A method for near-ground warning of a helicopter

Abstract

The application discloses a helicopter near-ground alarm method, and relates to the technical field of flight control, which comprises the following steps: judging the flight state according to a wheel load mechanism, acquiring flight data in real time and comparing the flight data with pre-set alarm boundaries to determine an alarm mode, adjusting a stability augmentation coefficient in real time according to the current state of the helicopter and the alarm boundaries, and then generating a control strategy; determining a transition mechanism according to the control strategy combined with core associated parameters, determining a solution according to the transition mechanism, and eliminating the alarm; driving bidirectional smooth switching through the wheel load mechanism, real-time association of the alarm system and the flight control system, setting the alarm mode and the alarm boundaries, adjusting the stability augmentation coefficient, and acquiring the core associated parameters; precisely adapting to the take-off characteristics of the helicopter, recognizing the take-off state in advance, improving the response speed, shortening the reaction time of the pilot, and preventing attitude out-of-control.

Description

Technical Field

[0001] This invention relates to the field of flight control technology, and in particular to a method for ground proximity warning of helicopters. Background Technology

[0002] In the field of helicopter flight safety, traditional ground proximity warning systems mainly rely on fixed warning boundaries, often based on conservative design principles. This leads to a high false alarm rate in actual flight, limiting the helicopter's maneuverability, especially in complex terrain and low-altitude flight environments. Furthermore, pilot manuals for large, conventional helicopters require pilots to disconnect the automatic flight control function during the ground taxiing phase before takeoff and after landing. This is to prevent the automatic control system from generating corrective signals to the helicopter at the end of pilot operation, which could damage the main and tail rotors or cause unintended attitude corrections to the helicopter.

[0003] However, existing systems often separate safety control during ground taxiing from that during near-ground flight, lacking comprehensive adaptive protection across all phases. During ground taxiing and near-ground flight, the lack of an effective flight control system means pilots must operate the helicopter manually. Improper operation or unexpected situations can easily lead to accidents such as rotor strikes. Summary of the Invention

[0004] This application provides a helicopter ground proximity warning method that solves the problems of high false alarm rate and poor adaptability at all stages in the prior art. It realizes dynamic modal coordination and accurate identification of flight status, effectively reducing the false alarm rate and improving the handling stability and safety of helicopters.

[0005] This application provides a helicopter ground proximity warning method, the method includes: S1: judging the flight status according to the wheel-mounted mechanism, acquiring flight data in real time and comparing it with a pre-set warning boundary to determine the warning mode, adjusting the stability enhancement coefficient in real time according to the current helicopter status and the warning boundary, and then generating a control strategy; S2: Determine the transition mechanism based on the control strategy and core related parameters, and determine the solution based on the transition mechanism to eliminate the alarm; The wheel-mounted mechanism performs logical operations on the discrete signals of the left, right, and front wheel-mounted sensors to determine the helicopter's flight status. The flight status includes both ground status and air status. If any wheel-mounted sensor outputs an "air" signal, the entire aircraft is determined to be in an air status. The core associated parameters include radio altitude value, ground stabilization output value, air control target value, and radio altitude integral abrupt change. The standard deviation is calculated based on the ground stabilization output value and the air control target value. The standard deviation is used to quantify the switching amplitude and provide a transition benchmark. The transition mechanism includes the takeoff process and the landing process.

[0006] Furthermore, the alarm modes include alarm mode one: the descent rate is greater than the alarm boundary; alarm mode two: the terrain proximity rate is greater than the alarm boundary. The alarm boundary is a set of critical safety data values ​​corresponding to various safety indicators; the stability enhancement coefficient is a key parameter in the flight control law, used to adjust the feedback gain. The descent rate refers to the risk of a helicopter crashing due to its high vertical speed, while the terrain approach rate refers to the risk of a helicopter hitting the ground due to changes in terrain slope during low-altitude flight.

[0007] Furthermore, the core correlation parameter is a key dynamic variable driving the bidirectional smooth switching, determining the transition ratio and switching sequence; the ground stabilization output is calculated in real time based on the wheel-mounted ground state and stabilization coefficient, and the air control target value is calculated based on the wheel-mounted air state and dynamic stabilization coefficient. The transition ratio is a dynamic weighting coefficient for the smooth transition of control commands from the current mode to the target mode. It is used to quantify the degree of mixing of the two modes in the current control command, and its value range is [0%, 100%]. The switching timing is the key time node control of the state transition process to ensure that the change of the transition ratio matches the helicopter dynamic response.

[0008] Furthermore, regarding the takeoff process: a transition phase is set based on the standard deviation, which includes two phases: rapid integration and altitude decay. The rapid integration phase involves uniformly superimposing 50% of the standard deviation onto the air control output through an integrator within 0.2 seconds. The altitude decay phase refers to calculating the real-time transition amount by attenuating the remaining standard deviation according to altitude using a predefined decay function. For the landing process: a pre-buffering mechanism is implemented, in which 10% of the air control output value is added to the ground stabilization command in advance, and the addition ratio is increased linearly as the altitude decreases.

[0009] Furthermore, the method also includes: acquiring terrain features based on a visual sensor, generating a fused image, and outputting pixel-level semantic labels; assigning friction coefficients according to the semantic labels, generating a terrain digital elevation model according to a point cloud registration algorithm, and calculating the base slope; fusing the friction coefficients and the base slope to generate a terrain feature vector; acquiring real-time flight parameters and calculating the downwash velocity according to the base slope, and correcting the terrain feature vector according to the downwash velocity; the real-time flight parameters include rotor speed, airspeed, and flight altitude.

[0010] Furthermore, a three-dimensional terrain model is regenerated based on the corrected terrain feature vector, and the terrain contour features are extracted; historical safe trajectories with a similarity greater than 90% to the terrain contour features are retrieved from the pre-established safe flight database; if the match is successful, the corresponding safe boundary is invoked; if the match fails, the safe boundary is reconstructed; the final safe boundary output after the matching is completed is used as the alarm boundary.

[0011] Furthermore, based on the acquisition of terrain features by visual sensors, a fused image is generated, including: the visual sensors including a visible light camera, an infrared thermal imager, and a millimeter-wave radar; the visible light camera and the infrared thermal imager simultaneously scan the terrain in front to generate an infrared image; the millimeter-wave radar scans the terrain at a 10-degree cone angle and outputs point cloud data; the infrared image and the point cloud image are fused to obtain a fused image.

[0012] Furthermore, the pixel-level semantic label refers to assigning a specific category label to each pixel in the image, representing terrain features; The friction coefficient is assigned based on semantic tags, including: pre-determining the range of terrain friction coefficients corresponding to each semantic tag according to the physical characteristics of different terrain types; acquiring pixel-level semantic tag images obtained after semantic segmentation, traversing the image pixels, searching for the corresponding friction coefficient value from the mapping table according to its semantic tag; assigning the found friction coefficient value to the pixel, and finally obtaining an image of the same size as the semantic tag image, where each pixel stores the corresponding friction coefficient, i.e., a friction coefficient distribution map, and assigning the friction coefficient.

[0013] Furthermore, the friction coefficient and the base slope are fused to generate a terrain feature vector, including: associating the friction coefficient image and the base slope image with the same spatial location, and combining the friction coefficient value and the base slope value at each location into a vector as the terrain feature vector at that location; The downwash velocity refers to the speed of the downward airflow generated below the rotor during the flight of the aircraft, which is used to reflect the degree of influence of the aircraft on the airflow environment of the terrain below. Based on the influence of downwash flow on terrain friction coefficient and base slope, a single-factor correction model is pre-established. Based on the established single-factor correction models, the terrain feature vector of each pixel in the terrain feature vector map is corrected and recombined into a corrected terrain feature vector.

[0014] One or more technical solutions provided in this application have at least the following technical effects or advantages: By employing a wheel-mounted mechanism, establishing a real-time connection between the alarm system and the flight control system, setting alarm modes and alarm boundaries, adjusting stability augmentation coefficients, and acquiring core correlation parameters to drive smooth bidirectional switching, this system precisely adapts to the helicopter's takeoff characteristics. It can identify takeoff status in advance, avoiding premature switching to ground mode. When alarm boundaries are breached, it quickly identifies abnormal values, issues visual / audio alarms, and automatically adjusts flight control parameters, transforming passive warnings into active protection, improving response speed, shortening pilot reaction time, and preventing attitude loss of control. This solves the problem of alarm and flight control disconnect in traditional systems, shortening response time and reducing accident risks through active control intervention. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a helicopter ground proximity warning method according to an embodiment of the present invention. Detailed Implementation

[0016] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0018] Example 1: As Figure 1 As shown, a helicopter ground proximity warning method includes: S1: Determines the flight status based on the wheel-mounted mechanism, acquires flight data in real time and compares it with the pre-set alarm boundaries to determine the alarm mode, and adjusts the stability augmentation coefficient in real time according to the current helicopter status and alarm boundaries, thereby generating a control strategy.

[0019] The wheel-mounted mechanism performs logical operations on the discrete signals from the left, right, and front wheel-mounted sensors to determine the helicopter's flight status, which includes both ground and air states. If any one wheel-mounted sensor outputs an "air" signal, the helicopter is determined to be in an air state; only when all three wheel-mounted sensors output "ground" signals is the helicopter determined to be in a ground state. Preferably, the wheel-mounted mechanism can use two JGW-3019A opto-MOS relays (JK1 and JK2) to build the logic circuit. The circuit design is equivalent to a three-input OR gate, with each relay performing an OR operation on two signals; the two-stage series connection covers all three signals. Signal input must be continuous for >10ms to be acknowledged (to prevent false judgments due to momentary jitter); the relays are optocoupled to avoid electromagnetic interference.

[0020] This application overcomes the limitations of traditional systems where single-wheel sensors are prone to falsely triggering the "airborne" state due to mechanical vibration or partial ground lift, and where single logic (such as "both main wheels lift off simultaneously") cannot adapt to scenarios like slope landing or single-side ground contact. Even in the event of a single-wheel failure, the other two wheels can still correctly determine the state (e.g., a left wheel failure outputs "airborne," while the right / front wheel outputs "ground": the combined output is "airborne," conforming to the principle of safety conservatism). It precisely adapts to the characteristics of helicopter takeoff, where helicopters often lift off with one main wheel or front wheel first (affected by uneven lift from the main rotor). The wheel-mounted mechanism can identify the takeoff state 50-200ms in advance; the probability of all three points touching the ground simultaneously during landing is low, avoiding premature switching to ground mode.

[0021] In some embodiments, the alarm mode and linkage mechanism are a dynamically coupled process, referring to the establishment of a real-time association between the alarm system and the flight control system. When the alarm system identifies the helicopter as being in a threatening state (such as excessive descent rate or terrain approach rate), the flight control system no longer passively waits for pilot intervention but actively adjusts the control law parameters. Specifically, the alarm system monitors flight parameters (such as descent rate, airspeed, and terrain slope) in real time. If the parameters exceed preset alarm boundaries, the system immediately triggers an alarm and determines the alarm mode. Simultaneously, based on the exceeded boundary values ​​and the current flight state, the flight control system dynamically calculates the correction amount of the stability augmentation coefficient. The corrected stability augmentation coefficient is directly applied to the control law to enhance the feedback signals of the pitch, roll, or azimuth channels, thereby improving helicopter stability.

[0022] In some embodiments, the number of alarm modes shall not be less than two, including alarm mode one: descent rate greater than alarm threshold; and alarm mode two: terrain approach rate greater than alarm threshold. The descent rate addresses the risk of the helicopter crashing due to excessive vertical speed, and the terrain approach rate addresses the risk of the helicopter hitting the ground due to changes in terrain slope during low-altitude flight. For alarm mode one, the real-time descent rate (obtained from the barometric altimeter or inertial measurement unit) and wheel load status are acquired. If the descent rate continuously exceeds the alarm threshold (e.g., >20 feet / second for more than 5 seconds) and the wheel load status is "in the air" (not in the ground taxiing phase), alarm mode one is determined and activated. For alarm mode two, the terrain slope (calculated from the radar altimeter and terrain database), airspeed (obtained from the airspeed meter), and economic speed threshold (efficiency speed defined in the helicopter manual) are acquired. If the terrain slope exceeds the alarm threshold (steep slope) and the airspeed > economic speed (e.g., >80 knots), alarm mode two is determined and activated.

[0023] The alarm boundary is a set of critical safety data values ​​corresponding to various safety indicators, including at least the descent rate and terrain approach rate. The specific values ​​need to be pre-set according to the type of alarm mode; this application does not impose specific limitations here. A large amount of historical data is collected, including extensive flight data (such as test flight records), and statistical threat events (such as accident precursors caused by excessive descent rates). A boundary fitting algorithm (such as optimization using the least squares method) is used to calculate the safety thresholds for parameters (descent rate, slope). The alarm boundary is dynamically calibrated based on real-time flight altitude and speed, taking into account helicopter type and environmental conditions to avoid overly strict thresholds leading to false alarms or overly lenient thresholds leading to missed detections.

[0024] When the alarm boundary is breached, the system quickly identifies abnormal values ​​and issues visual / audio alerts to the pilot; it automatically adjusts flight control parameters, i.e., the stability enhancement factor, to improve stability; and it transforms passive warnings into active protection. This improves response speed, shortens pilot reaction time, prevents loss of attitude control, and ensures that threat signals are seamlessly translated into control actions.

[0025] In some embodiments, the stability augmentation coefficient is a key parameter in the flight control law, used to adjust the feedback gain and ensure the helicopter's attitude stability. Specifically, it refers to the scaling factor (e.g., 75% indicates a gain of 75% of the standard air value) used in the control law to scale the angular rate feedback signals of the pitch, roll, or yaw channels, determining the system's response strength to disturbances: a higher coefficient results in stronger stability but may reduce maneuverability. In this embodiment, simulation using low-speed wind data at 0 altitude verifies that a coefficient of 75% meets the requirements of GJB902B. Therefore, the ground stability augmentation coefficient is optimized to 75% of the air value (verified through simulation) to balance stability during the taxiing phase.

[0026] In actual dynamic acquisition, the base stability augmentation coefficient is 75%. When alarm mode one is activated, the coefficient dynamically increases to 85% (base value + 10% correction). When alarm mode two is activated, the heading coefficient increases by 40% (e.g., from 75% to 105%). Automatic calibration based on historical control data (coefficient = 75% + 0.1 × (average control error - standard value)) ensures personalized adaptation. The stability augmentation coefficient changes in real time with the alarm status and is calculated in real time using sensor data (such as altitude and airspeed).

[0027] The stability enhancement factor, alarm mode, and alarm boundary form a closed-loop data flow. The alarm mode defines the threat type, the alarm boundary serves as the trigger condition, and the stability enhancement factor is the execution mechanism. The alarm mode determines the type of alarm boundary. When the alarm boundary is breached, the stability enhancement factor is adjusted, and the adjusted value is output, thus creating a coupling between threat perception and enhanced control.

[0028] In some embodiments, the stability augmentation coefficient is adjusted in real time based on the current helicopter status and alarm boundaries to generate a control strategy, which includes dynamic control commands for the pitch / heading channels (such as angular rate limit value +20%).

[0029] S2: Determine the transition mechanism based on the control strategy and core related parameters, and determine the solution based on the transition mechanism to eliminate the alarm.

[0030] The core correlation parameters are key dynamic variables driving smooth bidirectional handover, directly determining the transition ratio and handover timing. These core correlation parameters include radio altitude, ground stabilization output, airborne control target, and radio altitude integral mutation. The radio altitude is a real-time measured altitude above ground, used to calculate the transition ratio; the ground stabilization output refers to the currently executed control quantity, generated by the ground control law, including the stabilization signal after amplitude limiting; the airborne control target refers to the expected control quantity, dynamically calculated by the airborne control law based on flight status (e.g., warning mode) (e.g., pitch rate target); and the radio altitude integral mutation is used to correct the transition ratio to avoid misjudgment of terrain threats.

[0031] The ground stabilization output is calculated in real time based on the wheel-mounted ground state and stabilization coefficient, while the airborne control target value is calculated based on the wheel-mounted airborne state and dynamic stabilization coefficient. The airborne control target value = baseline control quantity × (1 + stabilization coefficient). The baseline control quantity is the standard control command of the helicopter in stable flight, generated in real time through the airborne control law. The control law dynamically calculates the baseline value based on the flight state (airspeed, altitude, attitude angle). These two represent control targets in different modes; direct switching will result in command jumps. A standard deviation is calculated based on the ground stabilization output value and the airborne control target value. This standard deviation is used to quantify the switching amplitude and provide a transition baseline; the larger the absolute value, the longer the transition time and distance, ensuring seamless connection between the switching start point and the current system state. If the standard deviation is greater than 0, it indicates that the ground command is greater than the airborne command, requiring a downward transition; conversely, it indicates that the ground command is less than the airborne target, requiring an upward transition.

[0032] The transition mechanism includes the takeoff and landing processes. For the takeoff process, a transition phase is set based on the standard deviation. This transition phase addresses the takeoff process and eliminates alarms. The transition phase includes two stages: rapid integration and altitude decay. The rapid integration stage involves uniformly superimposing 50% of the standard deviation onto the airborne control output via an integrator within 0.2 seconds. Specifically, 0.2 seconds is divided into 10 control cycles (each cycle being 20ms), and the superposition amount per cycle is calculated as follows: ,in It is a single-cycle superposition amount. It is the absolute value of the standard deviation; the output control output to the air is superimposed with the single-cycle superposition amount (the sign is determined by the positive or negative of the standard deviation) to set the output limit (e.g., ±10 units / second) to avoid instantaneous overload. Actual experiments have verified that a response delay of less than 0.2 seconds in the matched helicopter rotor system will cause pitch overshoot; a delay exceeding 0.2 seconds will fail to suppress attitude oscillations in the initial takeoff phase (flight test data shows that pitch angle fluctuations exceed ±1 degree when the delay exceeds 0.3 seconds). During the rapid integration phase, it can eliminate the sudden change in main rotor anti-torque at takeoff, solving the problem of emergency tail rotor compensation in traditional switching, and achieving rapid elimination of attitude angle fluctuations after takeoff. The altitude attenuation refers to attenuating the remaining standard deviation according to altitude, calculating the real-time transition amount using a predefined attenuation function, the formula of which is: Where Q is the transition amount, which refers to the absolute value of the remaining control command to be migrated (unit: degrees / second). It is the real-time value of the standard deviation. H is the current altitude, and H is the maximum altitude. Altitude attenuation prevents rotor downwash disturbances caused by high-command transitions at low altitudes, avoids time blind spots, and reduces power system load. In this embodiment, by segmenting the transition of the standard deviation, the active control optimization triggered by the alarm is seamlessly embedded into the state switching process.

[0033] The transition ratio is a dynamic weighting coefficient used to smoothly transition control commands from the current mode to the target mode. It quantifies the degree of mixing between the two modes (ground stabilization mode and air control mode) in the current control command, and its value ranges from [0%, 100%]. It is dynamically determined by the radio altitude value because altitude changes directly reflect the relative position of the helicopter to the ground and are the most reliable physical basis for state switching. The specific calculation formula is as follows: For the takeoff process: Regarding the landing process: Among them, the transition ratio of B, H represents the current altitude, and H is the maximum altitude the helicopter can reach. A transition ratio of 0% indicates that ground-based stabilization output is used entirely (the helicopter is not off the ground), while a transition ratio of 100% indicates that airborne control target values ​​are used entirely (stable flight state). The switching sequence is a key time node control in the state transition process, ensuring that the change in the transition ratio matches the helicopter's dynamic response. Specifically, it needs to be dynamically set according to the actual situation and the altitude changes during the state transition to ensure that the timing of the state adjustment perfectly matches the time changes of the helicopter's dynamic response.

[0034] For the air-to-ground transition, i.e., the landing process: a pre-buffering mechanism is implemented. This pre-buffering mechanism is a solution specifically for the landing process, aiming to eliminate alarms. The pre-buffering mechanism involves adding 10% of the airborne control output value to the ground stabilization command in advance, and then linearly increasing the addition ratio as altitude decreases. The specific transition ratio and switching sequence are obtained in the same way as during the landing process, and will not be elaborated upon here.

[0035] In this embodiment, a smooth two-way switching is driven by a wheel-mounted mechanism, establishing a real-time association between the alarm system and the flight control system, setting alarm modes and alarm boundaries, adjusting the stability augmentation coefficient, and acquiring core correlation parameters (radio altitude value, ground stability augmentation output value, airborne control target value, and radio altitude integral mutation). This precisely adapts to the helicopter's takeoff characteristics, allowing for early identification of takeoff status and preventing premature switching to ground mode. When alarm boundaries are breached, abnormal values ​​are quickly identified, visual / voice alarms are issued, and flight control parameters are automatically adjusted, transforming passive warnings into active protection, improving response speed, shortening pilot reaction time, and preventing attitude loss of control. It achieves smooth switching between ground and airborne control commands, eliminating sudden changes in main rotor anti-torque at takeoff, preventing rotor downwash disturbances caused by excessive low-altitude commands, avoiding time blind spots, and reducing power system load. It solves the problem of alarm and flight control disconnect in traditional systems (such as escape failure due to pilot operation delays) by actively intervening, shortening response time (by 300ms) and reducing accident risk.

[0036] Example 2: In unknown or complex terrain, rotor airflow interference leads to terrain measurement errors, and preset boundaries cannot dynamically adapt to the real physical environment. This solution, based on Example 1, further improves upon this by using terrain semantic segmentation, rotor flow field modeling, and topology matching to achieve boundary self-reconstruction, thereby enhancing alarm accuracy.

[0037] In this embodiment, multi-sensor fusion is used to identify terrain material and structure in real time, eliminating the dependence of traditional preset models on the static environment, thereby achieving dynamic compensation and boundary reconstruction.

[0038] The method further includes: acquiring terrain features based on a visual sensor, generating a fused image, and outputting pixel-level semantic labels; assigning friction coefficients based on the semantic labels, generating a digital elevation model of the terrain based on a point cloud registration algorithm, and calculating the base slope; fusing the friction coefficients and the base slope to generate a terrain feature vector; acquiring real-time flight parameters and calculating the downwash velocity based on the base slope, and correcting the terrain feature vector based on the downwash velocity. The real-time flight parameters include rotor speed, airspeed, and flight altitude.

[0039] A 3D terrain model is regenerated based on the corrected terrain feature vector, and the terrain contour features are extracted. Historical safe trajectories with a similarity greater than 90% to the terrain contour features are retrieved from the safe flight database. If a match is successful, the corresponding safety boundary is invoked. If a match fails, the safety boundary is reconstructed. The final safety boundary output after the matching is completed is used as the alarm boundary.

[0040] In some embodiments, the complementary characteristics of visual sensors (visible light + infrared) are utilized to capture terrain texture and thermal radiation features. Combined with the high-precision distance measurement capabilities of millimeter-wave radar, semantic segmentation and physical parameter extraction are achieved through a deep learning model. Based on the terrain features acquired by the visual sensors, a fused image is generated, including: the visual sensors comprising a visible light camera, an infrared thermal imager, and a millimeter-wave radar. The visible light camera and infrared thermal imager simultaneously scan the terrain ahead to generate an infrared image; the millimeter-wave radar scans the terrain at a 10-degree cone angle to output point cloud data; the infrared image and the point cloud image are fused to obtain a fused image.

[0041] In some embodiments, pixel-level semantic labels are output from the fused image. These pixel-level semantic labels assign a specific category label to each pixel in the image, representing the semantic category of the terrain or object corresponding to that pixel, such as grass, rocks, water, or buildings. In this way, the image is no longer merely a collection of pixels, but is endowed with rich semantic information, clearly expressing the meaning of different regions within the image. Transforming images from low-level pixel data into high-level semantic representations facilitates deeper image analysis and processing, enabling more accurate execution of tasks such as object detection and scene understanding based on semantic labels.

[0042] In flight safety scenarios, pixel-level semantic labeling helps identify terrain features, providing fundamental information for subsequent steps such as friction coefficient assignment and terrain feature vector generation. For example, knowing whether an area is grass or rock is crucial for assessing the safety of an aircraft landing or taxiing in that area. Suppose we have an image containing grass, rocks, and roads. After pixel-level semantic segmentation, each pixel in the image is labeled with its corresponding category. For instance, all pixels in the grass area are labeled "grass," all pixels in the rock area are labeled "rock," and all pixels in the road area are labeled "road." This clearly indicates the distribution of different terrain regions in the image, thereby enabling different levels of alerts.

[0043] In some embodiments, assigning friction coefficients based on semantic tags includes: pre-determining the range of terrain friction coefficients corresponding to each semantic tag based on the physical characteristics of different terrain types. For example, grassland typically has a relatively high friction coefficient, while smooth rock surfaces have a low friction coefficient, and water surfaces have a very low friction coefficient. A reasonable friction coefficient value or range is then set for each semantic tag using experimental or empirical data.

[0044] Specifically, the process involves reading the pixel-level semantic label image, obtaining the pixel-level semantic label image after semantic segmentation, where each pixel carries a corresponding semantic label. The image pixels are then traversed, and for each pixel, the corresponding friction coefficient value is looked up from the mapping table based on its semantic label. The found friction coefficient value is assigned to that pixel, resulting in an image of the same size as the semantic label image, where each pixel stores its corresponding friction coefficient—a friction coefficient distribution map. For example, the mapping table might specify a friction coefficient of 0.6 for grass, 0.3 for rock, and 0.1 for water. When traversing the semantic label image, if a pixel's semantic label is "grass," its friction coefficient is assigned 0.6; if it's "rock," it's 0.3; and if it's "water," it's 0.1.

[0045] In some embodiments, a terrain digital elevation model is generated based on a point cloud registration algorithm, and the basic slope is calculated. This includes registering multiple frames of point cloud data obtained from millimeter-wave radar scanning using a point cloud registration algorithm (such as the Iterative Closest Point (ICP) algorithm). The purpose of registration is to unify point cloud data acquired from different viewpoints or at different times into the same coordinate system, eliminate errors caused by changes in sensor position and attitude, and obtain a complete and accurate terrain point cloud dataset.

[0046] Interpolation processing is performed on the registered point cloud data to convert the discrete point cloud data into a continuous terrain surface model, i.e., a digital elevation model (DEM). Commonly used interpolation methods include inverse distance weighted interpolation and kriging interpolation. The DEM represents the terrain in the form of a regular grid, with each grid point storing the terrain height information at its corresponding location.

[0047] For each grid point in the DEM, calculate the height variation within a certain range around it (e.g., a 3x3 or 5x5 grid). The base slope of that grid point is obtained by calculating the ratio of the height difference to the horizontal distance. For example, for a grid point, calculate the height difference between it and its neighboring grid points, and then, based on these height differences and the grid spacing, calculate the base slope of that point using mathematical formulas (e.g., the maximum slope method, the average slope method, etc.).

[0048] In some embodiments, fusing the friction coefficient and the base slope to generate a terrain feature vector includes: mapping the friction coefficient image and the base slope image to the same spatial location, and combining the friction coefficient value and the base slope value at each location into a vector as the terrain feature vector for that location. For example, for a pixel location in the image with a friction coefficient value of f and a base slope value of s, the terrain feature vector for that location is (f, s).

[0049] Each pixel in the image is fused using the method described above, resulting in an image of the same size as the original image, where each pixel stores the corresponding terrain feature vector, i.e., a terrain feature vector map. This method comprehensively considers the frictional properties and undulations of the terrain, providing a more complete description of its features.

[0050] In some embodiments, downwash velocity refers to the speed of the downward-flowing airflow generated beneath the rotor or wing of an aircraft during flight, reflecting the degree of influence of the aircraft on the airflow environment of the terrain below. Specifically, parameters such as rotor speed, airspeed, and flight altitude of the aircraft are acquired in real time. Based on fluid dynamics principles and experimental data, a mathematical model is established regarding the relationship between rotor speed, airspeed, flight altitude, and downwash velocity. For example, downwash velocity may be directly proportional to rotor speed and inversely proportional to the square of flight altitude, etc. Specific parameters in the model are determined by fitting experimental data. The real-time acquired flight parameters are substituted into the downwash velocity model to calculate the downwash velocity value at the current moment. Downwash velocity affects the interaction between the aircraft and the terrain, such as changing the airflow distribution on the terrain surface, affecting the stability and safety of the aircraft during low-altitude flight. Accurate calculation of downwash velocity helps to more precisely assess the dynamic interaction between the aircraft and the terrain, providing a more reliable basis for flight safety decisions.

[0051] In some embodiments, the study investigates how downwash velocity affects the friction coefficient and base slope of the terrain. For example, downwash may stir up loose material (such as sand, gravel, etc.) on the ground, changing the surface roughness and thus affecting the friction coefficient; the impact force of downwash on the terrain may cause minor deformation of the terrain surface, affecting the measurement of the base slope.

[0052] Based on the influence of downwash flow on the terrain friction coefficient and foundation slope, a single-factor correction model is pre-established. For example, for the friction coefficient, a correction function based on the downwash flow velocity is established: ,in, It is the original coefficient of friction; is the corrected friction coefficient; k is a pre-set convergence coefficient used to prevent over-correction, and its value range is [0,1]. In this embodiment, the value is 0.2 after experimental verification. This refers to the downwash velocity; a similar single-factor correction model is also needed for the base slope. Based on the established single-factor correction models, the terrain feature vector of each pixel in the terrain feature vector map is corrected. For each pixel, its friction coefficient and base slope are first corrected according to the downwash velocity and the correction model, and then recombined to form the corrected terrain feature vector. This more accurately reflects the terrain characteristics of the aircraft under the current downwash environment.

[0053] In some embodiments, a 3D terrain model is regenerated based on the corrected terrain feature vectors, including: collecting the corrected terrain feature vectors, including key information such as the basic slope and friction coefficient of the terrain; and dividing the terrain region into regular grids according to the terrain extent and accuracy requirements. The size and density of the grids affect the accuracy of the final model; the denser the grid, the higher the model accuracy, but the greater the computational cost. For each grid point, the height value of that point is calculated based on the basic slope information in the terrain feature vectors, combined with the principles of a digital elevation model (DEM). The basic slope reflects the basic undulation trend of the terrain, and the height variation of different locations relative to the reference surface can be deduced from the slope. Simultaneously, the influence of the friction coefficient on local terrain variations is considered, and the height value is fine-tuned. For example, in areas with a high friction coefficient, the terrain may be relatively more rugged, and the fluctuation of the height value is appropriately increased.

[0054] By connecting the height values ​​of all grid points, a continuous three-dimensional surface is formed, thus constructing a three-dimensional terrain model. Methods such as triangulation can be used to connect adjacent grid points into triangular patches, and these triangular patches combine to form a complete terrain surface.

[0055] The various parameters in the terrain feature vector play different roles in the generation of the 3D terrain model. The base slope is mainly used to determine the general direction and undulation of the terrain, which determines the basic calculation framework for the height of grid points. The friction coefficient, on the other hand, acts as a correction factor, locally adjusting the height value calculated based on the base slope to more accurately reflect the actual characteristics of the terrain. For example, when calculating the height of a grid point, a preliminary height is first calculated based on the base slope and the point's position in the grid. Then, this preliminary height is corrected based on the friction coefficient of the surrounding area to obtain the final height value.

[0056] In some embodiments, extracting terrain contour features includes: rendering the generated 3D terrain model to obtain a 2D terrain image. Then, preprocessing operations such as denoising and smoothing are performed on the image to remove noise and interference information, making the terrain contour clearer. For example, a Gaussian filter is used to smooth the image, reducing high-frequency noise. Edge detection algorithms, such as the Canny edge detection algorithm, are used to detect edge information in the terrain image. Edges are important features of terrain contours; detecting edges can identify the boundaries between different regions in the terrain. The Canny edge detection algorithm calculates the gradient of the image and then determines edge points based on the magnitude and direction of the gradient, effectively detecting edges in the image. The detected edges are connected and organized to extract the complete terrain contour. A contour tracking algorithm can be used, starting from an edge point in the image, sequentially tracking adjacent edge points according to the edge direction until returning to the starting point, forming a closed contour. The extracted terrain contour is simplified, removing some small, unimportant contour details while retaining the main contour features. For example, the Douglas-Puk algorithm can be used to simplify the contour. This algorithm reduces the number of contour points by setting a threshold and deleting intermediate points when the deviation between contour points is less than the threshold, while preserving the main shape of the contour.

[0057] Image preprocessing aims to improve the accuracy of subsequent edge detection and contour extraction. Denoising eliminates random noise in the image, preventing it from being mistaken for edge information; smoothing smooths edges, reducing jaggedness and facilitating contour extraction. Edge detection is a crucial step in extracting terrain contours. The Canny edge detection algorithm boasts excellent edge detection performance, accurately detecting edges while suppressing noise. It identifies potential edge points by calculating image gradients, finding points with dramatic gray-level changes. Contour extraction connects the detected edge points into meaningful contours. Contour tracking algorithms connect scattered edge points according to edge continuity, forming complete terrain contours.

[0058] In some embodiments, a safe flight database is pre-established, including: collecting a large amount of historical flight data, including flight trajectories, terrain features, flight parameters (such as rotor speed, airspeed, flight altitude, etc.), and meteorological conditions. The collected historical flight data is labeled to determine whether each flight trajectory is a safe trajectory. Labeling can be based on criteria such as whether an accident occurred during the flight or whether flight rules were violated. Simultaneously, the terrain feature vector and flight parameters corresponding to each safe trajectory are recorded. The structure of the safe flight database is designed, including table design and field definition. Multiple tables can be created, such as a flight trajectory table, a terrain feature table, and a flight parameter table, linked by foreign keys to store and manage various data. The labeled historical flight data and related feature information are stored in the safe flight database. Relational databases (such as MySQL) or non-relational databases (such as MongoDB) can be used for storage; the appropriate database type is selected based on the characteristics and requirements of the data.

[0059] For each safe flight path, a corresponding safety boundary is determined based on terrain features and flight parameters. The safety boundary can be defined as a region within a certain radius around the flight path where flight is safe. Geometric methods, such as calculating the safety distance based on terrain elevation and flight altitude, can be used to determine the shape and extent of the safety boundary. The safety boundary information is also stored in a database and associated with the corresponding safe flight path.

[0060] In some embodiments, retrieving historical safe trajectories with a similarity greater than 90% to terrain contour features from a safe flight database includes: reading historical terrain contour features from the safe flight database and calculating the similarity with the extracted terrain contour features; iterating through all similarity calculation results; and selecting historical terrain contour features with a similarity greater than 90%, their corresponding historical safe trajectories, and historical safe boundaries. If multiple trajectories meet the conditions, further selection is made based on the similarity of flight parameters and meteorological conditions to identify the best-matching historical safe trajectory and historical safe boundary. For example, using cosine similarity or Euclidean distance, the similarity between the current terrain contour feature vector and each historical terrain contour feature vector is calculated. During comparison, these historical feature vectors undergo the same preprocessing and dimensionality reduction as the current feature vector to ensure they are compared in the same feature space.

[0061] In some embodiments, if a match is successful, the corresponding safety boundary is invoked; if a match fails, the safety boundary is reconstructed; after matching is completed, the final safety boundary is output as the alarm boundary. Reconstructing the safety boundary includes: analyzing the characteristics of the current flight environment based on the corrected terrain feature vector and real-time flight parameters (rotor speed, airspeed, flight altitude, etc.). For example, the complexity of the terrain is determined based on the base slope and friction coefficient, and the flight state and performance are determined based on the flight parameters. Combining terrain features and flight parameters, the safe distance for flight under the current terrain conditions is calculated. The safe distance should consider factors such as terrain elevation changes, obstacle distribution, and aircraft performance. For example, in areas with rugged terrain, the safe distance needs to be increased to avoid collisions; at higher flight speeds, the safe distance also needs to be appropriately increased to ensure sufficient reaction time.

[0062] The shape of the safety boundary is determined based on the safety distance and terrain contour features. The safety boundary can be a polygon, a circle, or other suitable shape, capable of encompassing the flight path and ensuring flight safety. For example, if the terrain contour is relatively regular, a circle or rectangle can be used as the safety boundary; if the terrain contour is complex, an irregular polygon can be used based on the shape of the contour. The determined safety boundary is then optimized by removing unreasonable parts to make the boundary smoother and more reasonable. Methods such as curve fitting can be used to smooth the boundary, while considering the continuity and feasibility of the flight path to ensure that the aircraft can fly safely within the safety boundary.

[0063] The reconstructed safety boundaries are verified through methods such as flight simulation or expert evaluation. The verified safety boundaries are then output as the final safety boundaries and added to the safe flight database. Flight simulation uses a flight simulator, inputting current terrain features, flight parameters, and the reconstructed safety boundaries to observe the aircraft's safety and feasibility during simulated flight. Expert evaluation involves inviting experienced pilots or flight experts to assess the safety boundaries and provide suggestions for improvement. Based on the verification results, the safety boundaries are further adjusted and refined.

[0064] In this embodiment, the dependence of traditional preset models on static environments is eliminated, achieving dynamic compensation and boundary reconstruction, thus solving the problem that they cannot adapt to dynamic environmental changes in unknown or complex terrains; the images are transformed into high-level semantic representations, solving the problem that images lack semantic information and are difficult to accurately identify terrain features; different terrain friction coefficients are obtained, providing basic information for subsequent steps to assess the landing or gliding safety of the aircraft, thus solving the problem of not being able to accurately obtain different terrain friction coefficients.

[0065] Accurately acquiring terrain height information and basic slope provides data support for terrain feature vector generation, solving the problem of difficulty in accurately acquiring terrain height and undulation; comprehensively considering terrain friction characteristics and undulation, it provides a more comprehensive description of terrain features, solving the problem that a single parameter cannot accurately describe terrain features; it more accurately reflects the terrain features of the aircraft in the current downwash environment, providing a reliable basis for flight safety decisions, solving the problem that the downwash affects terrain features and causes inaccurate measurements.

[0066] It intuitively presents terrain features, providing a foundation for extracting landform contour features and solving the problem that two-dimensional terrain information cannot meet the needs of complex flight environments; it accurately extracts landform contour features, providing a basis for safe flight database retrieval and solving the problem of difficulty in accurately obtaining landform contour features; it provides data support for safe flight, facilitating the retrieval and matching of historical safe trajectories and safety boundaries, solving the problem of difficulty in determining safety boundaries due to the lack of historical flight data reference; and it finally outputs the safety boundary as the alarm boundary to ensure flight safety, solving the problem of not being able to dynamically determine the safety boundary based on real-time terrain and flight conditions.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A helicopter ground proximity warning method, characterized in that, include: S1: Determine the flight status based on the wheel-mounted mechanism, acquire flight data in real time and compare it with the pre-set alarm boundaries to determine the alarm mode, and adjust the stability augmentation coefficient in real time according to the current helicopter status and alarm boundaries, thereby generating a control strategy; S2: Determine the transition mechanism based on the control strategy and core related parameters, and determine the solution based on the transition mechanism to eliminate the alarm; The wheel-mounted mechanism performs logical operations on the discrete signals of the left, right, and front wheel-mounted sensors to determine the helicopter's flight status. The flight status includes both ground status and air status. If any wheel-mounted sensor outputs an "air" signal, the entire aircraft is determined to be in an air status. The core associated parameters include radio altitude value, ground stabilization output value, air control target value, and radio altitude integral abrupt change. The standard deviation is calculated based on the ground stabilization output value and the air control target value. The standard deviation is used to quantify the switching amplitude and provide a transition baseline. The transition mechanism includes the takeoff process and the landing process. The alarm modes include alarm mode one: descent rate greater than alarm boundary; alarm mode two: terrain approach rate greater than alarm boundary; the alarm boundary is a set of critical safety data values ​​corresponding to various safety indicators; the stability enhancement coefficient is a key parameter in the flight control law, used to adjust the feedback gain; The descent rate refers to the risk of a helicopter crashing due to its high vertical speed, while the terrain approach rate refers to the risk of a helicopter hitting the ground due to changes in terrain slope during low-altitude flight. The core correlation parameters are key dynamic variables that drive bidirectional smooth switching, determining the transition ratio and switching sequence; the ground stabilization output is calculated in real time based on the wheel-mounted ground state and stabilization coefficient, and the air control target value is calculated based on the wheel-mounted air state and dynamic stabilization coefficient. The transition ratio is a dynamic weighting coefficient used to smoothly transition the control command from the current mode to the target mode. It is used to quantify the degree of mixing between the two modes in the current control command, and its value range is [0%, 100%]. The switching timing is the key time node control of the state transition process to ensure that the change of the transition ratio matches the helicopter dynamic response. For the takeoff process: a transition phase is set based on the standard deviation, which includes two phases: rapid integration and altitude decay. The rapid integration phase involves uniformly superimposing 50% of the standard deviation onto the air control output through an integrator within 0.2 seconds. The altitude decay phase refers to attenuating the remaining standard deviation according to altitude, with a predefined decay function used to calculate the real-time transition amount. For the landing process: a pre-buffering mechanism is implemented, in which 10% of the air control output value is added to the ground stabilization command in advance, and the addition ratio is increased linearly as the altitude decreases.

2. The helicopter ground proximity warning method as described in claim 1, characterized in that, The method further includes: acquiring terrain features based on a visual sensor, generating a fused image, and outputting pixel-level semantic labels; assigning friction coefficients according to the semantic labels, generating a digital elevation model of the terrain according to a point cloud registration algorithm, and calculating the base slope; fusing the friction coefficients and the base slope to generate a terrain feature vector; acquiring real-time flight parameters and calculating the downwash velocity according to the base slope, and correcting the terrain feature vector according to the downwash velocity; the real-time flight parameters include rotor speed, airspeed, and flight altitude.

3. The helicopter ground proximity warning method as described in claim 2, characterized in that, A 3D terrain model is regenerated based on the corrected terrain feature vector, and the terrain contour features are extracted. Historical safe trajectories with a similarity greater than 90% to the terrain contour features are retrieved from the pre-established safe flight database. If a match is successful, the corresponding safe boundary is invoked. If a match fails, the safe boundary is reconstructed. The final safe boundary output after the matching is completed is used as the alarm boundary.

4. A helicopter ground proximity warning method as described in claim 2, characterized in that, The method for acquiring terrain features based on visual sensors and generating a fused image includes: the visual sensors including a visible light camera, an infrared thermal imager, and a millimeter-wave radar; simultaneously scanning the terrain ahead using the visible light camera and the infrared thermal imager to generate an infrared image; scanning the terrain at a 10-degree cone angle using the millimeter-wave radar to output point cloud data; and fusing the infrared image and the point cloud image to obtain a fused image.

5. A helicopter ground proximity warning method as described in claim 2, characterized in that, The pixel-level semantic label refers to assigning a specific category label to each pixel in the image, representing terrain features; The friction coefficient is assigned based on semantic tags, including: pre-determining the range of terrain friction coefficients corresponding to each semantic tag according to the physical characteristics of different terrain types; acquiring pixel-level semantic tag images obtained after semantic segmentation, traversing the image pixels, searching for the corresponding friction coefficient value from the mapping table according to its semantic tag; assigning the found friction coefficient value to the pixel, and finally obtaining an image of the same size as the semantic tag image, where each pixel stores the corresponding friction coefficient, i.e., a friction coefficient distribution map, and assigning the friction coefficient.

6. A helicopter ground proximity warning method as described in claim 2, characterized in that, The friction coefficient and the base slope are fused to generate a terrain feature vector, including: associating the friction coefficient image and the base slope image with the same spatial location, and combining the friction coefficient value and the base slope value at each location into a vector as the terrain feature vector at that location; The downwash velocity refers to the speed of the downward airflow generated below the rotor during the flight of the aircraft, which is used to reflect the degree of influence of the aircraft on the airflow environment of the terrain below. Based on the influence of downwash flow on terrain friction coefficient and base slope, a single-factor correction model is pre-established. Based on the established single-factor correction models, the terrain feature vector of each pixel in the terrain feature vector map is corrected and recombined into a corrected terrain feature vector.

7. A helicopter ground proximity warning method as described in claim 3, characterized in that, The process of regenerating a 3D terrain model based on the corrected terrain feature vectors includes: collecting the corrected terrain feature vectors; dividing the terrain region into regular grids according to the terrain range and accuracy requirements; calculating the height value for each grid point based on the basic slope information in the terrain feature vectors; and connecting the height values ​​of all grid points to form a continuous 3D surface, thereby constructing a 3D terrain model. The safe flight database includes historical terrain features, historical safe trajectories, and historical safe boundaries.

Images