborne camera operates at a fixed sampling frequency during landing. Continuous acquisition of image sequences Extract ORB feature point set in each frame of image :

[0023]

[0024] in, For pixel coordinates, For ORB operators.

[0025] Based on this, feature matching is performed on adjacent frames, and the relative pose transformation between adjacent frames is solved by minimizing the multi-frame reprojection error:

[0026]

[0027] in, For relative rotation matrices, Relative translation vector. By recursively accumulating the relative pose at each moment, the continuous visual odometry trajectory of the UAV in the initial coordinate system is obtained. :

[0028]

[0029] This results in the output of a high update rate and a continuous, smooth relative motion trajectory of the drone.

[0030] Step 2.2: High-resolution remote sensing imagery of the landing area is introduced as a global positioning reference. During the UAV's descent, cross-view feature matching is performed between the current airborne imagery and the remote sensing imagery to construct the spatial transformation relationship between the two.

[0031]

[0032] in, These are the pixel coordinates of a feature point in the current airborne image; These are the pixel coordinates in the remote sensing image corresponding to this feature; This is the homography transformation matrix between airborne images and remotely sensed images.

[0033] By combining the current flight altitude information of the UAV, the airborne image features are mapped to the world coordinate system, thereby calculating the absolute two-dimensional position of the UAV in the world coordinate system. :

[0034]

[0035] Furthermore, in step 3, the laser quantum TDOA hyperboloid geometric constraints obtained in step 1 and the visual pose information obtained in step 2 are fused under a unified coordinate system. Asynchronous data are time-aligned, and a high-precision, low-drift fused pose estimate is output within a sliding window graph optimization framework. Assume the UAV is in... The pose to be estimated at any given time is ,in, For the drone's attitude, For the drone's location, firstly, the TDOA data is time-aligned; then, within a sliding window graph optimization framework, the visual motion residual and the TDOA geometric constraint residual are jointly minimized to construct a sliding window joint optimization objective function:

[0036]

[0037] Among them, visual residual Described as the relative motion error between adjacent poses, TDOA residual The deviation described is the amount of deviation of the UAV's current position from the hyperboloid geometric constraints, and its expressions are as follows:

[0038]

[0039] in, The visual relative pose measurement obtained in step 2, The coordinates of the laser base station At the speed of light, This represents the arrival time difference between the two base stations.

[0040] By using nonlinear least squares to obtain the fused pose estimation results of the UAV at each moment, high-precision, low-drift UAV positioning and navigation can be achieved, providing stable and reliable pose input for subsequent landing guidance.

[0041]

[0042] Furthermore, in step 4, a segmented landing trajectory model is established. Using real-time acquired UAV pose and velocity information, the trajectory altitude and rate of change are adjusted online to achieve dynamic trajectory planning under energy constraints. PID control is then employed to achieve stable trajectory tracking, thereby guiding the UAV to a stable, safe, and precise landing. This includes the following two steps:

[0043] Step 4.1: Design the trajectory for the four stages: level flight upon entry, guided descent, index pull-up, and shallow descent.

[0044] Approach Level Flight Phase: The drone smoothly approaches the landing point at the predetermined altitude, preparing for the descent phase. At that time, the drone tracked the approach altitude. The trajectory line is:

[0045]

[0046] when During the initial approach, the drone transitions from level flight to a straight descent, and its trajectory is as follows:

[0047]

[0048] Guided descent phase: The drone gradually descends in altitude with controlled speed to ensure attitude stability upon entering the exponential pull-up phase. When the drone's distance meets the requirements... At that time, the trajectory line for this stage is:

[0049]

[0050] Exponential pull-up phase: The drone's altitude change exhibits an exponential decay characteristic, achieving a transition from a relatively high vertical velocity to the desired descent rate, reducing impact; the exponential pull-up phase begins when the drone's actual relative altitude is ≤10m, and ends when the relative altitude is <3m or... At that time, the upward phase of the index ended. At that time, the trajectory during this stage is as follows:

[0051]

[0052] Shallow glide phase: The drone enters the approaching ground phase, descending further and more gently, preparing for final touchdown; when the sinking rate is ≥ Or, when the height is ≤3m, the trajectory for this stage is:

[0053]

[0054] Step 4.2: During the landing process, the system acquires the fused pose of the UAV in real time. Using speed information and the real-time position of the exponential pull-up phase as a reference point, the trajectory altitude and altitude change rate are adjusted online to achieve dynamic trajectory planning under energy constraints. Combined with traditional PID control, the trajectory line is tracked, allowing the UAV to have a high vertical velocity at the initial landing stage, gradually descend smoothly, and transition to the desired descent rate, ensuring the UAV smoothly enters the shallow glide phase and lands safely.

[0055] Beneficial Effects: This invention provides a method for guiding UAV landing using dual-base station laser correlation information and visual fusion. By introducing the fusion of dual-base station laser quantum TDOA geometric constraints and airborne visual perception information, it can achieve high-precision, low-drift estimation of UAV attitude in complex environments, effectively overcoming the problems of easy drift accumulation in single visual positioning and insufficient stability of single external positioning systems. By establishing segmented landing trajectories and combining real-time fused attitude results for online adjustment and PID control tracking, the UAV achieves continuous altitude change rate and controllable descent speed throughout the landing process, thereby significantly improving the safety, stability, and landing accuracy of UAV automatic landing, and possessing good engineering application value. Attached Figure Description

[0056] Figure 1 A schematic diagram of the overall framework of the UAV autonomous landing guidance system based on laser quantum correlation and visual perception provided by the present invention;

[0057] Figure 2 This is a schematic diagram illustrating the principle and timing relationship of dual-base station laser quantum TDOA measurement provided by the present invention. Detailed Implementation

[0058] The invention will now be further described with reference to the accompanying drawings.

[0059] The technical solution provided by this invention is as follows: First, based on the time difference of arrival extracted from the laser quantum correlation signals of dual base stations, a three-dimensional geometric constraint for the UAV's position is constructed; second, visual odometry positioning is performed using an airborne camera, and the absolute position is obtained by combining remote sensing image matching to correct drift; then, the laser constraint and visual pose information are fused and time-aligned to obtain a high-precision fused pose within a sliding window optimization framework; finally, a segmented landing trajectory model is constructed based on the fused pose and velocity information, the trajectory is adjusted online under energy constraints, and stable, safe, and accurate landing is achieved through PID control.

[0060] Furthermore, the method for guiding UAV landing by dual-base station laser association information and visual fusion provided in this application includes:

[0061] Step 1: Based on the laser quantum signals from the dual base stations, extract the time difference of arrival (TDOA) of the signals through cross-correlation operation, and convert the TDOA into a hyperboloid geometric constraint on the position of the UAV in three-dimensional space.

[0062] Step 2: Use the UAV's onboard downward-view camera to acquire ground images in real time, obtain the UAV's continuous relative pose information through visual odometry, and match the current onboard images with remote sensing images to obtain the UAV's absolute position, which is used to correct the cumulative drift of the visual trajectory.

[0063] Step 3: In a unified coordinate system, the geometric constraints of TDOA and the visual pose information are fused, the asynchronously acquired data is time-aligned, and the high-precision, low-drift UAV fused pose estimation results are obtained by solving in the sliding window graph optimization framework.

[0064] Step 4: Construct a segmented landing trajectory model. Based on the real-time acquired fusion pose and velocity information of the UAV, adjust the trajectory altitude and altitude change rate online to achieve dynamic trajectory planning under energy constraints. Use PID control to achieve stable trajectory tracking, thereby guiding the UAV to complete a stable, safe and accurate landing.

[0065] The embodiments of the present invention are described in detail below:

[0066] Step 1, as follows Figure 1 As shown, at least two laser quantum base stations are first deployed on the ground to send time-stamped quantum signals to the UAV. After receiving the signals from each base station, the UAV extracts the time-to-offset alignment (TDOA) between different signals through cross-correlation calculations. Based on the light speed propagation model, the TDOA is then converted into hyperboloid geometric constraints for the UAV in three-dimensional space, thereby providing stable and reliable global spatial prior information for the UAV's position. Specifically, this involves the following two steps:

[0067] Step 1.1: Deploy two laser quantum emission base stations on the ground in the UAV landing area. and Assume the coordinates of the base station are known, and denoted as . and Each base station transmits entangled laser quantum signals modulated by pseudo-random codes. For example... Figure 2 As shown, the drone is equipped with a laser quantum signal detector to receive and record the arrival time of quantum signals from various base stations. and The signal arrival time (TDOA) is estimated through intensity cross-correlation operations, and then calculated. Utilizing the time-energy entanglement properties of photon pairs, the signal arrival time is extracted through a cross-correlation function.

[0068]

[0069] in, To receive signal strength time series, This is a copy of the corresponding pseudo-random code stored locally on the drone. Due to the orthogonality of the pseudo-random codes, and The cross-correlation between the signals is close to zero, achieving signal separation. This is based on the arrival time of the signals from the two base stations. and Calculate the TDOA of the signals from the two base stations:

[0070]

[0071] Step 1.2, combining TDOA and the locations of the two base stations and the location of the drone to be estimated A hyperboloid geometric observation model with the UAV's position as the variable is constructed to form a geometric constraint on the UAV's position.

[0072] Convert time difference to distance difference:

[0073]

[0074] in, It is the speed of light. This distance difference can be used to constrain the drone's current position. Satisfies the hyperboloid equation:

[0075]

[0076] This refers to the three-dimensional geometric priors provided by TDOA.

[0077] Step 2, as follows Figure 1 As shown, the UAV uses a downward-view camera to continuously collect ground image sequences during descent; through feature extraction, matching and optimization (such as ORB-SLAM), the relative pose change of the UAV relative to the starting point is calculated in real time, the visual odometry (VO) trajectory is output, and the UAV position is calculated by combining the absolute positioning information obtained by matching UAV images and remote sensing images.

[0078] Step 2.1: Using continuous ground images acquired by the airborne downward-view camera as input, the ORB-SLAM framework based on features is employed to perform high-frequency estimation of the UAV's relative motion. Assume the airborne camera operates at a fixed sampling frequency during landing. Continuous acquisition of image sequences Extract ORB feature point set in each frame of image :

[0079]

[0080] in, For pixel coordinates, For ORB operators.

[0081] Based on this, feature matching is performed on adjacent frames, and the relative pose transformation between adjacent frames is solved by minimizing the multi-frame reprojection error:

[0082]

[0083] Its optimization objective function is:

[0084]

[0085] in, For relative rotation matrices, Relative translation vector, For 3D map points, This is a camera projection model. By recursively accumulating the relative pose at each time step, the continuous visual odometry trajectory of the UAV in the initial coordinate system is obtained. :

[0086]

[0087] This results in the output of a high update rate and a continuous, smooth relative motion trajectory of the drone.

[0088] Step 2.2: To suppress the cumulative drift generated by the visual odometry during continuous operation, this invention introduces high-resolution remote sensing imagery of the landing area as a global positioning reference. A mapping relationship between the pixel coordinates of the remote sensing imagery and the world coordinate system is pre-established. During the UAV's descent, cross-view feature matching is performed between the current airborne imagery and the remote sensing imagery to construct the spatial transformation relationship between the two.

[0089]

[0090] in, These are the pixel coordinates of a feature point in the current airborne image; These are the pixel coordinates in the remote sensing image corresponding to this feature; This is the homography transformation matrix between airborne images and remotely sensed images.

[0091] Key feature points extracted from images captured by the UAV's downward-view camera during flight are used for visual odometry and localization matching with remote sensing imagery. Combined with the UAV's current flight altitude information, the airborne image features are mapped to a world coordinate system, thereby calculating the UAV's absolute two-dimensional position in that world coordinate system. :

[0092]

[0093] Step 3: Under a unified spatiotemporal coordinate system, the laser quantum TDOA hyperboloid geometric constraints obtained in Step 1 are deeply fused with the visual odometry relative pose and absolute positioning observations obtained in Step 2. By constructing a laser-visual coupled observation model, the asynchronously acquired multi-source data are time-aligned, and the visual motion residual and TDOA geometric constraint residual are jointly minimized in the sliding window graph optimization framework, thereby outputting a high-precision, low-drift UAV fusion pose estimation result.

[0094] Assuming the drone is in The pose to be estimated at any given time is ,in, For the drone's attitude, The location of the UAV is shown. Due to the time asynchrony between TDOA measurement and visual image sampling, the TDOA data needs to be time-aligned based on the visual time axis. Then, within the sliding window graph optimization framework, the visual motion residual and the TDOA geometric constraint residual are jointly minimized to construct a sliding window joint optimization objective function:

[0095]

[0096] Among them, visual residual Described as the relative motion error between adjacent poses, TDOA residual The deviation described is the amount of deviation of the UAV's current position from the hyperboloid geometric constraints, and its expressions are as follows:

[0097]

[0098] in, The visual relative pose measurement obtained in step 2, For the coordinates of the laser base station, At the speed of light, This represents the arrival time difference between the two base stations.

[0099] By using nonlinear least squares to obtain the fused pose estimation results of the UAV at each moment, high-precision, low-drift UAV positioning and navigation can be achieved, providing stable and reliable pose input for subsequent landing guidance.

[0100]

[0101] Step 4: Establish a segmented landing trajectory model. Utilize real-time acquired UAV attitude and velocity information to adjust the trajectory altitude and rate of change of altitude online, achieving dynamic trajectory planning under energy constraints. Employ PID control to achieve stable trajectory tracking, resulting in a stable, safe, and precise landing. This specifically includes the following two sub-steps:

[0102] Step 4.1: To ensure the smoothness of the UAV landing process and avoid abrupt changes in altitude change rate, the UAV's automatic landing trajectory is divided into five segments: approach level flight segment, guided glide segment, exponential pull-up segment, shallow glide segment, and ground taxiing segment. In this embodiment, the trajectory design is performed for the first four stages:

[0103] The approach level flight phase: used to stabilize the drone's landing altitude and speed, align it with the airport runway centerline, change its aerial configuration, and prepare to capture the descent trajectory. At that time, the drone tracked the approach altitude. The trajectory line is:

[0104]

[0105] when During the initial approach, the drone transitions from level flight to a straight descent, and its trajectory is as follows:

[0106]

[0107] Straight-line descent phase: In this phase, a landing trajectory line is pre-defined along the runway centerline to ensure the drone reaches near-ground smoothly and quickly, preparing for attitude pull-up. When the drone's distance meets the requirements... At that time, the trajectory line for this stage is:

[0108]

[0109] Exponential pull-up phase: In this phase, the drone pulls up smoothly, seamlessly connecting the straight descent phase and the final shallow descent phase, ensuring that the trajectory transition between different phases does not produce large jumps. This is also a crucial part of online trajectory adjustment. The exponential pull-up phase begins when the drone's actual relative altitude is ≤10m, and ends when the relative altitude is <3m or... At that time, the upward phase of the index ended. At that time, the trajectory during this stage is as follows:

[0110]

[0111] Shallow glide phase: During this phase, the drone's descent trajectory is disrupted, maintaining the desired descent rate until a safe and stable touchdown is achieved. Laterally, the drone's various states must meet the touchdown requirements. When the descent rate ≥ Or, when the height is ≤3m, the trajectory for this stage is:

[0112]

[0113] Step 4.2: During the landing process, the system acquires the fused pose of the UAV in real time. Using speed information and the real-time position of the exponential pull-up phase as a reference point, the trajectory altitude and altitude change rate are adjusted online to achieve dynamic trajectory planning under energy constraints. Combined with traditional PID control, the trajectory line is tracked, allowing the UAV to have a high vertical velocity at the initial landing stage, gradually descend smoothly, and transition to the desired descent rate, ensuring the UAV smoothly enters the shallow glide phase and lands safely.

Claims

1. A method for guiding UAV landing by fusion of dual-base station laser correlation information and visual information, characterized in that, The drone landing guidance method includes the following steps: Step 1: Obtain the laser quantum signals from the two base stations, extract the time difference of arrival (TDOA) of the signals received by the UAV from the two base stations through cross-correlation calculation, and convert it into a hyperboloid geometric prior that constrains the position of the UAV in three-dimensional space. Step 2: Real-time acquisition of ground images using an airborne downward-view camera; real-time estimation of the UAV's continuous relative pose using visual odometry; matching with remote sensing images to obtain the UAV's absolute position and correct visual trajectory drift. Step 3: In a unified coordinate system, fuse the geometric constraints of Time Difference of Arrival (TDOA) and visual pose information, perform time alignment on asynchronous data, and output a high-precision, low-drift fused pose estimate in the sliding window graph optimization framework. Step 4: Establish a segmented landing trajectory model. Utilize real-time acquired UAV pose estimation and velocity information to adjust the trajectory altitude and altitude change rate online, achieving dynamic trajectory planning under energy constraints. PID control is then employed to achieve stable trajectory tracking, thereby guiding the UAV to complete the landing.

2. The method for guiding UAV landing by dual-base station laser correlation information and visual fusion according to claim 1, characterized in that, In step 1, two laser quantum emission base stations are deployed on the ground in the UAV landing area. and Assume the coordinates of the base station are known, and denote them as... and After receiving signals from each base station, the UAV extracts the time difference of arrival (TDOA) between different signals through cross-correlation calculation. Based on the light speed propagation model, the TDOA is converted into hyperboloid geometric constraints of the UAV in three-dimensional space, thereby providing stable and reliable global spatial prior information for the UAV's position. Assume the arrival time of the signals from the two base stations and The time difference of arrival (TDOA) between the two base station signals is expressed as: ； Convert time difference to distance difference: ； in, At the speed of light, this distance difference can be used to constrain the drone's current position. Satisfies the hyperboloid equation: ； Three-dimensional geometric priors are provided to achieve Time Difference of Occurrence (TDOA).

3. The method for guiding UAV landing by dual-base station laser correlation information and visual fusion according to claim 1, characterized in that: In step 2, the UAV uses a downward-view camera to continuously acquire ground image sequences during descent; the relative pose change of the UAV relative to the starting point is calculated in real time, the visual odometry (VO) trajectory is output, and the UAV position is calculated by combining the absolute positioning information obtained by matching the UAV images and remote sensing images. Assume the airborne camera uses a fixed sampling frequency during landing. Continuous acquisition of image sequences Extract ORB feature point set in each frame of image ,in, These are the pixel coordinates of a feature point in the current airborne image. For ORB operators; Feature matching is performed on adjacent frames, and the relative pose transformation between adjacent frames is solved by minimizing the multi-frame reprojection error. ； in, For relative rotation matrices, Relative translation vector, For 3D map points, For camera projection model; By recursively accumulating the relative pose at each moment, the continuous visual odometry trajectory of the UAV in the initial coordinate system is obtained. : ； This results in the output of a high update rate and a continuous, smooth relative motion trajectory of the UAV. Constructing the spatial transformation relationship between airborne images and remote sensing images ,in, These are the pixel coordinates of a feature point in the current airborne image; These are the pixel coordinates in the remote sensing image corresponding to this feature; This represents the homography transformation matrix between the airborne image and the remotely sensed image. Combining the UAV's current flight altitude information, the airborne image features are mapped to the world coordinate system, and the UAV's absolute two-dimensional position in the world coordinate system is calculated. : 。 4. The method for guiding UAV landing by dual-base station laser correlation information and visual fusion according to claim 1, characterized in that: In step 3, the laser quantum TDOA hyperboloid geometric constraints obtained in step 1 are deeply fused with the visual odometry relative pose and absolute positioning observations obtained in step 2. By constructing a laser-visual coupled observation model, the asynchronously acquired multi-source data are time-aligned, and the visual motion residual and TDOA geometric constraint residual are jointly minimized in the sliding window graph optimization framework to output the UAV fused pose estimation result. Assuming the drone is in The pose to be estimated at any given time is ,in, For the drone's attitude, Let the location be the drone's position. The sliding window joint optimization objective function is expressed as: ； Among them, visual residual Described as the relative motion error between adjacent poses, TDOA residual Described as the deviation of the drone's current position from the hyperboloid geometric constraints; The fused pose estimation results of the UAV at each time step are obtained by nonlinear least squares solution: 。 5. The method for guiding UAV landing by dual-base station laser correlation information and visual fusion according to claim 1, characterized in that: In step 4, a segmented landing trajectory model is established, including the approach level flight segment, the guided glide segment, the exponential pull-up segment, and the shallow glide segment trajectory design; by using real-time acquired fused attitude and velocity information of the UAV, the trajectory altitude and altitude change rate are adjusted online to achieve dynamic trajectory planning under energy constraints; PID control is used to achieve stable trajectory tracking and precise landing.

Images