Cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and ROS architecture
By employing a cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture, the positioning accuracy problem of UAVs in environments with weak or denied GPS signals is solved. This method achieves high-precision robust mapping and lightweight computing, supports unified control across different platforms, and enhances the autonomous navigation capability of UAVs in complex environments.
Patent Information
- Application Number
- CN202511318733.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-16
- Publication Date
- 2025-12-23
AI Technical Summary
Existing UAV autonomous navigation systems suffer from a sharp drop in positioning accuracy in environments with weak or denied GPS signals. Traditional SLAM solutions are prone to failure in complex environments. Multi-sensor fusion solutions lack robustness, and incompatibility between different platform protocols leads to poor scalability, making it difficult to achieve high-precision robust mapping, lightweight computing, and unified cross-platform control.
By employing multi-sensor fusion and ROS architecture, and through the fusion of fisheye vision and LiDAR data, a cross-platform protocol adaptation mechanism is designed to achieve unified control. Furthermore, dynamic weight adjustment of visual reprojection error and LiDAR point cloud matching error is performed at the computational layer of the ROS architecture, thereby constructing an environment-adaptive, tightly coupled, autonomous mapping optimization function.
Achieving high-precision positioning and mapping in GPS-denied environments improves the robustness and positioning continuity of UAVs in complex environments, reduces computing power consumption, supports seamless control across different platforms, and enhances the accuracy and stability of autonomous mapping.
Smart Images
Figure CN121186809A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of unmanned aerial vehicle autonomous navigation, in particular to a cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and ROS architecture. BACKGROUND
[0002] At present, the unmanned aerial vehicle autonomous navigation system mainly relies on the combination of GPS and inertial measurement unit (IMU) for navigation in a good GPS signal environment. However, in complex scenarios such as indoor, tunnel, urban canyon or disaster area where GPS signal is weak or denied, the positioning accuracy of this scheme decreases sharply, leading to mapping failure and navigation out of control. For such environments, the existing technical solutions and their limitations mainly include: Pure vision SLAM schemes such as ORB-SLAM and VINS-Fusion rely on camera data and are prone to tracking failure in low light, texture missing or strong dynamic interference conditions; Laser SLAM schemes such as LOAM and FAST-LIO are based on laser radar point cloud and have significantly increased error in sparse point cloud, rain, fog and dust environment or feature repetitive scene; Multi-sensor fusion schemes such as LIO-SAM mostly use a loosely coupled architecture, and vision and laser data are only fused at the result level of independent processing, failing to fully exploit the synergistic complementary advantages of multi-modal data. When a single sensor fails due to environmental factors, it is easy to cause a cliff-like decline in performance, lacking an environment-adaptive robustness guarantee mechanism.
[0003] Traditional high-precision SLAM algorithms have huge computational overhead and usually rely on high-performance industrial computers for operation, resulting in high power consumption and large size, making it difficult to achieve efficient deployment and real-time operation on lightweight embedded unmanned aerial vehicle platforms, severely limiting the application range.
[0004] In addition, mainstream unmanned aerial vehicle platforms such as open-source flight control PX4 / APM and commercial DJI unmanned aerial vehicles use heterogeneous communication protocols, and existing solutions lack a unified control framework and intelligent protocol adaptation mechanism, requiring developers to customize development for different platforms, resulting in poor scalability, high maintenance cost, and difficulty in achieving "plug and play" and unified scheduling of heterogeneous unmanned aerial vehicle platforms.
[0005] In summary, the existing technology cannot simultaneously meet the core needs of high-precision robust mapping, lightweight low-power computing architecture deployment and cross-platform unmanned aerial vehicle unified seamless control for unmanned aerial vehicle autonomous operation in complex denied environments. SUMMARY
[0006] In view of the problems in the prior art, the present application provides a cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and a ROS architecture, which realizes high-precision positioning and environment mapping in a GPS denial environment by designing a protocol adaptation mechanism to be compatible with a unified autonomous control of heterogeneous unmanned aerial vehicle platforms and by fusing fisheye vision and laser radar perception data.
[0007] To achieve the above technical purposes, the present application adopts the following technical solutions: a cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and a ROS architecture, comprising the following steps: Step S1: respectively deploying a fisheye camera and a laser radar on different types of unmanned aerial vehicles, the fisheye camera being used to collect visual image data, and the laser radar being used to collect three-dimensional point cloud data; Step S2: a control layer of the ROS architecture dynamically loads a corresponding protocol stack according to the type of the unmanned aerial vehicle, thereby realizing unified control of the cross-platform unmanned aerial vehicle; Step S3: a sensor layer of the ROS architecture is used to collect the visual image data and the three-dimensional point cloud data collected by the unmanned aerial vehicle in real time and transmit them to a calculation layer of the ROS architecture; Step S4: the calculation layer of the ROS architecture generates visual re-projection error weights and laser point cloud matching error weights according to the visual image data and the three-dimensional point cloud data, and constructs an environment-adaptive tightly coupled autonomous mapping optimization function; Step S5: taking minimization of the tightly coupled autonomous mapping optimization function as an objective, the optimal visual-radar joint pose of the unmanned aerial vehicle is determined through nonlinear optimization, and cross-platform unmanned aerial vehicle autonomous mapping is performed.
[0008] Further, the specific process of step S2 is as follows: the middleware of the control layer of the ROS architecture continuously monitors the physical connection of the specified physical port, when a unmanned aerial vehicle is connected, the middleware starts MAVLink heartbeat monitoring or tries PSDK device enumeration to identify the type of the connected unmanned aerial vehicle, and configures the unmanned aerial vehicle, thereby realizing unified control of the cross-platform unmanned aerial vehicle.
[0009] Further, for an open-source unmanned aerial vehicle, the middleware identifies the type of the unmanned aerial vehicle by listening to and analyzing the MAVLink heartbeat message packet sent by the unmanned aerial vehicle connected to the MAVLink compatible port, and according to the autopilot field and the type field in the MAVLink heartbeat message packet.
[0010] Further, for the identified open-source drone, the middleware sends ROS standard control instructions using MAVROS, including: sending target height and hovering time for takeoff or landing through mavros_msgs / CommandTOL message, specifying target position and attitude for pose control using geometry_msgs / PoseStamped, and sending linear and angular velocity instructions for velocity control through geometry_msgs / TwistStamped.
[0011] Further, for the commercial drone, the middleware enumerates available drones through the device management interface DJI::PSDK::get_device_list( ) provided by the PSDK library, and extracts the unique identifier SN, type, and firmware version information of the drone from the enumeration result.
[0012] Further, for the identified commercial drone, the middleware parses the ROS standard control instructions, extracts the target position and attitude, and encapsulates them into PSDK protocol format.
[0013] Further, step S4 includes the following sub-steps: Step S4.1: Convert the visual image data into an image grayscale matrix, calculate the global average grayscale of the visual image, and determine the illumination intensity of the visual image in combination with the underexposure threshold and overexposure threshold; Step S4.2: Calculate the effective point cloud density according to the three-dimensional point cloud data, and determine the point cloud density evaluation value in combination with the point cloud sparsity threshold and the point cloud ideal density threshold; Step S4.3: Generate the visual re-projection error weight and the laser point cloud matching error weight according to the illumination intensity of the visual image and the point cloud density evaluation value of the corresponding three-dimensional point cloud data; Step S4.4: Estimate the visual-laser joint pose according to the visual image data and the corresponding three-dimensional point cloud data, determine the visual re-projection error and the laser point cloud matching error, and construct an environment-adaptive tight-coupling autonomous mapping optimization function combining the visual re-projection error weight and the laser point cloud matching error weight:
[0014] wherein, represents the visual-laser joint pose represented by Lie algebra, represents the mathematical expression of K represents the ORB feature point in the visual image data, k represents K the index of represents the re-projection error of the k th ORB feature point under the visual-laser joint pose, represents a visual re-projection error weight; S represents a quantity of three-dimensional point cloud data after noise reduction, s represents S an index of represents a matching error of the s th three-dimensional point cloud data under a visual-laser combined pose, represents a laser point cloud matching error weight.
[0015] Further, the calculation process of the light intensity of the visual image is: , wherein, represents the light intensity of the visual image, represents the global average gray scale of the visual image, represents an underexposure threshold value, represents an underexposure region scaling coefficient, ; represents an overexposure threshold value, represents an overexposure region attenuation coefficient, , represents a maximum gray scale value.
[0016] Further, the calculation process of the point cloud density evaluation value is: , wherein, represents the point cloud density evaluation value, represents an effective point cloud density, , represents the number of point clouds in the effective region of the three-dimensional point cloud data, represents the convex hull area of the three-dimensional point cloud data projected onto the horizontal plane; represents a point cloud sparsity threshold value, represents a point cloud ideal density threshold value, represents a density gain coefficient, .
[0017] Further, the generation process of the visual re-projection error weight is as follows:
[0018] The generation process of the laser point cloud matching error weight is as follows:
[0019] wherein, represents the light intensity of the visual image, represents the point cloud density evaluation value, denotes a normalization factor, .
[0020] Compared with the prior art, the present application has the following beneficial effects: the cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and ROS architecture solves the problem of incompatible protocols of different types of unmanned aerial vehicle manufacturers, sets cross-platform middleware in the control layer of the ROS architecture, supports the conversion between ROS standard messages and different unmanned aerial vehicle special protocols, and thus realizes unified control of different types of unmanned aerial vehicles; secondly, the visual image data and three-dimensional point cloud data are used to generate visual re-projection error weights and laser point cloud matching error weights to dynamically affect the proportion in the fusion process of visual SLAM and laser SLAM, and thus affect the robustness of unmanned aerial vehicle autonomous mapping; when the environment changes, the corresponding weight dynamically decreases to reduce the proportion of error items of the corresponding sensor, while improving the dominant role of reliable sensors, which can inhibit environmental interference from the root and significantly enhance the robustness and positioning continuity of unmanned aerial vehicle autonomous mapping in scenes such as light mutation and single structure. BRIEF DESCRIPTION OF DRAWINGS
[0021] Figure 1 The flowchart of the cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and ROS architecture of the present application. DETAILED DESCRIPTION
[0022] The technical solutions of the present application will be further explained and described below in combination with the drawings.
[0023] As Figure 1 The flowchart of the cross-platform unmanned aerial vehicle autonomous mapping method based on multi-sensor fusion and ROS architecture of the present application, the cross-platform unmanned aerial vehicle autonomous mapping method comprises the following steps: step S1: respectively deploying an fisheye camera, a laser radar and a physical communication interface on different types of unmanned aerial vehicles, wherein the field of view angle of the fisheye camera is ≥180°, the resolution is 1920x1080, and the frame rate is 30Hz, which is used to collect visual image data under a wide field of view angle, and the lens needs to be ensured to be unobstructed during installation, and the hardware synchronization line is aligned with the time stamp of the laser radar; the laser radar adopts MID-360, the working frequency is 10Hz, the maximum ranging is 100m, and the angle resolution is 0.18°, which is used to collect three-dimensional point cloud data, and the installation position needs to avoid overlapping with the field of view of the fisheye lens to reduce interference; the physical communication interface includes USB, serial port or network port, which is used to connect the middleware of the ROS architecture.
[0024] Step S2: the control layer of the ROS architecture dynamically loads the corresponding protocol stack according to the type of the unmanned aerial vehicle, and realizes unified control of the cross-platform unmanned aerial vehicle; specifically: in view of the problem that the protocols of different types of unmanned aerial vehicles such as PX4, ArduPilot, APM, DJI PSDK and the like are incompatible, the control layer of the ROS architecture sets a cross-platform middleware, the middleware continuously monitors the physical connection of a specified physical port such as a USB, a serial port and a network port, when an unmanned aerial vehicle is connected, the middleware starts MAVLink heartbeat monitoring or tries PSDK device enumeration to identify the type of the connected unmanned aerial vehicle, the middleware automatically loads and activates the corresponding protocol conversion inside, supports conversion between ROS standard messages and different unmanned aerial vehicle special protocols, for example, when PX4, ArduPilot or APM is identified, the MAVROS adaptation layer is loaded, when DJI is identified, the PSDK conversion layer is loaded, to configure the unmanned aerial vehicle, and realize unified control of the cross-platform unmanned aerial vehicle.
[0025] For open source unmanned aerial vehicles, the middleware listens to and analyzes the MAVLink heartbeat message packet sent by the unmanned aerial vehicle connected to the MAVLink compatible port, identifies the type of the unmanned aerial vehicle according to the autopilot field and the type field in the MAVLink heartbeat message packet, including identification of PX4, ArduPilot and APM unmanned aerial vehicles. If the middleware continuously receives the MAVLink heartbeat message packet of the unmanned aerial vehicle, it indicates that the unmanned aerial vehicle is online; otherwise, it indicates that the unmanned aerial vehicle is in an abnormal or offline state. The middleware sends ROS standard control instructions using MAVROS, including: sending target height and hovering time for takeoff or landing through mavros_msgs / CommandTOL message, specifying target position and attitude for pose control using geometry_msgs / PoseStamped, and sending linear velocity and angular velocity instructions for speed control using geometry_msgs / TwistStamped.
[0026] For commercial unmanned aerial vehicles such as DJI PSDK, the middleware enumerates available unmanned aerial vehicles through the device management interface DJI::PSDK::get_device_list( ) provided by the PSDK library, and extracts the unique identifier SN, type and firmware version information of the unmanned aerial vehicle from the enumeration result to explicitly identify the specific model of the connected DJI unmanned aerial vehicle. The middleware analyzes the ROS standard control instructions, extracts the target position and attitude, encapsulates them into the PSDK protocol format, adds the PSDK check code CRC16 and the timeout retransmission mechanism for safety check, and sends the instructions through the asynchronous interface of DJI MSDK / OSDK / PSDK to avoid blocking the ROS main thread.
[0027] Step S3: The sensor layer of the ROS architecture is used to collect visual image data and three-dimensional point cloud data collected by the unmanned aerial vehicle in real time, which is transmitted to the computing layer of the ROS architecture.
[0028] Step S4: The computing layer of the ROS architecture generates visual re-projection error weights and laser point cloud matching error weights according to the visual image data and the three-dimensional point cloud data, and constructs an environment-adaptive tightly-coupled autonomous mapping optimization function. On the one hand, the visual image data and the three-dimensional point cloud data are tightly coupled and optimized in the pose estimation stage. On the other hand, a real-time dynamic weight adjustment mechanism is introduced, which can adaptively adjust the weight proportion of the visual re-projection error and the laser point cloud matching error in the overall optimization target according to the environment perception information of the light intensity and the point cloud density, maximize the complementary advantages of the two sensors in complex environments, and thus improve the mapping accuracy. The following sub-steps are included: Step S4.1: Convert the visual image data into an image grayscale matrix G , the gray value of each pixel point G , i , j in the image grayscale matrix , calculate the global average gray of the visual image, and combine the underexposure threshold and the overexposure threshold to determine the light intensity of the visual image: , wherein, represents the light intensity of the visual image, represents the global average gray of the visual image, , m represents the number of rows of the image grayscale matrix G , n represents the number of columns of the image grayscale matrix G ; represents the underexposure threshold, represents the underexposure area scaling coefficient, ; represents the overexposure threshold, represents the overexposure area decay coefficient, , represents the maximum gray value, . In one technical solution of the present application , .
[0029] Step S4.2: Calculate the effective point cloud density according to the three-dimensional point cloud data, and combine the point cloud sparsity threshold and the point cloud ideal density threshold to determine the point cloud density evaluation value: , wherein, represents the point cloud density evaluation value, represents the effective point cloud density, , represents the number of point clouds in the effective area of three-dimensional point cloud data, represents the convex hull area of three-dimensional point cloud data projected onto the horizontal plane; represents the point cloud sparsity threshold, represents the point cloud ideal density threshold, represents the density gain coefficient, In one technical solution of the present application, , .
[0030] Step S4.3: generating visual re-projection error weight according to the light intensity of the visual image and the point cloud density evaluation value of the corresponding three-dimensional point cloud data and laser point cloud matching error weight :
[0031]
[0032] wherein, represents the light intensity of the visual image, represents the point cloud density evaluation value, represents the normalization factor, .
[0033] Step S4.4: estimating visual-laser joint pose according to visual image data and corresponding three-dimensional point cloud data, determining visual re-projection error and laser point cloud matching error, combining visual re-projection error weight and laser point cloud matching error weight to construct environment-adaptive tight-coupling autonomous mapping optimization function; specifically, including the following sub-steps: Step S4.4.1: performing de-distortion processing on the visual image data, extracting ORB feature points and matching with adjacent frames to establish the correspondence of ORB feature points; Step S4.4.2: performing noise reduction processing on the corresponding three-dimensional point cloud data, and matching the noise reduction processed three-dimensional point cloud data with the ORB feature points, and using PnP algorithm to estimate the visual-laser joint pose; Step S4.4.3: projecting the ORB feature points in the camera coordinate system to the world system, projecting to the image plane based on the current visual-laser joint pose to generate projection coordinates , calculating the Euclidean distance between the actual observation coordinates of the ORB feature points and the projection coordinates as the visual re-projection error of the ORB feature points; Step S4.4.4: Search for nearest neighbor point sets in the sliding window local map using the denoised 3D point cloud data and fit a plane. Transform the denoised point cloud to the world coordinate system using the current vision-laser joint pose. Calculate the vertical distance from each 3D point cloud data point to the corresponding fitted plane, which is used as the laser point cloud matching error. ,in, Indicates the first s The normal vector of the fitted plane corresponding to each 3D point cloud data. T Indicates transpose. This represents the variance of radar ranging noise. This represents the rotation matrix from the radar frame to the world coordinate system. Indicates the first s 3D coordinates of a 3D point cloud data point in a radar system Indicates the first s The intercept of the fitting plane corresponding to each 3D point cloud data; Step S4.4.5: Construct an environment-adaptive, tightly coupled autonomous mapping optimization function by combining visual reprojection error weights and laser point cloud matching error weights:
[0034] in, Represents the vision-laser joint pose as expressed by Lie algebra. express The mathematical representation of K Represents ORB feature points in visual image data. k express K index, Indicates the first under vision-laser joint pose k Reprojection error of each ORB feature point Indicates the weight of visual reprojection error; S This represents the amount of 3D point cloud data after noise reduction. s express S index, Indicates the first under vision-laser joint pose s Matching error of 3D point cloud data This represents the weight of the laser point cloud matching error.
[0035] Using visual reprojection error weights Matching error weights with laser point clouds The dynamic influence of the light intensity and the point cloud density on the environment perception information is used to adaptively adjust the weight proportion of the visual error term and the laser error term in the overall optimization target, the complementary advantages of the two sensors in the complex environment are maximized, and the mapping precision is improved.
[0036] Step S5: The tight-coupling autonomous mapping optimization function minimization is taken as an objective, nonlinear optimization is performed, the optimal camera-radar joint position of the unmanned aerial vehicle is determined, and cross-platform autonomous mapping of the unmanned aerial vehicle is realized; specifically, the LM algorithm is used to iteratively optimize the camera-radar joint position, the Jacobian matrix chain derivation of the visual and laser errors is used to update the camera-radar joint position increment, and the optimal camera-radar joint position of the unmanned aerial vehicle is used for laser dense map construction.
[0037] In one technical solution of the present application, the RK3588 chip with 6 TOPS AI computing power is used, 16GB RAM and 128GB eMMC storage are configured, and the cross-platform autonomous mapping method of the unmanned aerial vehicle is run, so that the power consumption can be reduced by 60% compared with the traditional industrial computer, and the on-board computing demand is met.
[0038] The cross-platform autonomous mapping method of the unmanned aerial vehicle based on the multi-sensor fusion and the ROS architecture sets the cross-platform middleware in the control layer of the ROS architecture, supports the conversion between the ROS standard message and the different unmanned aerial vehicle special protocol, thereby realizing the unified control on different types of unmanned aerial vehicles; meanwhile, the visual image data and the three-dimensional point cloud data are tightly coupled and optimized in the position estimation stage, and the real-time dynamic weight adjustment mechanism is introduced, the weight proportion of the visual error term and the laser error term in the overall optimization target can be adaptively adjusted according to the light intensity and the point cloud density, the complementary advantages of the two sensors in the complex environment are maximized, thereby the mapping precision is improved, and high-precision positioning and environment mapping can be realized even in the GPS denial environment. It has been verified that the absolute trajectory error of the autonomous mapping of the cross-platform autonomous mapping method of the unmanned aerial vehicle is reduced by 42% compared with the loose-coupling autonomous mapping, and the instruction transmission delay of the cross-platform unmanned aerial vehicle is lower than 50ms, which can meet the real-time data transmission demand. The cross-platform autonomous mapping method of the unmanned aerial vehicle can be widely applied to the fields of industrial inspection, disaster emergency rescue, three-dimensional surveying and mapping and the like which need to be operated in complex or unknown environments.
[0039] In one technical solution of the present application, a computer readable storage medium is also provided, which stores a computer program, and the computer program enables a computer to execute the cross-platform autonomous mapping method of the unmanned aerial vehicle based on the multi-sensor fusion and the ROS architecture.
[0040] In one technical solution of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and when the processor executes the computer program, the method for autonomous mapping of a cross-platform unmanned aerial vehicle based on multi-sensor fusion and ROS architecture is implemented.
[0041] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store programs for use by or in connection with an instruction execution system, apparatus or device. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the above. More specific examples of computer storage medium can include one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.
[0042] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiment. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principle of the present application shall be considered as falling within the protection scope of the present application.
Claims
1. A cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture, characterized in that, Includes the following steps: Step S1: Deploy fisheye cameras and lidar on different types of drones respectively. The fisheye cameras are used to collect visual image data, and the lidar is used to collect three-dimensional point cloud data. Step S2: The ROS architecture's control layer dynamically loads the corresponding protocol stack based on the type of drone, enabling unified control of cross-platform drones; Step S3: The sensor layer of the ROS architecture is used to collect visual image data and 3D point cloud data acquired by the UAV in real time and transmit them to the computing layer of the ROS architecture. Step S4: The computational layer of the ROS architecture generates visual reprojection error weights and laser point cloud matching error weights based on visual image data and 3D point cloud data, and constructs an environment-adaptive, tightly coupled autonomous mapping optimization function. Step S5: With the goal of minimizing the tightly coupled autonomous mapping optimization function, the optimal vision-radar joint pose of the UAV is determined through nonlinear optimization, and cross-platform UAV autonomous mapping is performed.
2. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture as described in claim 1, characterized in that, The specific process of step S2 is as follows: The middleware of the control layer of the ROS architecture continuously monitors the physical connection of the specified physical port. When a drone connects, the middleware starts MAVLink heartbeat listening or attempts PSDK device enumeration to identify the type of connected drone and configure the drone to achieve unified control of cross-platform drones.
3. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 2, characterized in that, For open-source drones, the middleware listens to and parses MAVLink heartbeat message packets sent by drones connected to the MAVLink compatible port, and identifies the type of the drone based on the autopilot and type fields in the MAVLink heartbeat message packet.
4. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 3, characterized in that, For the identified open-source drone, the middleware uses MAVROS to send ROS standard control commands, including: sending target altitude and hovering time via mavros_msgs / CommandTOL messages for takeoff or landing; using geometry_msgs / PoseStamped to specify target position and attitude for pose control; and sending linear velocity and angular velocity commands via geometry_msgs / TwistStamped for velocity control.
5. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 2, characterized in that, For commercial drones, the middleware enumerates available drones through the device management interface DJI::PSDK::get_device_list() provided by the PSDK library, and extracts the drone's unique identifier SN, type, and firmware version information from the enumeration results.
6. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 5, characterized in that, For the identified commercial drones, the middleware parses the ROS standard control commands, extracts the target position and attitude, and encapsulates them into the PSDK protocol format.
7. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 1, characterized in that, Step S4 includes the following sub-steps: Step S4.1: Convert the visual image data into an image grayscale matrix, calculate the global average grayscale of the visual image, and combine the underexposure threshold and the overexposure threshold to determine the illumination intensity of the visual image; Step S4.2: Calculate the effective point cloud density based on the 3D point cloud data, and determine the point cloud density evaluation value by combining the point cloud sparsity threshold and the point cloud ideal density threshold. Step S4.3: Generate visual reprojection error weights and laser point cloud matching error weights based on the illumination intensity of the visual image and the point cloud density evaluation value of the corresponding 3D point cloud data; Step S4.4: Estimate the visual-laser joint pose based on the visual image data and the corresponding 3D point cloud data, determine the visual reprojection error and the laser point cloud matching error, and construct an environment-adaptive, tightly coupled, autonomous mapping optimization function by combining the visual reprojection error weights and the laser point cloud matching error weights. in, Represents the vision-laser joint pose as expressed by Lie algebra. express The mathematical representation of K Represents ORB feature points in visual image data. k express K index, Indicates the first under vision-laser joint pose k Reprojection error of each ORB feature point Indicates the weight of visual reprojection error; S This represents the amount of 3D point cloud data after noise reduction. s express S index, Indicates the first under vision-laser joint pose s Matching error of 3D point cloud data This represents the weight of the laser point cloud matching error.
8. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 7, characterized in that, The calculation process for the illumination intensity of the visual image is as follows: ， in, Indicates the light intensity of a visual image. Represents the global average gray level of a visual image. Indicates the underexposure threshold. Indicates the scaling factor for the underexposed area. ; Indicates the overexposure threshold. Indicates the attenuation coefficient in the overexposed area. , This represents the maximum grayscale value.
9. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 7, characterized in that, The calculation process for the point cloud density evaluation value is as follows: ， in, This represents the point cloud density assessment value. Indicates the effective point cloud density. , This indicates the number of points within the valid area of the 3D point cloud data. This represents the area of the convex hull projected onto the horizontal plane from a 3D point cloud data. This represents the point cloud sparsity threshold. This represents the ideal density threshold for point clouds. Represents the density gain coefficient. .
10. The cross-platform UAV autonomous mapping method based on multi-sensor fusion and ROS architecture according to claim 7, characterized in that, The visual reprojection error weight The generation process is as follows: The laser point cloud matching error weight The generation process is as follows: in, Indicates the light intensity of a visual image. This represents the point cloud density assessment value. Represents the normalization factor. .