Unmanned sailboat adaptive environment perception method and perception system

Abstract

The application discloses an unmanned sailboat self-adaptive environment sensing method and a sensing system, and relates to the technical field of unmanned sailboats.The application realizes the leap from passive geometric sensing to active cognition based on physical laws through physical trigger sampling, sea wave same-frequency denoising, true wind field reconstruction and active smoothing interaction, effectively solves the problem of sensing failure of the unmanned sailboat in severe sea conditions, and can guarantee stable sailing of the unmanned sailboat.

Description

Technical Field

[0001] This invention relates to the field of unmanned sailboat technology, and in particular to an adaptive environment perception method and perception system for unmanned sailboats. Background Technology

[0002] Currently, environmental perception technology for unmanned autonomous vessels, especially unmanned sailboats (USV-S), is receiving widespread attention.

[0003] With its advantages such as sustainable energy and unlimited endurance, unmanned sailing vessels have shown great potential in long-endurance missions such as oceanographic research, meteorological and hydrological monitoring, and anti-submarine reconnaissance. However, compared with traditional diesel-electric propulsion motorized unmanned vessels, unmanned sailing vessels rely entirely on wind power. Since traditional unmanned sailing vessels usually do not have active roll reduction devices, they are affected by large hull sway and power source fluctuations during actual navigation, which affects their stable navigation. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned defects in the prior art and provide an adaptive environment perception method and perception system for unmanned sailboats, which can solve the problem of perception failure of unmanned sailboats in severe sea conditions and ensure that unmanned sailboats can sail stably.

[0005] To achieve the above objectives, the present invention provides an adaptive environment perception method for unmanned sailboats, which includes the following steps:

[0006] S1: Real-time monitoring and acquisition of data from the ship's inertial measurement unit. A synchronous trigger signal is generated only when the hull motion stability of the unmanned sailboat meets the preset quasi-static window threshold to activate the lidar and visual sensor for discrete sampling. The target is obtained through the lidar and the original image is obtained through the visual sensor.

[0007] S2: Construct a virtual stable window based on the real-time gravity vector, reverse rotate and dynamically lock the region of interest in the original image output by the visual sensor, and output image data with the sea-line horizontally centered;

[0008] S3: Calculate the vertical motion frequency of the target and filter out the wave interference point cloud that is strongly correlated with the heave and pitch frequencies of the ship.

[0009] S4: Establish a rigid body motion model of the mast to eliminate the sway-induced wind speed in the anemometer measurement, obtain the ground velocity vector and heading information through the global navigation satellite system, fuse them to generate meteorological data containing the true wind field vector and gust forward prediction, and adaptively adjust the fusion weight of each sensor data according to the ship's heel angle at the time of collection.

[0010] S5: Based on the processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system, spatiotemporal registration is performed to construct a wave potential energy field map that includes wave height and impact energy distribution;

[0011] S6: Spatial mapping and overlay of the wind field distribution map determined by wind field dynamics with the physical safety boundary determined by obstacles and high-energy ocean waves to generate an environmental grid map or topology map representing the dynamic joint navigable domain, which is then sent as output data to the external planning system.

[0012] S7: Real-time assessment of the integrity and confidence of the environmental model. When N consecutive frames of sensing data are found to be unreliable or the environment is too harsh to cause the loss of positioning features, a stabilization observation request signal is generated and sent to the external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, the observation is reset.

[0013] Further, S1: Real-time monitoring and acquisition of data from the ship's inertial measurement unit, generating a synchronous trigger signal only when the ship's motion stability meets a preset quasi-static window threshold to activate the lidar and vision sensor for discrete sampling, obtaining the detection target through the lidar, and obtaining the original image through the vision sensor, including the following steps:

[0014] The angular velocity data output by the ship's inertial measurement unit is monitored in real time to obtain the ship's real-time roll angular velocity and real-time pitch angular velocity.

[0015] Set the preset quasi-static window threshold to obtain the preset roll quasi-static threshold and pitch quasi-static threshold;

[0016] When the real-time roll rate of the hull is less than the roll quasi-static threshold and the real-time pitch rate of the hull is less than the pitch quasi-static threshold, the synchronization trigger signal is generated to activate the lidar and the vision sensor to perform discrete sampling, thereby obtaining the detected target and the original image.

[0017] Furthermore, the angular velocity data output by the ship's inertial measurement unit is used to generate a trigger signal through logical operations, triggering the logic... This is determined by the following system of inequalities:

[0018] ;

[0019] in, for The state of the trigger signal at any given moment. =1 triggers data collection. =0 indicates standby mode; This refers to the ship's real-time roll rate. This refers to the ship's real-time pitch rate. The preset quasi-static threshold for roll is... This is the preset pitch quasi-static threshold.

[0020] Further, S2: Constructing a virtual stable viewport based on the real-time gravity vector, performing reverse rotation and dynamic region of interest locking on the original image output by the visual sensor, and outputting image data with the sea-line horizontally centered, includes the following steps:

[0021] The ship's pitch angle and roll angle are obtained by real-time monitoring of the ship's inertial measurement unit.

[0022] Obtain the initial coordinates of the optical center of the vision sensor and the equivalent focal length of the camera;

[0023] Based on the ship's pitch angle, roll angle, initial coordinates of the optical center of the vision sensor, and equivalent focal length of the camera, a reverse compensation matrix is ​​constructed. Geometric correction is performed on the original image to obtain the center ordinate of the region of interest. The initial ordinate of the optical center of the vision sensor is then moved to the center ordinate of the region of interest to obtain the image data.

[0024] Further, S3: Calculate the vertical motion frequency of the detected target and filter out wave interference point clouds that are strongly correlated with the ship's heave and pitch frequencies, including the following steps:

[0025] By comparing the vertical displacement frequency of the target object clustering in the point cloud data obtained by the lidar with the heave frequency of the ship's hull, the correlation coefficient is obtained.

[0026] When the correlation coefficient is greater than or equal to the preset denoising threshold, the point cloud of the target object cluster is filtered out.

[0027] When the correlation coefficient is less than the preset denoising threshold, the target object is determined to be clustered as an obstacle.

[0028] Furthermore, S4: Establish a rigid body motion model of the mast to eliminate the sway-induced wind speed in the anemometer measurements, obtain the ground velocity vector and heading information through the Global Navigation Satellite System, and fuse them to generate meteorological data that includes the true wind field vector and forward gust prediction. At the same time, based on the ship's heel angle at the time of acquisition, adaptively adjust the fusion weights of the data from each sensor, including the following steps:

[0029] The apparent wind speed vector obtained by the anemometer, the three-axis angular velocity vector of the hull obtained by the hull inertial measurement unit, the displacement vector of the anemometer relative to the center of the hull obtained according to the height and position of the anemometer on the mast, and the hull velocity vector relative to the ground obtained by the global navigation satellite system.

[0030] Treating the mast as a rigid lever, calculate the tangential linear velocity at the anemometer and perform vector subtraction to obtain the true wind field vector. Determined by the following vector equation:

[0031] ;

[0032] in, This is the apparent wind speed vector of the anemometer; The vector of angular velocity along the three axes of the hull; This is the displacement vector of the anemometer relative to the center of the ship's hull; Represents the vector cross product operation; The hull-to-ground velocity vector provided for global navigation satellite systems.

[0033] Furthermore, it also includes the following steps: configuring a directional sensor covering the port side field of view and a directional sensor covering the starboard side field of view, as well as a lidar with omnidirectional detection capability;

[0034] The critical angle threshold at which the directional sensor's field of view is blocked by waves or diffused into the sky is preset.

[0035] The fusion weights of the directional sensors are dynamically adjusted based on the ship's heel angle. Determined by the following S-shaped function formula:

[0036] ;

[0037] in, This is the current hull heel angle; The critical angle threshold at which the field of view of the directional sensor is obstructed by sea waves or diffused into the sky; The gain coefficient used to adjust the weight decay rate;

[0038] When the current port roll angle of the hull is greater than the preset critical angle threshold, the sensor weight of the directional sensor covering the port field of view is less than the sensor weight of the directional sensor covering the starboard field of view.

[0039] When the current starboard heel angle is greater than the preset critical angle threshold, the sensor weight of the directional sensor covering the port side field of view is greater than the sensor weight of the directional sensor covering the starboard side field of view.

[0040] Furthermore, S5: Based on the processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system, spatiotemporal registration is performed to construct a wave potential energy field map containing wave height and impact energy distribution, including the following steps:

[0041] The wave energy mapping principle based on gravitational potential energy field is adopted to quantify the potential threat of ocean waves to sailing, and the geometric elevation of ocean waves is converted into energy distribution to construct the ocean wave potential energy field.

[0042] The wave potential energy density of the ocean wave potential energy field Determined by the following physical formula:

[0043] ;

[0044] in, The density of seawater; It is the acceleration due to gravity; The reconstructed wave surface for the sensing system The height at the coordinates; The mean static sea level is the height of the sea.

[0045] Furthermore, S7: Real-time assessment of the integrity and confidence of the environmental model. When it detects that N consecutive frames of sensing data are unreliable or the environment is too harsh to cause the loss of positioning features, it generates a stabilization observation request signal and sends it to the external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, it triggers a reset observation, which includes the following steps:

[0046] A sensing lockout determination principle based on multi-dimensional feature consistency is adopted to achieve bidirectional interaction between the sensing system and the control system;

[0047] By calculating the confidence level of environmental perception in real time An interaction request is triggered when the confidence level is insufficient; the confidence level is determined by the following weighted summation formula:

[0048] ;

[0049] in, This is the normalized value of the number of valid feature points extracted in the current frame. The residual value for matching maps from previous and next frames; For attitude stability scoring based on hull inertial measurement unit data; , , These are the weighting coefficients;

[0050] when Less than the safety threshold ,and Less than The duration exceeded At that time, a stabilization request is triggered, requesting the ship to perform roll reduction or stop against the wind.

[0051] when Greater than the safety threshold At that time, the sensing system re-initializes its positioning, and triggers a reset of observation after the ship stabilizes.

[0052] According to a second aspect of the present invention, an adaptive environmental perception system for an unmanned sailboat includes a physical trigger acquisition module, a multi-source data processing and fusion module, a wave and wind field environment modeling module, and a perception state decision module. The perception system outputs environmental situation information and interaction requests to an external navigation controller. The physical trigger acquisition module acquires raw environmental data, integrates attitude-based discrete sampling and image preprocessing mechanisms, and can monitor the data from the ship's inertial measurement unit in real time. It generates a synchronous trigger signal only when the ship's motion stability meets a preset quasi-static window threshold to activate the lidar and visual sensor for discrete sampling. It constructs a virtual stable window based on the real-time gravity vector, reverses the rotation of the raw image output by the visual sensor, and dynamically locks the region of interest, outputting image data with the sea-line horizontally centered. The multi-source data processing and fusion module cleans and extracts features from the acquired data, calculates the vertical motion frequency of the detected target, filters out wave interference point clouds strongly correlated with the ship's heave and pitch frequencies, establishes a mast rigid body motion model to eliminate sway-induced wind speeds from anemometer measurements, and combines the ground velocity provided by the global navigation satellite system. Vector and heading information are fused to generate meteorological data containing true wind field vectors and forward gust predictions. Simultaneously, the fusion weights of each sensor data are adaptively adjusted based on the ship's heel angle at the time of data acquisition. The wave and wind field environment modeling module generates an environmental situation map and can perform spatiotemporal registration based on processed lidar data, ship inertial measurement unit data, GNSS data, and absolute positioning coordinates provided by the GNSS system to construct a wave potential energy field map containing wave height and impact energy distribution. It also integrates the wind field distribution map determined by wind dynamics with the physical distribution map determined by obstacles and high-energy waves. The safety boundary is spatially mapped and overlaid to generate an environmental grid map or topology map representing the dynamic joint navigable domain, which is sent as output data to the external planning system. The perception state decision module is used to monitor the operating status and confidence level of the perception system itself, and has the function of active perception interaction. It can evaluate the integrity and confidence level of the environmental model in real time. When it detects that the perception data is unreliable for N consecutive frames or the environment is too harsh to cause the loss of positioning features, it generates a stabilization observation request signal and sends it to the external navigation controller to request the ship to perform roll reduction or stop in the wind. After the ship stabilizes, it triggers a reset observation.

[0053] Compared with existing technologies, this invention has the following advantages: The inertial measurement unit (IMU) data is used to acquire raw environmental data. Addressing the severe swaying characteristics of unmanned sailboats, an attitude-based discrete sampling mechanism is employed: The IMU data is monitored in real time, and a synchronous trigger signal is generated only when the hull motion stability of the unmanned sailboat meets a preset quasi-static window threshold. This activates the lidar and visual sensors for instantaneous data acquisition, eliminating motion blur from its physical source. Simultaneously, a virtual stable window is constructed using real-time gravity vectors to reverse the original image and dynamically lock the region of interest, ensuring that the sea-line in the output image remains horizontally centered, thus improving the efficiency of subsequent algorithms. The vertical motion frequency of the detected target is calculated, filtering out wave interference point clouds that are strongly correlated with the hull's heave and pitch frequencies. A rigid body motion model of the mast is established for wind field perception, and the velocity vector from the global navigation satellite system is used to eliminate sway-induced wind speeds from anemometer measurements, generating meteorological data containing true wind field vectors and gust forward predictions. Furthermore, based on… The system collects the ship's heel angle at the moment of acquisition and adaptively adjusts the fusion weights of each sensor to automatically shield visual sensor data with limited field of view at large heel angles. It then constructs a wave potential energy field map containing wave height and impact energy distribution. By quantifying wave height and impact energy distribution, it can identify high-energy wave regions. Subsequently, it spatially maps and overlays the wind field distribution map determined by wind dynamics with the physical safety boundaries determined by obstacles and high-energy waves to generate an environmental grid map or topology map representing the dynamic joint navigable domain. This map is then sent to the external planning system as the final perception output data. The system evaluates the integrity and confidence of the environmental model in real time. When N consecutive frames of perception data are detected as unreliable, the module generates a stabilization observation request signal and sends it to the external navigation controller, requesting the ship to perform roll reduction or stop against the wind. Once the ship's attitude returns to stability, the observation is reset. This achieves an active survival strategy of control for perception, reducing the impact of large-scale rolling and power source fluctuations on the ship during actual navigation. Attached Figure Description

[0054] To more clearly illustrate the technology in the embodiments of the present invention or 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating the adaptive environment perception method for unmanned sailboats according to the present invention.

[0056] Figure 2 This is a schematic diagram of the workflow of the unmanned sailboat adaptive environment perception system of the present invention;

[0057] Figure 3 This is another flowchart illustrating the adaptive environment perception method for unmanned sailboats according to the present invention. Detailed Implementation

[0058] The technology of this embodiment of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiment is one embodiment of the present invention, and not all embodiments thereof. Based on this embodiment 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.

[0059] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0060] Furthermore, if the embodiments of the present invention involve descriptions such as "first" or "second", such descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0061] Please see Figures 1 to 3 This invention provides an adaptive environment perception method and perception system for unmanned sailboats.

[0062] Please see Figures 1 to 3 The adaptive environment perception method for unmanned sailboats according to embodiments of the present invention includes the following steps:

[0063] S1: Real-time monitoring and acquisition of data from the ship's inertial measurement unit. A synchronous trigger signal is generated only when the hull motion stability of the unmanned sailboat meets the preset quasi-static window threshold to activate the lidar and visual sensor for discrete sampling. The lidar is used to obtain the detection target, and the visual sensor is used to obtain the original image.

[0064] S2: Construct a virtual stable window based on the real-time gravity vector, reverse rotate and dynamically lock the region of interest in the original image output by the visual sensor, and output image data with the sea-line horizontally centered.

[0065] S3: Calculate the vertical motion frequency of the target and filter out the wave interference point cloud that is strongly correlated with the heave and pitch frequencies of the ship.

[0066] S4: Establish a rigid body motion model of the mast to eliminate the sway-induced wind speed in the anemometer measurement, obtain the ground velocity vector and heading information through the global navigation satellite system, fuse them to generate meteorological data containing the true wind field vector and gust forward prediction, and adaptively adjust the fusion weight of each sensor data according to the ship's heel angle at the time of collection.

[0067] S5: Based on the processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system, spatiotemporal registration is performed to construct a wave potential energy field map that includes wave height and impact energy distribution;

[0068] S6: Spatial mapping and overlay of the wind field distribution map determined by wind field dynamics with the physical safety boundary determined by obstacles and high-energy ocean waves to generate an environmental grid map or topology map representing the dynamic joint navigable domain, which is then sent as output data to the external planning system.

[0069] S7: Real-time assessment of the integrity and confidence of the environmental model. When N consecutive frames of sensing data are found to be unreliable or the environment is too harsh to cause the loss of positioning features, a stabilization observation request signal is generated and sent to the external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, the observation is reset.

[0070] Data from the ship's inertial measurement unit (IMU) is used to acquire raw environmental data. Considering the severe swaying of the unmanned sailboat, an attitude-based discrete sampling mechanism is employed: IMU data is monitored in real time, and a synchronous trigger signal is generated only when the sailboat's hull motion stability meets a preset quasi-static window threshold to activate the lidar and visual sensors for instantaneous data acquisition, eliminating motion blur at its physical source. Simultaneously, a virtual stabilization window is constructed using real-time gravity vectors, and the original image is reverse-rotated and dynamically locked to ensure the sea-line in the output image remains horizontally centered, improving the efficiency of subsequent algorithms. The vertical motion frequency of the detected target is calculated, and wave interference point clouds strongly correlated with the ship's heave and pitch frequencies are filtered out. A rigid body motion model of the mast is established for wind field perception, and the velocity vector from the global navigation satellite system is used to eliminate sway-induced wind speeds from anemometer measurements, generating meteorological data containing true wind field vectors and gust forward predictions. Furthermore, based on the ship's heel at the time of data acquisition... The system adaptively adjusts the fusion weights of various sensors to automatically shield visual sensor data with limited field of view at large tilt angles, constructing a wave potential energy field map that includes wave height and impact energy distribution. By quantifying wave height and impact energy distribution, high-energy wave regions can be identified. Subsequently, the wind field distribution map determined by wind dynamics is spatially mapped and superimposed with the physical safety boundaries determined by obstacles and high-energy waves to generate an environmental grid map or topology map representing the dynamic joint navigable domain. This map is then sent to the external planning system as the final perception output data. The system also evaluates the integrity and confidence of the environmental model in real time. When N consecutive frames of perception data are detected as unreliable, the module generates a stabilization observation request signal and sends it to the external navigation controller, requesting the ship to perform roll reduction or stop against the wind. Once the ship's attitude returns to stability, the observation is reset, thus realizing an active survival strategy of control for perception, reducing the impact of large-scale rolling and power source fluctuations on the ship during actual navigation.

[0071] This invention achieves a leap from passive geometric perception to active cognition based on physical laws through physical trigger sampling, wave frequency noise reduction, real wind field reconstruction, and active stabilization interaction. It effectively solves the problem of perception failure of unmanned sailboats in severe sea conditions and can ensure that unmanned sailboats can sail stably.

[0072] Specifically, the target being detected is the target in the point cloud data generated by the lidar.

[0073] In some embodiments of the present invention, S1: Real-time monitoring and acquisition of data from the ship's inertial measurement unit, generating a synchronous trigger signal only when the ship's motion stability meets a preset quasi-static window threshold to activate the lidar and visual sensor for discrete sampling, obtaining the detection target through the lidar, and obtaining the original image through the visual sensor, includes the following steps:

[0074] The angular velocity data output by the ship's inertial measurement unit is monitored in real time to obtain the ship's real-time roll angular velocity and real-time pitch angular velocity.

[0075] Set a preset quasi-static window threshold to obtain preset roll quasi-static thresholds and pitch quasi-static thresholds;

[0076] When the ship's real-time roll rate is less than the roll quasi-static threshold and the ship's real-time pitch rate is less than the pitch quasi-static threshold, a synchronous trigger signal is generated to activate the lidar and vision sensor to perform discrete sampling and obtain the detected target and the original image.

[0077] A quasi-static window determination principle based on time-domain inertial threshold is adopted to accurately identify brief moments of stability in the hull during violent swaying.

[0078] The angular velocity data output by the ship's inertial measurement unit is used to generate a trigger signal through logical operations, triggering the logic... This is determined by the following system of inequalities:

[0079] ;

[0080] in, for The state of the trigger signal at any given moment. =1 triggers data collection. =0 indicates standby mode; This refers to the ship's real-time roll rate. This refers to the ship's real-time pitch rate. The preset quasi-static threshold for roll is... This is the preset pitch quasi-static threshold.

[0081] By monitoring the angular velocity data output by the ship's inertial measurement unit in real time, trigger signals are generated through logical operations. The roll quasi-static threshold and pitch quasi-static threshold are dynamically set based on the inherent oscillation period of the unmanned sailboat's hull. For example, the roll angular velocity threshold is set... =0.2 rad / s, pitch angular velocity threshold =0.1rad / s. When the unmanned sailboat experiences weightlessness after falling below a wave crest, and the inertial measurement unit measures its real-time roll rate as 0.15rad / s and pitch rate as 0.05rad / s, the logic is triggered. The conditions of the system of inequalities are simultaneously satisfied, and the system output is... =1, immediately triggering the lidar to emit a pulse; however, if the angular velocity exceeds the limit at any other time, then... =0.

[0082] By introducing this inertial threshold determination principle, the system achieves physical isolation of high dynamic motion noise at the hardware source, ensuring that the lidar and camera only work at the moment when the ship is relatively stationary, fundamentally eliminating motion blur and significantly reducing the average power consumption of the sensors.

[0083] In some embodiments of the present invention, S2: Constructing a virtual stable window based on the real-time gravity vector, performing reverse rotation and dynamic region of interest locking on the original image output by the visual sensor, and outputting image data with the sea-line horizontally centered, includes the following steps:

[0084] The ship's pitch and roll angles are obtained by real-time monitoring of the ship's inertial measurement unit.

[0085] Obtain the initial coordinates of the optical center of the vision sensor and the equivalent focal length of the camera;

[0086] Based on the ship's pitch angle, roll angle, initial coordinates of the optical center of the vision sensor, and equivalent focal length of the camera, a reverse compensation matrix is ​​constructed. Geometric correction is performed on the original image to obtain the center ordinate of the region of interest. The initial ordinate of the optical center of the vision sensor is then moved to the center ordinate of the region of interest to obtain image data.

[0087] This method employs affine transformation based on gravity vectors and dynamic window locking to eliminate the interference of ship attitude changes on image field stability. This principle utilizes real-time attitude angles to construct an inverse compensation matrix, performing geometric correction on the original image, and determining the center ordinate of the region of interest. Determined by the following trigonometric function relationships:

[0088] ;

[0089] in, The initial ordinate of the optical center of the image sensor; This is the equivalent focal length of the camera; The pitch angle of the ship; This refers to the roll angle of the ship.

[0090] For example, the camera's vertical resolution center =512 pixels, focal length =1000 pixels; When an unmanned sailboat encounters waves that cause the bow to rise 10 degrees, the hull pitch angle at this time is... When the roll angle is 10 degrees, i.e., tan(10∘)≈0.176, and the roll angle is close to 0 degrees, the center ordinate of the region of interest is used as the reference. Calculation of trigonometric function relationships, =512+1000×0.176×1=688; At this point, the algorithm automatically moves the center of the region of interest cropping box down from 512 to 688 pixels, thereby offsetting the shift of the horizon line caused by the bow of the ship rising, and ensuring that the captured image is still centered on the horizon line.

[0091] By introducing this affine transformation and locking principle, the system achieves virtual mechanical stabilization of the image field of view relative to the horizon, ensuring that the horizon in the output image remains horizontal and centered, significantly reducing the computational overhead of subsequent neural network processing of invalid sky or deck areas.

[0092] In some embodiments of the present invention, S3: Calculating the vertical motion frequency of the detected target and filtering out wave interference point clouds that are strongly correlated with the heave and pitch frequencies of the ship includes the following steps:

[0093] By comparing the vertical displacement frequency of target object clustering with the heave frequency of the ship hull in the point cloud data obtained by lidar, the correlation coefficient is obtained.

[0094] When the correlation coefficient is greater than or equal to the preset denoising threshold, the point cloud of the target object cluster is filtered out.

[0095] When the correlation coefficient is less than the preset denoising threshold, the target object is determined to be clustered as an obstacle.

[0096] A wave dynamics filtering principle based on frequency domain spectral correlation is employed to distinguish between drifting false targets and rigid obstacles. This principle compares the vertical displacement frequency of target clusters with the heave frequency of the ship's hull in point cloud data obtained from lidar, and determines target attributes by calculating the correlation coefficient. Determined using the following statistical formula:

[0097] ;

[0098] in, The centroid of the point cloud is at the 1st The vertical height of the frame; The height of the ship's heave during the same period; and These are the means of the two, respectively; This is the length of the sampling window.

[0099] For example, in a section Within a sampling window of 50 frames, the unmanned sailboat experiences periodic rises and falls of 1 meter with the swell. Simultaneously, the lidar detects a point cloud 20 meters ahead that is also experiencing a synchronous rise and fall of 1 meter. Since the motion phases of the two are highly synchronized, the correlation coefficient calculated using the formula is... If the value approaches 1.0, which is greater than the preset denoising threshold of 0.85, the system determines that the point cloud is a wave that rises and falls with the waves, rather than a stationary reef or ship, and thus filters it out.

[0100] By introducing this principle of frequency domain spectral correlation, the system achieves accurate elimination of false wave alarms, only when... It only identifies obstacles when the noise level is below a preset threshold, effectively solving the problem of low-profile unmanned sailboats misjudging swells as obstacles in rough sea conditions.

[0101] In some embodiments of the present invention, S4: A rigid body motion model of the mast is established to eliminate the sway-induced wind speed in the anemometer measurement; the ground velocity vector and heading information are obtained through the Global Navigation Satellite System; meteorological data including the true wind field vector and gust forward prediction are fused to generate meteorological data; and the fusion weights of the data from each sensor are adaptively adjusted according to the ship's heel angle at the time of data collection.

[0102] It includes the following steps:

[0103] The apparent wind speed vector obtained by the anemometer, the three-axis angular velocity vector of the hull obtained by the hull inertial measurement unit, the displacement vector of the anemometer relative to the center of the hull obtained according to the height and position of the anemometer on the mast, and the hull velocity vector relative to the ground obtained by the global navigation satellite system.

[0104] Treating the mast as a rigid lever, calculate the tangential linear velocity at the anemometer and perform vector subtraction to obtain the true wind field vector. Determined by the following vector equation:

[0105] ;

[0106] in, This is the apparent wind speed vector of the anemometer; The vector of angular velocity along the three axes of the hull; This is the displacement vector of the anemometer relative to the center of the ship's hull; Represents the vector cross product operation; The hull-to-ground velocity vector provided for global navigation satellite systems.

[0107] A mast lever effect compensation principle based on rigid body kinematics is adopted to eliminate the induced error in wind speed measurement caused by hull sway. This principle treats the mast as a rigid lever, calculates the tangential linear velocity at the anemometer, and performs vector subtraction; where, This is the displacement vector of the anemometer relative to the center of the ship, i.e., the height and position of the mast;

[0108] For example, in an ideal, windless environment, The value should be 0. The unmanned sailboat is violently pitching forward due to the waves, causing a forward tangential velocity of 3 m / s at the top of the 5-meter mast. At this time, the anemometer will incorrectly measure a headwind of 3 m / s. The anemometer's apparent wind speed vector... The true wind field vector is 3. The vector equation is the angular velocity measured by the ship's inertial measurement unit. Multiply by mast height The induced velocity was calculated to be 3 m / s, and then subtracted from the formula, i.e., 3-3=0, thus successfully restoring the true wind speed to 0.

[0109] By introducing this leverage effect compensation principle, the system achieves accurate reconstruction of the real sea surface wind field under severe swaying conditions, providing distortion-free meteorological data support for the sail adjustment and course planning of unmanned sailboats.

[0110] In some embodiments of the present invention, the following steps are also included:

[0111] It is equipped with a directional sensor covering the port and starboard fields of view, as well as a lidar with omnidirectional detection capabilities;

[0112] The critical angle threshold at which the directional sensor's field of view is blocked by waves or diffused into the sky is preset.

[0113] The fusion weights of the directional sensors are dynamically adjusted based on the ship's heel angle. This is determined by the following S-shaped function relationship:

[0114] ;

[0115] in, This is the current hull heel angle; The critical angle threshold at which the field of view of a directional sensor is blocked by waves or diffused into the sky; The gain coefficient used to adjust the weight decay rate;

[0116] When the current port roll angle of the hull is greater than the preset critical angle threshold, the sensor weight of the directional sensor covering the port field of view is less than the sensor weight of the directional sensor covering the starboard field of view.

[0117] When the current starboard heel angle is greater than the preset critical angle threshold, the sensor weight of the directional sensor covering the port side field of view is greater than the sensor weight of the directional sensor covering the starboard side field of view.

[0118] Specifically, gain coefficient The parameters need to be set manually based on the sensor hardware configuration, carrier platform dynamics, and system stability.

[0119] The system employs an attitude confidence-based sigmoid adaptive weighting principle to address the limited field of view of a single sensor at large tilt angles. It is equipped with directional sensors, such as cameras, that cover the port and starboard fields of view, as well as sensors with omnidirectional detection capabilities, such as lidar.

[0120] For example, the system sets the critical angle of the effective field of view for the port side camera. =20.

[0121] When the unmanned sailboat's port heel reaches 30°, the hull heel angle is 30°. > At this point, the field of view of the directional sensor covering the port side is directed towards the sea surface waves, resulting in a sharp decrease in the amount of effective information. Substituting this into the sensor weights... The S-shaped function formula calculation shows that the exponent term −k(20−30) becomes a large positive number, causing a sharp increase in the denominator and affecting the weight of the directional sensor covering the port side field of view. The confidence level quickly approaches 0; at this point, the multi-source fusion algorithm automatically reduces the confidence level of the port side visual data and instead relies entirely on the directional sensor covering the starboard side field of view. The directional sensor covering the starboard side field of view is located at a high position with a wide field of view, and environmental modeling is performed using data from the omnidirectional lidar.

[0122] By introducing this adaptive weighting principle, the system achieves smooth switching of multimodal data. When the unmanned sailboat tilts at a large angle, causing the directional sensors in the port or starboard fields of view to look at the sky or be blocked by waves, the system automatically reduces their weights and increases the weights of the omnidirectional lidar, ensuring the robustness of all-weather perception.

[0123] In some embodiments of the present invention, S5: Spatiotemporal registration is performed based on the processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system to construct a wave potential energy field map containing wave height and impact energy distribution, including the following steps:

[0124] The wave energy mapping principle based on gravitational potential energy field is adopted to quantify the potential threat of ocean waves to unmanned sailboats, and the geometric elevation of ocean waves is converted into energy distribution to construct an ocean wave potential energy field.

[0125] Wave potential energy density of ocean wave potential energy field Determined by the following physical formula:

[0126] ;

[0127] in, The density of seawater; It is the acceleration due to gravity; The reconstructed wave surface for the sensing system The height at the coordinates; The mean static sea level is the height of the sea.

[0128] Specifically, wave potential energy density describes the gravitational potential energy of a unit area of ​​sea surface relative to the mean sea level. The wave potential energy field is a whole consisting of the continuous spatial distribution of wave potential energy density values ​​of all grids.

[0129] Specifically, It is the representation of calibrated and filtered lidar altimetry data on a geographic grid. The system uses the absolute position provided by the Global Navigation Satellite System as a reference and projects these filtered point clouds onto a rasterized horizontal plane. Specifically, for each grid cell on the horizontal plane, the average or highest z-coordinate of all nearby lidar points is taken as the estimated height of the wave surface at that location. .

[0130] It serves as a reference datum. The Global Navigation Satellite System (GNSS) can provide its own geodetic height. By using the ship's inertial measurement unit (IMU) and attitude data, the instantaneous sea level height at the point of contact between the ship and the waterline can be estimated. This is achieved by continuously recording the estimated instantaneous sea level height over a period of time during which the sea level remains stable, and then taking the average over that time. .

[0131] The wave energy mapping principle based on gravitational potential energy field is adopted to quantify the potential threat of ocean waves to unmanned sailboats. This principle transforms the geometric elevation of ocean waves into energy distribution and constructs the ocean wave potential energy field.

[0132] For example, a wave crest height was detected in the sea area ahead. For giant waves exceeding 3 meters above calm sea surface, substituting the wave potential energy density into the wave potential energy field... The physical formula for its potential energy density It is proportional to the square of the height difference, that is, 3 2 =9, compared to its height of only 0.5 meters, that is, 0.5 2 =0.2, this area is marked as a high-risk energy zone, i.e., a red obstacle; the path planning algorithm will guide the unmanned sailboat around this energy high wall, or find a saddle with a lower wave height to cut in.

[0133] By introducing this potential energy field mapping principle, the system realizes the digital representation of the "soft obstacle" of ocean waves. The generated potential energy map can guide unmanned sailboats to avoid high-energy wave peaks and find saddle areas with lower energy to "cut waves" and thus improve the stability and safety of navigation.

[0134] In some embodiments of the present invention, S7: Real-time evaluation of the integrity and confidence of the environmental model; when it is detected that N consecutive frames of sensing data are unreliable or the environment is too harsh to cause the loss of positioning features, a stabilization observation request signal is generated and sent to an external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, the observation is reset, including the following steps:

[0135] A sensing lockout determination principle based on multi-dimensional feature consistency is adopted to achieve bidirectional interaction between the sensing system and the control system;

[0136] By calculating the confidence level of environmental perception in real time An interaction request is triggered when the confidence level is insufficient; the confidence level is determined by the following weighted summation formula:

[0137] ;

[0138] in, This is the normalized value of the number of valid feature points extracted in the current frame. The residual value for matching maps from previous and next frames; For attitude stability scoring based on hull inertial measurement unit data; , , These are the weighting coefficients;

[0139] when Less than ,and Less than The duration exceeded At that time, a stabilization request is triggered, requesting the ship to perform roll reduction or stop against the wind.

[0140] when Greater than the safety threshold At that time, the sensing system re-initializes its positioning, and triggers a reset of observation after the ship stabilizes.

[0141] Specifically, This is the safety threshold.

[0142] A sensing lockout determination principle based on multi-dimensional feature consistency is adopted to achieve bidirectional interaction between the sensing system and the control system. This principle calculates the environmental perception confidence level in real time. An interaction request is triggered when the confidence level is insufficient.

[0143] < Duration exceeding At times, a stabilization request is triggered; for example, during a nighttime storm, insufficient lighting in the camera reduces the number of feature points. The value plummeted to 0.1, and the high waves further reduced the ship's inertial measurement unit score. The confidence level drops to 0.2, at which point the weighted total confidence level is calculated. The value is only 0.25, far below the safety threshold of 0.6, and the duration is greater than [missing value]. The sensing system determines that the road is no longer clearly visible, and immediately sends a "stabilization request" to the controller. Upon receiving the request, the controller executes a head-to maneuver, and once the vessel is relatively stationary, the sensing system re-initializes its positioning.

[0144] By introducing this sensing loss judgment principle, the system realizes an active survival mechanism of "control for sensing". Before the algorithm fails due to extreme environment, it actively requests the hull to enter the Head-to (stopping the ship against the wind) state, using physical stillness to restore sensing data, which greatly reduces the risk of capsizing.

[0145] It involves processing the current frame image using a feature extraction algorithm to identify significant features such as corners and edges in the image, and obtaining a series of normalized values ​​for the number of "feature points", which are between 0 and 1; The process involves registering the latest frame of lidar scan data (filtered and denoised) with the local map from the previous moment to find an optimal rigid body transformation (rotation + translation). Under this optimal transformation, the root mean square error of the distance between all matching point pairs is the value. The larger the value, the more it indicates that the current scan does not match the historical map, which may indicate that the location has drifted or the environment has changed drastically. By calculating statistical characteristics of inertial measurement unit (IMU) data in real time, such as the root mean square of angular velocity over a past period, the proportion exceeding the attitude stability threshold, and the severity of heave as reflected by the accelerometer, a predefined scoring function (such as weighted summation or the sigmoid function) is input, and a score between 0 and 1 is output. The closer to 1, the more stable the ship.

[0146] , , These are pre-set parameters, not obtained in real-time from sensor data; they are empirical parameters derived through experiments. The goal is to achieve a high overall confidence level. It can sensitively and accurately reflect the overall risk of the system under different failure modes (such as pure visual failure, violent shaking, etc.), and the number of frames N is based on the visual sensor.

[0147] Most existing unmanned surface vessel (USV) environmental perception systems are designed based on large-tonnage motorized vessels or platforms on stable waterways, assuming that the platform is in a relatively quasi-static or low-dynamic state. When these systems are directly applied to unmanned sailboats, they suffer from fundamental flaws:

[0148] Loss of field of view due to drastic attitude changes: In order to utilize wind energy, unmanned sailboats often need to maintain a heel angle of 10° to 30°, and are accompanied by severe pitching in the waves. This causes the field of view of fixed cameras or lidar to frequently shift significantly - one side of the sensor "looks at the sky" and the other side "looks at the sea", the horizon is severely tilted or even lost, causing conventional target detection algorithms to fail.

[0149] High false alarm rate in high sea states: Unmanned sailboats have low freeboard and strong wave-following behavior (rising and falling with the waves). Existing perception algorithms have difficulty distinguishing between nearby swells and real rigid obstacles, often misjudging broken waves as obstacles, leading to "seeing enemies everywhere" and frequently triggering false obstacle avoidance alarms.

[0150] Distortion of dynamic environment perception: The navigation of unmanned sailboats depends on the accurate capture of wind. However, the anemometer installed on the top of the towering mast generates a huge tangential linear velocity (lever effect) as the hull rolls, resulting in a serious motion-induced error superimposed on the measured apparent wind speed, which cannot reflect the true sea surface wind field.

[0151] Lack of proactive "perception-control" interaction: Existing systems typically have the perception module passively transmitting data to the control module. When the environment is too harsh (such as a stormy night) and the perception fails completely, the system often lacks a mechanism to proactively request the ship to adjust its attitude to restore perception, which can easily lead to a dangerous situation of blind navigation.

[0152] This invention provides an adaptive environment perception system for an unmanned sailboat. The system includes a physical trigger acquisition module, a multi-source data processing and fusion module, a wave and wind field environment modeling module, and a perception state decision module. The perception system outputs environmental situation information and interaction requests to an external navigation controller. The physical trigger acquisition module acquires raw environmental data, integrates attitude-based discrete sampling and image preprocessing mechanisms, and can monitor the data from the ship's inertial measurement unit in real time. It generates a synchronous trigger signal only when the ship's motion stability meets a preset quasi-static window threshold to activate the lidar and visual sensor for discrete sampling. It constructs a virtual stable window based on the real-time gravity vector, reverses the rotation of the raw image output by the visual sensor, and dynamically locks the region of interest, outputting image data with the sea-line horizontally centered. The multi-source data processing and fusion module cleans and extracts features from the acquired data, calculates the vertical motion frequency of the detected target, filters out wave interference point clouds strongly correlated with the ship's heave and pitch frequencies, establishes a mast rigid body motion model to eliminate sway-induced wind speeds in anemometer measurements, and combines the ground velocity vector provided by the global navigation satellite system. The system integrates heading information with meteorological data to generate true wind field vectors and forward gust predictions. It also adaptively adjusts the fusion weights of each sensor's data based on the ship's heel angle at the time of data collection. A wave and wind field environment modeling module generates environmental situation maps and performs spatiotemporal registration based on processed lidar data, ship inertial measurement unit data, GNSS data, and absolute positioning coordinates provided by GNSS. This constructs a wave potential energy field map including wave height and impact energy distribution, and integrates the wind field distribution map determined by wind dynamics with the physical safety data determined by obstacles and high-energy waves. The entire boundary is spatially mapped and overlaid to generate an environmental grid map or topology map representing the dynamic joint navigable domain, which is sent as output data to the external planning system. The perception state decision module is used to monitor the operation status and confidence level of the perception system itself, and has the function of active perception interaction. It evaluates the integrity and confidence level of the environmental model in real time. When it detects that the perception data is unreliable for N consecutive frames or that the environment is too harsh to cause the loss of positioning features, it generates a stabilization observation request signal and sends it to the external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, it triggers a reset observation.

[0153] This system can achieve highly robust and low-power accurate sensing in complex marine environments through discrete triggering at the physical layer, dynamic denoising and fusion at the data layer, and active interaction at the decision layer.

[0154] The physical trigger acquisition module integrates hardware such as the ship's inertial measurement unit, global navigation satellite system, lidar, and visual sensors to acquire raw environmental data. Addressing the severe swaying characteristics of unmanned sailboats, this module does not employ the traditional continuous sampling mode but instead uses an attitude-based discrete sampling mechanism: it monitors the ship's inertial measurement unit data in real time, generating a synchronous trigger signal to activate high-power sensors for instantaneous acquisition only when the ship's motion stability meets a preset quasi-static window threshold, thus eliminating motion blur at its physical source. Simultaneously, the module utilizes real-time gravity vectors to construct a virtual stable window, performing reverse rotation and dynamic region of interest locking on the original image to ensure that the sea-line in the output image remains horizontally centered, improving the efficiency of subsequent algorithms.

[0155] The multi-source data processing and fusion module is responsible for cleaning, denoising, and feature extraction of heterogeneous data. For wave interference, this module employs wave dynamics filtering based on frequency domain spectral correlation to calculate the target's vertical motion frequency and filter out "wave interference point clouds" that are strongly correlated with the ship's heave and pitch frequencies. For wind field perception, this module establishes a rigid body motion model of the mast, combines the velocity vector from the Global Navigation Satellite System to eliminate sway-induced wind speeds from anemometer measurements, and integrates visual analysis of dark sea surface textures (cat's claw winds) to generate meteorological data containing true wind field vectors and forward gust predictions. Furthermore, this module adaptively adjusts the fusion weights of various sensors (such as port and starboard cameras) based on the ship's heel angle at the time of data acquisition, automatically shielding sensor data with limited field of view at large heel angles.

[0156] The wave and wind field environment modeling module is used to generate an environmental situation map. This module goes beyond the traditional binary obstacle map and constructs a wave potential energy field map with dynamic constraints. By quantifying wave height and impact energy distribution, it identifies high-energy wave regions. Subsequently, the module spatially maps and overlays the "wind field distribution map" determined by wind field dynamics with the "physical safety boundary" determined by obstacles and high-energy waves to generate an environmental grid map or topology map representing the "dynamic joint navigable domain". This map is then sent to the external planning system as the final perception output data.

[0157] The perception state decision module is used to monitor the system's own operational status. Unlike one-way data output, this module has an active perception interaction function: it evaluates the integrity and confidence of the environmental model in real time; when it detects that the perception data is unreliable for N consecutive frames (such as feature loss due to extreme wind and waves), the module generates a stabilization observation request signal and sends it to the external navigation controller, requesting the ship to perform roll reduction or headwind stop (Heave-to) actions. After the ship's attitude returns to stability, it triggers a reset of observation, thereby realizing an active survival strategy of "control for perception".

[0158] This invention, by introducing core technologies such as physical trigger sampling, wave frequency noise reduction, real wind field reconstruction, and active stabilization interaction, achieves a leap from "passive geometric perception" to "active cognition based on physical laws," effectively solving the problem of perception failure of unmanned sailboats in severe sea conditions and ensuring stable navigation of unmanned sailboats.

[0159] This system utilizes sensors such as inertial measurement units, lidar, visual sensors, anemometers, and global navigation satellite systems, and integrates functions such as physical trigger acquisition, deep fusion of multimodal data, joint environmental modeling, and active perception interaction. At the physical acquisition level, it employs discrete trigger sampling based on the ship's "quasi-static window" and a dynamic region of interest locking mechanism for gravity vectors to address motion blur and field-of-view loss at the source. At the data processing level, it uses the ship-wave resonance characteristics to filter out wave interference point clouds, combines mast motion compensation and visual texture analysis to reconstruct the true wind field vector and predict gusts, and adaptively adjusts the sensor fusion weights based on the ship's attitude. At the environmental modeling level, it constructs a wave potential energy field and generates an environmental grid map representing a "dynamically joint navigable domain" as output by intersecting wind field dynamics, obstacle space, and wave energy distribution in real time. At the interactive decision-making level, it monitors perception confidence in real time and has the interactive capability to actively request an external controller to perform ship stabilization when perception fails. This invention represents a leap from passive geometric perception to active environmental cognition based on physical laws, effectively enhancing the survivability of unmanned sailboats in harsh sea conditions.

[0160] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive environment perception method for unmanned sailboats, characterized in that, include: S1: Real-time monitoring and acquisition of data from the ship's inertial measurement unit. A synchronous trigger signal is generated only when the hull motion stability of the unmanned sailboat meets the preset quasi-static window threshold to activate the lidar and visual sensor for discrete sampling. The target is obtained through the lidar and the original image is obtained through the visual sensor. S2: Construct a virtual stable window based on the real-time gravity vector, reverse rotate and dynamically lock the region of interest in the original image output by the visual sensor, and output image data with the sea-line horizontally centered; S3: Calculate the vertical motion frequency of the target and filter out the wave interference point cloud that is strongly correlated with the heave and pitch frequencies of the ship. S4: Establish a rigid body motion model of the mast to eliminate the sway-induced wind speed in the anemometer measurement, obtain the ground velocity vector and heading information through the global navigation satellite system, fuse them to generate meteorological data containing the true wind field vector and gust forward prediction, and adaptively adjust the fusion weight of each sensor data according to the ship's heel angle at the time of collection. S5: Based on the processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system, spatiotemporal registration is performed to construct a wave potential energy field map that includes wave height and impact energy distribution; S6: Spatial mapping and overlay of the wind field distribution map determined by wind field dynamics with the physical safety boundary determined by obstacles and high-energy ocean waves to generate an environmental grid map or topology map representing the dynamic joint navigable domain, which is then sent as output data to the external planning system. S7: Real-time assessment of the integrity and confidence of the environmental model. When N consecutive frames of sensing data are found to be unreliable or the environment is too harsh to cause the loss of positioning features, a stabilization observation request signal is generated and sent to the external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, the observation is reset.

2. The sensing method according to claim 1, characterized in that, S1: Real-time monitoring and acquisition of data from the ship's inertial measurement unit; generating a synchronous trigger signal only when the ship's motion stability meets a preset quasi-static window threshold to activate the lidar and vision sensor for discrete sampling; obtaining the target through the lidar; and obtaining the original image through the vision sensor; including the following steps: The angular velocity data output by the ship's inertial measurement unit is monitored in real time to obtain the ship's real-time roll angular velocity and real-time pitch angular velocity. Set the preset quasi-static window threshold to obtain the preset roll quasi-static threshold and pitch quasi-static threshold; When the real-time roll rate of the hull is less than the roll quasi-static threshold and the real-time pitch rate of the hull is less than the pitch quasi-static threshold, the synchronization trigger signal is generated to activate the lidar and the vision sensor to perform discrete sampling, thereby obtaining the detected target and the original image.

3. The sensing method according to claim 2, characterized in that, The angular velocity data output by the ship's inertial measurement unit is used to generate a trigger signal through logical operations, triggering the logic. This is determined by the following system of inequalities: ； in, for The state of the trigger signal at any given moment. =1 triggers data collection. =0 indicates standby mode; This refers to the ship's real-time roll rate. This refers to the ship's real-time pitch rate. The preset quasi-static threshold for roll is... This is the preset pitch quasi-static threshold.

4. The sensing method according to claim 2, characterized in that, S2: Construct a virtual stable viewport based on the real-time gravity vector, reverse rotate and dynamically lock the region of interest in the original image output by the visual sensor, and output image data with the sea-line horizontally centered, including the following steps: The ship's pitch angle and roll angle are obtained by real-time monitoring of the ship's inertial measurement unit. Obtain the initial coordinates of the optical center of the vision sensor and the equivalent focal length of the camera; Based on the ship's pitch angle, roll angle, initial coordinates of the optical center of the vision sensor, and equivalent focal length of the camera, a reverse compensation matrix is ​​constructed. Geometric correction is performed on the original image to obtain the center ordinate of the region of interest. The initial ordinate of the optical center of the vision sensor is then moved to the center ordinate of the region of interest to obtain the image data.

5. The sensing method according to claim 4, characterized in that, S3: Calculate the vertical motion frequency of the detected target and filter out wave interference point clouds that are strongly correlated with the ship's heave and pitch frequencies, including the following steps: By comparing the vertical displacement frequency of the target object clustering in the point cloud data obtained by the lidar with the heave frequency of the ship's hull, the correlation coefficient is obtained. When the correlation coefficient is greater than or equal to the preset denoising threshold, the point cloud of the target object cluster is filtered out. When the correlation coefficient is less than the preset denoising threshold, the target object is determined to be clustered as an obstacle.

6. The sensing method according to claim 5, characterized in that, S4: Establish a rigid body motion model of the mast to eliminate the sway-induced wind speed in the anemometer measurements. Obtain the ground velocity vector and heading information through the Global Navigation Satellite System, and fuse them to generate meteorological data that includes the true wind field vector and forward gust prediction. At the same time, based on the ship's heel angle at the time of acquisition, adaptively adjust the fusion weights of the data from each sensor. This includes the following steps: The apparent wind speed vector obtained by the anemometer, the three-axis angular velocity vector of the hull obtained by the hull inertial measurement unit, the displacement vector of the anemometer relative to the center of the hull obtained according to the height and position of the anemometer on the mast, and the hull velocity vector relative to the ground obtained by the global navigation satellite system. Treating the mast as a rigid lever, calculate the tangential linear velocity at the anemometer and perform vector subtraction to obtain the true wind field vector. Determined by the following vector equation: ； in, This is the apparent wind speed vector of the anemometer; The vector of angular velocity along the three axes of the hull; This is the displacement vector of the anemometer relative to the center of the ship's hull; Represents the vector cross product operation; The hull-to-ground velocity vector provided for global navigation satellite systems.

7. The sensing method according to claim 6, characterized in that, It also includes the following steps: It is equipped with a directional sensor covering the port and starboard fields of view, as well as a lidar with omnidirectional detection capabilities; The critical angle threshold at which the directional sensor's field of view is blocked by waves or diffused into the sky is preset. The fusion weights of the directional sensors are dynamically adjusted based on the ship's heel angle. Determined by the following S-shaped function formula: ； in, This is the current hull heel angle; The critical angle threshold at which the field of view of the directional sensor is obstructed by sea waves or diffused into the sky; The gain coefficient used to adjust the weight decay rate; When the current port roll angle of the hull is greater than the preset critical angle threshold, the sensor weight of the directional sensor covering the port field of view is less than the sensor weight of the directional sensor covering the starboard field of view. When the current starboard heel angle is greater than the preset critical angle threshold, the sensor weight of the directional sensor covering the port side field of view is greater than the sensor weight of the directional sensor covering the starboard side field of view.

8. The sensing method according to claim 7, characterized in that, S5: Based on the processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system, spatiotemporal registration is performed to construct a wave potential energy field map containing wave height and impact energy distribution. This includes the following steps: The wave energy mapping principle based on gravitational potential energy field is adopted to quantify the potential threat of ocean waves to sailing, and the geometric elevation of ocean waves is converted into energy distribution to construct the ocean wave potential energy field. The wave potential energy density of the ocean wave potential energy field Determined by the following physical formula: ； in, The density of seawater; It is the acceleration due to gravity; The reconstructed wave surface for the sensing system The height at the coordinates; The mean static sea level is the height of the sea.

9. The sensing method according to claim 8, characterized in that, S7: Real-time assessment of the integrity and confidence of the environmental model. When N consecutive frames of sensing data are detected to be unreliable or the environment is too harsh to cause the loss of positioning features, a stabilization observation request signal is generated and sent to the external navigation controller to request the ship to perform roll reduction or stop against the wind. After the ship stabilizes, the observation is reset, which includes the following steps: A sensing lockout determination principle based on multi-dimensional feature consistency is adopted to achieve bidirectional interaction between the sensing system and the control system; By calculating the confidence level of environmental perception in real time An interaction request is triggered when the confidence level is insufficient; the confidence level is determined by the following weighted summation formula: ； in, This is the normalized value of the number of valid feature points extracted in the current frame. The residual value for matching maps from previous and next frames; For attitude stability scoring based on hull inertial measurement unit data; , , These are the weighting coefficients; when Less than the safety threshold ,and Less than The duration exceeded At that time, a stabilization request is triggered, requesting the ship to perform roll reduction or stop against the wind. when Greater than the safety threshold At that time, the sensing system re-initializes its positioning, and triggers a reset of observation after the ship stabilizes.

10. An adaptive environmental perception system for unmanned sailboats, characterized in that: It includes a physical trigger acquisition module, a multi-source data processing and fusion module, a wave and wind field environment modeling module, and a perception state decision module; the perception system is used to output environmental situation information and interaction requests to an external navigation controller; The physical trigger acquisition module is used to acquire raw environmental data. It integrates attitude-based discrete sampling and image preprocessing mechanisms and can monitor the data of the ship's inertial measurement unit in real time. It generates a synchronous trigger signal only when the ship's motion stability meets the preset quasi-static window threshold to activate the lidar and vision sensor for discrete sampling. It constructs a virtual stable window based on the real-time gravity vector, reverses the original image output by the vision sensor and dynamically locks the region of interest, and outputs image data with the sea-line horizontally centered. The multi-source data processing and fusion module is used to clean and extract features from the collected data, and can calculate the vertical motion frequency of the detected target, filter out the wave interference point cloud that is strongly correlated with the heave and pitch frequencies of the ship, establish a rigid body motion model of the mast to eliminate the sway-induced wind speed in the anemometer measurement, and combine the ground velocity vector and heading information provided by the global navigation satellite system to generate meteorological data containing the true wind field vector and gust forward prediction. At the same time, it adaptively adjusts the fusion weight of each sensor data according to the ship's heel angle at the time of collection. The wave and wind field environment modeling module is used to generate an environmental situation map. It can perform spatiotemporal registration based on processed lidar data, ship inertial measurement unit data, global navigation satellite system data, and absolute positioning coordinates provided by the global navigation satellite system to construct a wave potential energy field map containing wave height and impact energy distribution. It also performs spatial mapping and overlay of the wind field distribution map determined by wind field dynamics with the physical safety boundary determined by obstacles and high-energy waves to generate an environmental grid map or topology map representing the dynamic joint navigable domain, which is then sent as output data to an external planning system. The perception state decision module is used to monitor the operating status and confidence level of the perception system itself, and has the function of active perception interaction. It can evaluate the integrity and confidence level of the environmental model in real time. When it detects that the perception data is unreliable for N consecutive frames or that the environment is too harsh to cause the loss of positioning features, it generates a stabilization observation request signal and sends it to the external navigation controller to request the ship to perform roll reduction or stop in the wind. After the ship stabilizes, it triggers a reset observation.

Images