An unmanned aerial vehicle near-ground complex obstacle pre-perception and emergency hovering method and system

Abstract

The embodiment of the application discloses a kind of unmanned plane near ground complex obstacle pre-perception and emergency hover method and system, through the synchronous acquisition of environment data of multiple sensor array;Data is carried out three-layer fusion perception processing including data layer, feature layer and decision layer, constructs real-time obstacle map, and generates virtual obstacle when point cloud is sparse through semantic completion mechanism;Based on obstacle map and flight state, the flight risk value is dynamically calculated using risk quantification model;When risk value exceeds emergency hover threshold, immediately generate hardware interrupt signal;In response to the interrupt signal, directly take over flight control, execute emergency hover logic including thrust boost and high gain attitude stability, so that unmanned plane instantaneously hovers.The present application solves the problem of inaccurate small obstacle perception and large obstacle avoidance response delay in near ground complex scenarios, realizes high detection rate, millisecond-level response active safety protection, and improves the flight safety and reliability of unmanned plane in narrow and dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) flight safety technology, specifically to a method and system for UAV near-ground complex obstacle pre-sensing and emergency hovering. Background Technology

[0002] With the widespread application of drones in complex near-ground scenarios such as logistics delivery, power line inspection, and emergency rescue, flight safety faces severe challenges. Small (less than 5 cm in diameter), low-reflectivity, and highly dynamic obstacles in these scenarios, such as power lines, tree branches, thin cables, and billboard supports, pose a significant threat to traditional drone sensing systems.

[0003] Single sensors (such as ultrasonic sensors, monocular vision, and 2D LiDAR) have significant limitations in complex near-ground environments. Ultrasonic sensors have short ranging ranges and are susceptible to interference; monocular vision lacks accurate depth information and is greatly affected by lighting conditions; 2D LiDAR can only scan a single plane, making it extremely easy to miss obstacles outside the scanning plane (such as overhead power lines).

[0004] Existing obstacle avoidance algorithms are mostly designed for large, static obstacles, and lack the ability to extract features and classify small, dynamic obstacles. The processing flow is usually "perception-modeling-planning-control", which is long and time-consuming. When faced with suddenly appearing obstacles, drones often collide because they do not have enough time to complete the obstacle avoidance path planning.

[0005] Most systems are pre-programmed to respond with either "detour" or "climb". However, in narrow alleys, there may not be enough space to detour; forced climbing may violate airspace height restrictions or result in collisions with overhead obstacles. There is a lack of a conservative safety strategy (such as emergency hovering) that can be decided and executed in a very short time. Summary of the Invention

[0006] To address this issue, this invention provides a method and system for anti-obstacle pre-sensing and emergency hovering of unmanned aerial vehicles (UAVs) in complex near-ground environments, thereby solving the technical problem in the prior art where UAVs are prone to collisions with small obstacles due to inaccurate perception and response delays in complex near-ground environments.

[0007] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0008] According to a first aspect of the present invention, a method for anti-obstacle prediction and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV) is provided, the method comprising:

[0009] Simultaneous acquisition of environmental data in front of and to the sides of the flight using a multi-sensor array;

[0010] The collected environmental data is processed by a three-layer fusion perception system to identify and construct a real-time obstacle map containing information on the location, type, and size of obstacles.

[0011] Based on the real-time obstacle map and the flight status information of the UAV, the risk value of the current flight is dynamically calculated through a risk quantification model.

[0012] The system detects in real time whether the risk value exceeds a preset emergency hovering threshold. When the risk value exceeds the emergency hovering threshold, it immediately generates a hardware interrupt signal with the highest priority.

[0013] In response to the hardware interrupt signal, the system bypasses the drone's conventional flight control loop and directly executes the emergency hovering control logic to control the drone to achieve stable hovering at its current position.

[0014] Furthermore, the multi-sensor array includes at least one solid-state LiDAR, one high frame rate global shutter camera, and a set of millimeter-wave radars;

[0015] Simultaneous acquisition of environmental data in front of and to the sides of the flight also includes:

[0016] The solid-state lidar and the camera are synchronized to collect data through a hardware synchronization signal, and the data from all sensors are unified into the UAV body coordinate system.

[0017] Furthermore, the collected environmental data undergoes three-layer fusion perception processing to identify and construct a real-time obstacle map containing obstacle location, category, and size information, including:

[0018] Data layer fusion: Associate and register the synchronously acquired LiDAR point cloud data with image pixels to form an enhanced point cloud with color features;

[0019] Feature layer fusion: The point cloud-based 3D obstacle detection neural network and the image-based 2D semantic segmentation neural network are run in parallel to output three-dimensional obstacle boxes and two-dimensional semantic segmentation results, respectively.

[0020] Decision-level fusion: Associate the three-dimensional obstacle box and the two-dimensional semantic segmentation result. When the vision identifies a specific type of obstacle but the point cloud data is sparse, generate virtual obstacles of the corresponding type and size in the real-time obstacle map.

[0021] Furthermore, in the decision-making layer fusion, the obstacles of a specific category are wires, tree branches, or nets; the virtual obstacles are generated as virtual cylinders or virtual planes with a preset diameter according to their category.

[0022] Furthermore, the factors on which the risk quantification model calculates the risk value include at least: the distance between the drone and the nearest obstacle, the relative speed between the drone and the obstacle, the preset category weight of the obstacle, and the current attitude angular velocity of the drone.

[0023] Furthermore, real-time detection of whether the risk value exceeds a preset emergency hovering threshold includes:

[0024] Two risk thresholds are preset: a warning threshold and an emergency hovering threshold; when the risk value exceeds the warning threshold but does not reach the emergency hovering threshold, a warning signal is sent to the conventional flight control system; emergency hovering is only executed when the risk value exceeds the emergency hovering threshold.

[0025] Furthermore, the emergency hovering control logic includes at least the following: immediately increasing the drone motor thrust to 115% of the current hovering thrust, enabling the high-gain attitude angular rate (PD) controller to stabilize the drone attitude, locking the current yaw angle, and setting the target vertical velocity to zero.

[0026] Furthermore, after executing the emergency hover control logic, the drone remains in hover until the risk value drops below the warning threshold and a release command is received from the pilot or the upper-level system, at which point it exits the emergency hover state and resumes normal control.

[0027] Furthermore, hardware interrupt signals are sent directly to the corresponding control module through a separate hardware interrupt path.

[0028] According to a second aspect of the present invention, a system for early detection and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV) is provided, the system comprising:

[0029] A multi-sensor array is used to simultaneously collect environmental data in front of and to the sides of the flight path;

[0030] The embedded fusion computing unit is used to perform three-layer fusion perception processing on the collected environmental data, identify and build a real-time obstacle map containing obstacle location, category and size information;

[0031] The rapid decision-making and arbitration module is used to dynamically calculate the risk value of the current flight based on the real-time obstacle map and the flight status information of the UAV through a risk quantification model; to detect in real time whether the risk value exceeds the preset emergency hovering threshold; and to immediately generate a hardware interrupt signal with the highest priority when the risk value exceeds the emergency hovering threshold.

[0032] The flight control execution module is used to respond to the hardware interrupt signal, bypass the conventional flight control loop of the UAV, and directly execute the emergency hovering control logic to control the UAV to achieve stable hovering at the current position.

[0033] The embodiments of the present invention have the following advantages:

[0034] This invention employs a multi-sensor array to synchronously collect environmental data; performs three-layer fusion perception processing (data layer, feature layer, and decision layer) on the data to construct a real-time obstacle map; and generates virtual obstacles when the point cloud is sparse through a semantic completion mechanism. Based on the obstacle map and flight status, a risk quantification model is used to dynamically calculate the flight risk value. When the risk value exceeds the emergency hovering threshold, a hardware interrupt signal is immediately generated. In response to this interrupt signal, flight control is directly taken over, and emergency hovering logic, including thrust boost and high-gain attitude stabilization, is executed, enabling the UAV to hover instantly. This invention solves the problems of inaccurate perception of small obstacles and large obstacle avoidance response delays in complex near-ground scenarios, achieving high detection rate and millisecond-level response for proactive safety protection, significantly improving the flight safety and reliability of UAVs in narrow and dynamic environments. Attached Figure Description

[0035] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0036] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0037] Figure 1 This is a schematic diagram of the logical structure of a UAV near-ground complex obstacle pre-sensing and emergency hovering system provided in an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of the overall hardware and software architecture of a UAV near-ground complex obstacle pre-sensing and emergency hovering system provided for an embodiment of the present invention;

[0039] Figure 3 This is a flowchart illustrating a method for anti-obstacle detection and emergency hovering of complex obstacles near the ground provided by an embodiment of the present invention;

[0040] Figure 4This is a schematic diagram of the software algorithm and logic control flow in a method for anti-obstacle pre-sensing and emergency hovering of unmanned aerial vehicles (UAVs) near-ground complex obstacles provided in an embodiment of the present invention.

[0041] Figure 5 A schematic diagram of three-layer fusion perception in a method for pre-sensing and emergency hovering of complex obstacles near the ground provided in an embodiment of the present invention, (a) original data, (b) neural network output, (c) fused map containing virtual obstacles;

[0042] Figure 6 This is a schematic diagram of the emergency hovering control logic in a method for early perception and emergency hovering of complex obstacles near the ground provided in an embodiment of the present invention, illustrating how the interruption signal bypasses the conventional control loop;

[0043] Figure 7 This is a comparative illustration of application scenarios in a method for anti-obstacles and emergency hovering of unmanned aerial vehicles (UAVs) near the ground, provided in an embodiment of the present invention.

[0044] Figure 8 This is a schematic diagram of sensor installation position and coordinate system calibration in a method for early sensing and emergency hovering of complex obstacles near the ground provided by an embodiment of the present invention;

[0045] Figure 9 This is a schematic diagram illustrating the risk quantification model calculation process in a method for pre-sensing and emergency hovering of complex obstacles near the ground for unmanned aerial vehicles (UAVs) according to an embodiment of the present invention.

[0046] Figure 10 This is a schematic diagram comparing the software timing of the normal mode and the emergency interruption mode in a method for anti-obstacle pre-sensing and emergency hovering of unmanned aerial vehicles near complex obstacles provided in an embodiment of the present invention. Detailed Implementation

[0047] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] To overcome the shortcomings of the existing technology, to achieve high-precision and robust real-time perception of small and complex obstacles near the ground, and to bypass the conventional path planner and directly trigger the underlying flight controller to perform emergency hovering when the collision risk exceeds the threshold, so as to avoid collisions to the greatest extent.

[0049] refer to Figure 1 and Figure 2This invention discloses a near-ground complex obstacle pre-sensing and emergency hovering system for unmanned aerial vehicles (UAVs). The system includes: a multi-sensor array 1; an embedded fusion computing unit 2; a rapid decision-making and arbitration module 3; and a flight control execution module 4.

[0050] Multi-sensor array 1:

[0051] Hardware configuration: Includes one forward-facing main sensing subarray and multiple auxiliary sensing subarrays (such as lateral and downward-facing subarrays). The forward-facing main sensing subarray includes at least:

[0052] One solid-state LiDAR: used to acquire high-precision 3D point cloud data within a 120° horizontal field of view and a 30° vertical field of view, focusing on detecting linear objects such as power lines and railings. During installation, its axis should be tilted downwards at a 5-10° angle to the direction of the drone's movement to optimize near-ground coverage.

[0053] A high frame rate global shutter monocular camera: triggered by a hardware synchronization signal (such as a synchronization pulse generated by an FPGA) with LiDAR to achieve spatiotemporal data synchronization, used for texture recognition and dynamic object detection.

[0054] A set of millimeter-wave radars (such as 77GHz) are used to provide relative speed information of obstacles and provide stable distance information in adverse weather conditions such as rain, fog, and dust.

[0055] All sensors are aligned with the UAV's body coordinate system through offline calibration (e.g., using a checkerboard calibration board for LiDAR-camera extrinsic parameter calibration).

[0056] Embedded Fusion Computing Unit 2:

[0057] Employ high-performance embedded AI computing platforms (such as NVIDIA Jetson AGX Orin).

[0058] Running the three-layer fusion sensing algorithm:

[0059] Data layer fusion: By projecting the point cloud and using bilinear interpolation, the synchronized LiDAR point cloud is associated with the image pixels, and the point cloud is given RGB color features to form an enhanced point cloud.

[0060] Feature layer fusion: running two optimized neural networks in parallel:

[0061] Branch 1: 3D Obstacle Detection Network: The PointPillars network is adopted, and its point cloud pillarization mesh size is set to 0.05m to enhance the feature extraction capability of small objects such as wires.

[0062] Branch 2: 2D Semantic Segmentation Network: Employs a lightweight DeepLabv3+ network with MobileNetV2 as the backbone to identify nearly 10 types of obstacles, including sky, building, wire, branch, pole, etc.

[0063] Decision-level fusion: Connecting the outputs of the two branches. When the visual system recognizes a "wire" but the LiDAR point cloud is sparse, the system generates a virtual cylindrical obstacle with a length of 5 meters and a diameter of 0.05 meters in 3D space based on the camera model and the UAV pose. This virtual obstacle is then injected into the final obstacle map, with a confidence level of 80% of the visual segmentation confidence level.

[0064] Rapid Decision-Making and Arbitration Module 3:

[0065] Used to continuously monitor the real-time obstacle map output by the fusion sensing unit.

[0066] An internal "emergency risk quantification model" is maintained. This model dynamically calculates the risk score (RiskScore, RS) based on the following formula: Where: α, β, γ, δ are adjustable weight coefficients (example values: 40, 0.5, 30, 5); D_min is the distance to the nearest obstacle (meters); V_rel is the relative velocity (meters / second); W_category is the obstacle category weight (wire = 1.0, tree branch = 0.7); A_attitude is the magnitude of the UAV's attitude angular velocity (radians / second).

[0067] Set two thresholds: a warning threshold (RS_warn) and an emergency hovering threshold (RS_emergency, e.g., 85).

[0068] When RS > RS_warn, the system sends a warning signal to the conventional flight control system.

[0069] When RS > RS_emergency, the module will immediately issue the highest priority hardware interrupt signal, directly skipping all upper-level path planning algorithms.

[0070] Flight control execution module 4 includes a proprietary interrupt service routine for receiving the aforementioned hardware interrupts.

[0071] Once an interrupt is triggered, the following emergency hover controller pseudocode logic will be executed immediately:

[0072] void ISR_EmergencyHover() {

[0073] disable_standard_controller(); / / Disable the standard controller

[0074] float hover_thrust = get_current_estimated_hover_thrust();

[0075] set_motor_thrust(hover_thrust * 1.15); / / Increase thrust to 115%

[0076] / / Enable high-gain angular rate PD controller (P=8.0, D=0.5)

[0077] set_angle_rate_PD_gain(roll, P=8.0, D=0.5);

[0078] set_angle_rate_PD_gain(pitch, P=8.0, D=0.5);

[0079] set_yaw_lock(current_yaw); / / Lock yaw

[0080] set_vertical_velocity_target(0); / / Target vertical velocity is zero

[0081] emergency_hover_flag = TRUE; / / Set the hover flag}

[0082] The control objective is to achieve attitude stability within a 1-meter cube space at the current location. The hovering state will remain until the risk value RS < RS_warn and is confirmed by the pilot or the upper-level system, at which point it can be released.

[0083] The invention will be further described in detail below using a specific example of a logistics drone flying in the alleys of an old urban area.

[0084] 1. System Initialization

[0085] Before takeoff, the system performs a self-check and loads calibration parameters. RS_warn = 50, RS_emergency = 85. In the risk model, the obstacle category weight is set to: power line = 1.0.

[0086] 2. Cruise and Sensing

[0087] The drone flew at a speed of 5 meters per second. The solid-state LiDAR (such as Livox Avia) operated in a forward non-repeating scan mode. When the drone flew to a distance of about 8 meters from the obstacle in front, the camera image was used by a semantic segmentation network to identify the horizontal "wires," but the LiDAR point cloud remained sparse.

[0088] 3. Integration and Risk Calculation

[0089] The decision-level fusion algorithm is activated, and a virtual cylindrical obstacle is generated in 3D space based on the identified wire pixel bands.

[0090] The rapid decision-making and arbitration module begins calculating the RS value. As the distance D_min decreases from 8 meters to 4.5 meters, the RS value rapidly rises to 60 due to the high weight of the "wire" category, exceeding RS_warn. The conventional flight control system's attempt to plan an obstacle avoidance path fails (narrow alleyway).

[0091] 4. Triggering and executing emergency hover

[0092] When D_min=3 meters, the RS value is calculated to be 90, which instantaneously exceeds the RS_emergency threshold.

[0093] The rapid decision-making and arbitration module immediately generates a hardware interrupt signal.

[0094] The interrupt service routine ISR_EmergencyHover() of the flight control execution module is triggered, instantly taking over control, setting the motor thrust to 115%, and enabling the high-gain PD controller.

[0095] The drone achieved a stable hover in about 1.5 seconds, with its nose about 1.2 meters away from the power line.

[0096] 5. Follow-up handling

[0097] The drone remains hovering, awaiting remote instructions. Once the pilot confirms, they manually maneuver the drone to increase altitude and clear the obstacle; the system then deactivates the emergency status.

[0098] Corresponding to the aforementioned UAV near-ground complex obstacle pre-sensing and emergency hovering system, this invention also discloses a UAV near-ground complex obstacle pre-sensing and emergency hovering method. The following details the UAV near-ground complex obstacle pre-sensing and emergency hovering method disclosed in this invention, in conjunction with the aforementioned UAV near-ground complex obstacle pre-sensing and emergency hovering system.

[0099] refer to Figures 3 to 10 This invention discloses a method for anti-obstacle prediction and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV), the method comprising:

[0100] Simultaneous acquisition of environmental data in front of and to the sides of the flight using a multi-sensor array;

[0101] The collected environmental data is processed by a three-layer fusion perception system to identify and construct a real-time obstacle map containing information on the location, type, and size of obstacles.

[0102] Based on the real-time obstacle map and the flight status information of the UAV, the risk value of the current flight is dynamically calculated through a risk quantification model.

[0103] The system detects in real time whether the risk value exceeds a preset emergency hovering threshold. When the risk value exceeds the emergency hovering threshold, it immediately generates a hardware interrupt signal with the highest priority.

[0104] In response to the hardware interrupt signal, the system bypasses the drone's conventional flight control loop and directly executes the emergency hovering control logic to control the drone to achieve stable hovering at its current position.

[0105] Furthermore, the multi-sensor array includes at least one solid-state LiDAR, one high frame rate global shutter camera, and a set of millimeter-wave radars;

[0106] Simultaneous acquisition of environmental data in front of and to the sides of the flight also includes:

[0107] The solid-state lidar and the camera are synchronized to collect data through a hardware synchronization signal, and the data from all sensors are unified into the UAV body coordinate system.

[0108] Furthermore, the collected environmental data undergoes three-layer fusion perception processing to identify and construct a real-time obstacle map containing obstacle location, category, and size information, including:

[0109] Data layer fusion: Associate and register the synchronously acquired LiDAR point cloud data with image pixels to form an enhanced point cloud with color features;

[0110] Feature layer fusion: The point cloud-based 3D obstacle detection neural network and the image-based 2D semantic segmentation neural network are run in parallel to output three-dimensional obstacle boxes and two-dimensional semantic segmentation results, respectively.

[0111] Decision-level fusion: Associate the three-dimensional obstacle box and the two-dimensional semantic segmentation result. When the vision identifies a specific type of obstacle but the point cloud data is sparse, generate virtual obstacles of the corresponding type and size in the real-time obstacle map.

[0112] Furthermore, in the decision-making layer fusion, the obstacles of a specific category are wires, tree branches, or nets; the virtual obstacles are generated as virtual cylinders or virtual planes with a preset diameter according to their category.

[0113] Furthermore, the factors on which the risk quantification model calculates the risk value include at least: the distance between the drone and the nearest obstacle, the relative speed between the drone and the obstacle, the preset category weight of the obstacle, and the current attitude angular velocity of the drone.

[0114] Furthermore, real-time detection of whether the risk value exceeds a preset emergency hovering threshold includes:

[0115] Two risk thresholds are preset: a warning threshold and an emergency hovering threshold; when the risk value exceeds the warning threshold but does not reach the emergency hovering threshold, a warning signal is sent to the conventional flight control system; emergency hovering is only executed when the risk value exceeds the emergency hovering threshold.

[0116] Furthermore, the emergency hovering control logic includes at least the following: immediately increasing the drone motor thrust to 115% of the current hovering thrust, enabling the high-gain attitude angular rate (PD) controller to stabilize the drone attitude, locking the current yaw angle, and setting the target vertical velocity to zero.

[0117] Furthermore, after executing the emergency hover control logic, the drone remains in hover until the risk value drops below the warning threshold and a release command is received from the pilot or the upper-level system, at which point it exits the emergency hover state and resumes normal control.

[0118] Furthermore, hardware interrupt signals are sent directly to the corresponding control module through a separate hardware interrupt path.

[0119] The solid-state lidar in the forward-facing main sensing subarray can be replaced with a multi-line rotating lidar; the camera can be replaced with an infrared thermal imager for nighttime detection.

[0120] The decision-level fusion can generate virtual planar obstacles (such as when identified as "mesh") based on the 2D segmentation results.

[0121] If space is detected above, a "emergency hover + slight vertical ascent" combination strategy can be triggered.

[0122] It can be integrated into the final approach and landing phase of large unmanned helicopters or eVTOL (electric vertical takeoff and landing aircraft).

[0123] This invention delivers four significant benefits: First, significantly enhanced perception capabilities. Through a unique three-layer fusion and semantic completion mechanism, the system's detection rate of small obstacles such as power lines is dramatically increased from 62% in traditional solutions to 98% (simulated test data). Second, extremely fast response speed. Utilizing a hardware interrupt pass-through control design, the latency of the complete "perception-decision-execution" link is reduced from the conventional 120 milliseconds to less than 15 milliseconds, achieving near-instantaneous response. Third, optimal safety strategy. The adopted emergency hovering mechanism provides the most reliable safety backup, successfully reducing the collision rate from 4.7% to 0.1% in 1000 tunnel flight simulations, greatly ensuring flight safety. Fourth, high overall system robustness. Thanks to the complementary advantages of multiple sensors and a well-considered risk quantification model, the system's reliability and adaptability are significantly enhanced in various complex and harsh environments.

[0124] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for anti-obstacle prediction and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV), characterized in that, The method includes: Simultaneous acquisition of environmental data in front of and to the sides of the flight using a multi-sensor array; The collected environmental data is processed by a three-layer fusion perception system to identify and construct a real-time obstacle map containing information on the location, type, and size of obstacles. Based on the real-time obstacle map and the flight status information of the UAV, the risk value of the current flight is dynamically calculated through a risk quantification model. The system detects in real time whether the risk value exceeds a preset emergency hovering threshold. When the risk value exceeds the emergency hovering threshold, it immediately generates a hardware interrupt signal with the highest priority. In response to the hardware interrupt signal, the system bypasses the drone's conventional flight control loop and directly executes the emergency hovering control logic to control the drone to achieve stable hovering at its current position.

2. The method for pre-sensing and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV) as described in claim 1, characterized in that, The multi-sensor array includes at least one solid-state lidar, one high frame rate global shutter camera, and a set of millimeter-wave radars; Simultaneous acquisition of environmental data in front of and to the sides of the flight also includes: The solid-state lidar and the camera are synchronized to collect data through a hardware synchronization signal, and the data from all sensors are unified into the UAV body coordinate system.

3. The method for pre-sensing and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV) as described in claim 2, characterized in that, The collected environmental data undergoes three-layer fusion perception processing to identify and construct a real-time obstacle map containing obstacle location, category, and size information, including: Data layer fusion: Associate and register the synchronously acquired LiDAR point cloud data with image pixels to form an enhanced point cloud with color features; Feature layer fusion: The point cloud-based 3D obstacle detection neural network and the image-based 2D semantic segmentation neural network are run in parallel to output three-dimensional obstacle boxes and two-dimensional semantic segmentation results, respectively. Decision-level fusion: Associate the three-dimensional obstacle box and the two-dimensional semantic segmentation result. When the vision identifies a specific type of obstacle but the point cloud data is sparse, generate virtual obstacles of the corresponding type and size in the real-time obstacle map.

4. The method for pre-sensing and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle as described in claim 3, characterized in that, In the decision-making layer fusion, the obstacles of a specific category are wires, tree branches, or nets; the virtual obstacles are generated as virtual cylinders or virtual planes with a preset diameter according to their category.

5. The method for pre-sensing and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle as described in claim 1, characterized in that, The risk quantification model calculates the risk value based on factors including at least: the distance between the drone and the nearest obstacle, the relative speed between the drone and the obstacle, the preset category weight of the obstacle, and the current attitude angular velocity of the drone.

6. The method for pre-sensing and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle as described in claim 1, characterized in that, Real-time detection of whether the risk value exceeds a preset emergency hovering threshold includes: Two risk thresholds are preset: a warning threshold and an emergency hovering threshold; when the risk value exceeds the warning threshold but does not reach the emergency hovering threshold, a warning signal is sent to the conventional flight control system; emergency hovering is only executed when the risk value exceeds the emergency hovering threshold.

7. The method for anti-obstacle prediction and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle as described in claim 1, characterized in that, The emergency hovering control logic includes at least the following: immediately increasing the drone motor thrust to 115% of the current hovering thrust, enabling the high-gain attitude angular rate (PD) controller to stabilize the drone attitude, locking the current yaw angle, and setting the target vertical velocity to zero.

8. A method for anti-obstacle prediction and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle as described in claim 7, characterized in that, After executing the emergency hover control logic, the drone remains hovered until the risk value drops below the emergency hover threshold and a release command is received from the pilot or the upper-level system before exiting the emergency hover state and resuming normal control.

9. A method for anti-obstacle prediction and emergency hovering of complex obstacles near the ground by an unmanned aerial vehicle (UAV) as described in claim 1, characterized in that, Hardware interrupt signals are sent directly to the corresponding control module through a separate hardware interrupt path.

10. A system for early detection and emergency hovering of complex obstacles near the ground on unmanned aerial vehicles (UAVs), characterized in that, The system includes: A multi-sensor array is used to simultaneously collect environmental data in front of and to the sides of the flight path; The embedded fusion computing unit is used to perform three-layer fusion perception processing on the collected environmental data, identify and build a real-time obstacle map containing obstacle location, category and size information; The rapid decision-making and arbitration module is used to dynamically calculate the risk value of the current flight based on the real-time obstacle map and the flight status information of the UAV through a risk quantification model; to detect in real time whether the risk value exceeds the preset emergency hovering threshold; and to immediately generate a hardware interrupt signal with the highest priority when the risk value exceeds the emergency hovering threshold. The flight control execution module is used to respond to the hardware interrupt signal, bypass the conventional flight control loop of the UAV, and directly execute the emergency hovering control logic to control the UAV to achieve stable hovering at the current position.

Images