Lidar point cloud modeling method, system, device and medium under dynamic sea conditions

Abstract

The present application relates to the field of laser radar, and provides a laser radar point cloud modeling method, system, device and medium under dynamic sea conditions, comprising obtaining unmanned vehicle radar data, unmanned vehicle motion data and water surface environment data of a measured unmanned vehicle; performing point cloud inter-frame registration and dynamic point elimination to obtain static point cloud data; performing movement compensation in the wake area through the unmanned vehicle motion data to obtain compensated radar data, performing kernel density estimation to obtain Poisson wake parameters, calculating Poisson wake model parameters through the Poisson wake parameters, obtaining Poisson term coefficients based on the kernel density estimation; calculating the total number of Gaussian components, calculating the mean vector and Gaussian component weight of the static point cloud data, obtaining a covariance matrix, and constructing a dynamic surface model; constructing a parallel maximum likelihood function, performing model parameter estimation through the parallel maximum likelihood function to obtain updated parameters, thereby obtaining target point cloud data and completing the modeling of the point cloud data.

Description

Technical Field

[0001] This invention relates to the field of lidar technology, and in particular to lidar point cloud modeling methods, systems, devices and media under dynamic sea conditions. Background Technology

[0002] As a multifunctional intelligent platform, unmanned surface vessels (USVs) exhibit enormous application potential due to their lightweight design, high maneuverability, long endurance, and adaptability to harsh environments. The core prerequisite for achieving autonomous navigation and local path planning is accurate and robust environmental perception, especially the detection and identification of obstacles on and near the water surface.

[0003] Traditional surface sensing relies primarily on optoelectronic equipment and marine radar. Optoelectronic equipment can provide high-resolution images, but it is severely limited by lighting and weather conditions, and it is difficult to provide accurate distance and three-dimensional scale information. Marine radar has all-weather operation capabilities and long detection range, but it suffers from low update frequency and insufficient resolution in close-range and high-speed scenarios. Furthermore, it is susceptible to interference from sea clutter and multipath effects in nearshore areas, leading to false alarms and missed detections.

[0004] As an active sensor, lidar combines the detail resolution capability of optical imaging with the precise ranging capability of radar. It can directly acquire the three-dimensional point cloud data of the target, providing information on the target's spatial position, outline, and size. It is basically unaffected by illumination and has a fast scanning speed, making it an ideal tool for short-range target detection and identification on unmanned vessels. However, the application of lidar in actual dynamic marine environments still faces severe challenges: (1) The six degrees of freedom motion of the unmanned vessel itself in the waves, such as roll, pitch, and heave, will cause severe distortion and registration errors in the point cloud; (2) Dynamic sea surface structures such as water waves and ship wakes will generate dense and time-varying interference point clouds, which are mixed with the point clouds of real static or dynamic obstacles and are difficult to distinguish; (3) Point cloud data may become sparse at long distances or in harsh sea conditions, reducing the detection capability for small targets; (4) Sensor noise, sea clutter, and variable environmental background further increase the complexity of data processing. Summary of the Invention

[0005] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a method for modeling lidar point clouds under dynamic sea conditions, which reduces the interference of wake and waves on the identification of unmanned aerial vehicles (UAVs) and accurately models the lidar point cloud data of UAVs.

[0006] This invention provides a method for modeling lidar point clouds under dynamic sea conditions, including:

[0007] S1: Identify the unmanned aerial vehicle (UAV) being measured, and acquire its radar data, motion data, and surface environment data.

[0008] S2: Perform inter-frame registration and dynamic point removal on the UAV radar data using UAV motion data to obtain multi-frame point cloud data. Smooth and weighted fusion of the multi-frame point cloud data to obtain static point cloud data.

[0009] S3: Determine the wake region, perform motion compensation using UAV motion data within the wake region to obtain compensated radar data, and obtain Poisson wake parameters by performing kernel density estimation based on surface water environment data and compensated radar data. Calculate Poisson wake model parameters using Poisson wake parameters and obtain Poisson term coefficients based on kernel density estimation.

[0010] S4: Calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of the static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. Construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0011] S5: Construct a parallel maximum likelihood function, estimate model parameters through the parallel maximum likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

[0012] According to the lidar point cloud modeling method under dynamic sea conditions provided by the present invention, step S1 further includes:

[0013] S11: Determine the measurement unmanned vehicle, and install motion acquisition sensors, lidar and water surface environment acquisition sensors on the measurement unmanned vehicle;

[0014] S12: The radar data of the unmanned vehicle is collected by the lidar, the motion data of the unmanned vehicle is obtained by the motion acquisition sensor, and the water surface environment data is collected by the water surface environment acquisition sensor.

[0015] According to the lidar point cloud modeling method under dynamic sea conditions provided by the present invention, step S2 further includes:

[0016] S21: The motion data of the unmanned vehicle includes the change of Euler angles of the unmanned vehicle and the displacement parameters of the unmanned vehicle. A rotation matrix is ​​constructed based on the change of Euler angles of the unmanned vehicle, and a displacement vector is obtained based on the displacement parameters of the unmanned vehicle. Initial multi-frame point cloud data is obtained using the rotation matrix, the radar data of the unmanned vehicle, and the displacement vector.

[0017] S22: Calculate the point cloud moving speed using the displacement vector, calculate the wave period term, ship speed term, and transient decay term, calculate the dynamic threshold based on the wave period term, ship speed term, and transient decay term, and dynamically remove points from the initial multi-frame point cloud data using the dynamic threshold and the point cloud moving speed to obtain the multi-frame point cloud data;

[0018] S23: The multi-frame point cloud data is smoothed using a Kalman filter and then weighted and fused to obtain the static point cloud data.

[0019] According to the lidar point cloud modeling method under dynamic sea conditions provided by the present invention, step S3 further includes:

[0020] S31: Determine the ship speed and ship yaw rate, and determine the wake region based on the ship speed and ship yaw rate;

[0021] S32: Calculate the point cloud rotation term, point cloud translation term, and point cloud dynamic displacement term based on the motion data of the unmanned vehicle, and perform motion compensation on the static point cloud data in the wake region using the point cloud rotation term, point cloud translation term, and point cloud dynamic displacement term to obtain the compensated radar data;

[0022] S33: Calculate the wake characteristic wavelength based on the motion data of the unmanned vehicle, calculate the Kelvin wake mode term based on the wake characteristic wavelength and the compensation radar data, calculate the ship's draft based on the surface environment data, obtain the draft-dependent attenuation through the ship's draft, and perform kernel density estimation to obtain kernel density estimation parameters.

[0023] S34: Calculate the Poisson wake model parameters based on the Poisson wake parameters, including kernel density estimation parameters, draft-dependent decay, and Kelvin wake model terms, and calculate the Poisson term coefficients based on the kernel density estimation parameters.

[0024] According to the lidar point cloud modeling method under dynamic sea conditions provided by the present invention, step S4 further includes:

[0025] S41: Calculate the total number of Gaussian components, obtain the number of cluster points and sample ratio based on the total number of Gaussian components, calculate the mean vector based on the number of cluster points, and obtain the weight of the Gaussian components through the mean vector and the sample ratio;

[0026] S42: Perform principal component analysis to obtain the local surface principal component matrix, obtain the covariance matrix based on the local surface principal component matrix, and construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0027] According to the lidar point cloud modeling method under dynamic sea conditions provided by the present invention, in step S5, the inlier probability density and the outlier probability density of the static point cloud data are calculated, and the parallel maximized likelihood function is constructed based on the inlier probability density and the outlier probability density.

[0028] According to the lidar point cloud modeling method under dynamic sea conditions provided by the present invention, in step S5, a backup model is constructed, the output of the dynamic surface model is used as the main target point cloud data, backup target point cloud data is obtained through the backup model, the health of the main target point cloud data is calculated, and a health threshold is determined. When the health is higher than the health threshold, the target point cloud data is obtained through the main target point cloud data and the backup target point cloud data; otherwise, the backup target point cloud data is used as the target point cloud data.

[0029] This invention also provides a lidar point cloud modeling system for dynamic sea conditions, including:

[0030] Unmanned Vehicle Data Module: Used to identify and measure unmanned vehicles, acquire radar data, motion data, and surface environment data of the unmanned vehicles being measured;

[0031] Static point cloud data module: Used to perform inter-frame registration and dynamic point removal of UAV radar data using UAV motion data to obtain multi-frame point cloud data, and to perform smoothing and weighted fusion of multi-frame point cloud data to obtain static point cloud data;

[0032] Poisson parameter module: used to determine the wake region, perform motion compensation using UAV motion data in the wake region to obtain compensated radar data, obtain Poisson wake parameters based on surface environment data and compensated radar data, perform kernel density estimation, calculate Poisson wake model parameters using Poisson wake parameters, and obtain Poisson term coefficients based on kernel density estimation.

[0033] Dynamic Surface Model Module: Used to calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. The dynamic surface model is constructed using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0034] Target point cloud data module: Used to construct a parallel maximize likelihood function, estimate model parameters through the parallel maximize likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the lidar point cloud modeling method under dynamic sea conditions as described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the lidar point cloud modeling method under dynamic sea conditions as described above.

[0037] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0038] The present invention provides a method, system, device, and medium for modeling lidar point clouds under dynamic sea conditions. Through a dynamic surface model, it achieves high-fidelity modeling of the dynamic water surface in the wake of a ship. Furthermore, by employing dynamic thresholding, motion compensation, and the introduction of wave disturbances, it significantly improves the model's accuracy, robustness, and adaptability in complex marine environments. Additionally, through health monitoring, model switching, and multiple hypothesis testing mechanisms, it ensures that the system can still operate robustly even when sensor data experiences short-term failure or severe interference, greatly improving the operational reliability of unmanned surface vessels in complex and unpredictable marine environments.

[0039] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in this 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 this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the lidar point cloud modeling method under dynamic sea conditions provided by the present invention.

[0042] Figure 2 This is a schematic diagram of the target point cloud data of the lidar point cloud modeling method under dynamic sea conditions provided by the present invention.

[0043] Figure 3 This is a schematic diagram of the structure of the lidar point cloud modeling system under dynamic sea conditions provided by the present invention.

[0044] Figure 4 This is a schematic diagram of the structure of the lidar point cloud modeling device under dynamic sea conditions provided by the present invention.

[0045] Figure label:

[0046] 100. Unmanned Aerial Vehicle Data Module; 200. Static Point Cloud Data Module; 300. Poisson Parameter Module; 400. Dynamic Surface Model Module; 500. Target Point Cloud Data Module; 810. Processor; 820. Communication Interface; 830. Memory; 840. Communication Bus. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0048] In the description of the embodiments of the present invention, it should be noted that the terms "first", "second" and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0049] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention based on the specific circumstances.

[0050] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0051] The following is combined Figures 1 to 4 Specific embodiments of the present invention are described below. Figure 1 A flowchart illustrating the lidar point cloud modeling method under dynamic sea conditions provided by this invention includes:

[0052] S1: Identify the unmanned aerial vehicle (UAV) being measured, and acquire its radar data, motion data, and surface environment data.

[0053] Furthermore, the objective of this stage is to measure the unmanned aerial vehicle (UAV) to obtain its radar data, motion data, and surface environment data. Specifically, step S1 further includes:

[0054] S11: Determine the measurement unmanned vehicle, and install motion acquisition sensors, lidar and water surface environment acquisition sensors on the measurement unmanned vehicle;

[0055] S12: The radar data of the unmanned vehicle is collected by the lidar, the motion data of the unmanned vehicle is obtained by the motion acquisition sensor, and the water surface environment data is collected by the water surface environment acquisition sensor.

[0056] The specific implementation method for the above steps in this embodiment is as follows:

[0057] First, the vessel to be measured, i.e. the unmanned aerial vehicle (UAV) to be tested, needs to be identified. Then, motion acquisition sensors are set on the UAV to be tested. These motion acquisition sensors include GPS devices and odometers, as well as inertial measurement sensors. A surface environment acquisition sensor, which is mainly responsible for collecting data such as wave height and water surface height, also needs to be set up. In addition, a lidar is also required.

[0058] Subsequently, the environment around the unmanned aerial vehicle (UAV) under test, including the ship in motion, is measured and collected by lidar to obtain UAV radar data; the motion acquisition sensor also needs to collect its own motion coordinates, speed, rotation angle, etc. to obtain UAV motion data; and the water surface environment acquisition sensor also needs to collect data such as water wave height and water surface height.

[0059] S2: Perform inter-frame registration and dynamic point removal on the UAV radar data using UAV motion data to obtain multi-frame point cloud data. Smooth and weighted fusion of the multi-frame point cloud data to obtain static point cloud data.

[0060] Furthermore, the objective of this stage is to perform inter-frame registration and dynamic point removal in the point cloud, followed by smoothing and weighted fusion to obtain static point cloud data. Specifically, step S2 further includes:

[0061] S21: The motion data of the unmanned vehicle includes the change of Euler angles of the unmanned vehicle and the displacement parameters of the unmanned vehicle. A rotation matrix is ​​constructed based on the change of Euler angles of the unmanned vehicle, and a displacement vector is obtained based on the displacement parameters of the unmanned vehicle. Initial multi-frame point cloud data is obtained using the rotation matrix, the radar data of the unmanned vehicle, and the displacement vector.

[0062] S22: Calculate the point cloud moving speed using the displacement vector, calculate the wave period term, ship speed term, and transient decay term, calculate the dynamic threshold based on the wave period term, ship speed term, and transient decay term, and dynamically remove points from the initial multi-frame point cloud data using the dynamic threshold and the point cloud moving speed to obtain the multi-frame point cloud data;

[0063] S23: The multi-frame point cloud data is smoothed using a Kalman filter and then weighted and fused to obtain the static point cloud data.

[0064] The specific implementation method for the above steps in this embodiment is as follows:

[0065] The motion data of an unmanned aerial vehicle (UAV) includes the change in Euler angles and the displacement parameters, which are essentially the displacement data of the UAV. First, the rotation matrix of the UAV needs to be constructed based on the change in Euler angles. ,in, Let be the change in the pitch angle of the unmanned aerial vehicle at a given moment. This represents the change in the rotation angle of the unmanned aerial vehicle at a given moment. This represents the change in the roll angle of the unmanned aerial vehicle (UAV) at a given moment. The displacement vector can also be obtained from the UAV's displacement parameters. This refers to the displacement vector of the unmanned aerial vehicle (UAV) between two UAV motion data points. This allows for inter-frame registration of point clouds, yielding initial multi-frame point cloud data. :

[0066]

[0067] in, This is radar data for unmanned aerial vehicles.

[0068] Then, the point cloud movement velocity Δv corresponding to the initial multi-frame point cloud data is calculated:

[0069] Δv = Δp / Δt

[0070] Here, Δt represents the time difference between the motion data of the two unmanned aerial vehicles. Additionally, the wave period term at time t also needs to be calculated. Ship speed item and transient decay term :

[0071]

[0072] in, The wave period coefficient. For ship speed coefficient, It is the transient decay coefficient, and , The wave period is derived from the fast Fourier transform analysis of water surface elevation measurements. The maximum wave speed is set based on experience; here we take 15 m / s. For the speed of unmanned aerial vehicles, The delay factor is set to 5. Let 0.8 be the velocity delay factor, and then we can calculate the dynamic threshold at time t. :

[0073]

[0074] Next, dynamic point removal is performed on the initial multi-frame point cloud data using a dynamic threshold. That is, the initial multi-frame point cloud data whose point cloud movement speed at a given moment exceeds the dynamic threshold at that moment is removed to obtain multi-frame point cloud data.

[0075] Finally, a Kalman filter is used to smooth the multi-frame point cloud data, and then weighted fusion is performed to obtain the static point cloud data. :

[0076]

[0077] in, The first weighted fusion coefficient, This is the second weighted fusion coefficient. This is the third weighted fusion coefficient. For smoothed multi-frame point cloud data, This is the static point cloud data from the previous time step. This is the static point cloud data from the previous two time points.

[0078] Furthermore, when obtaining static point cloud data, the resolution of the static point cloud data can be determined based on the speed of the unmanned aerial vehicle. This means that the distance between two adjacent data points changes according to the speed, specifically:

[0079]

[0080] in, For the speed of the unmanned aerial vehicle, when the distance between two data points is less than the resolution, one of them can be deleted. Furthermore, the sampling probability G of static point cloud data can be determined, allowing for probability sampling of the static point cloud data to reduce the computational burden. Specifically:

[0081]

[0082] Where exp() represents the exponential function operation. The distance sensitivity coefficient, This is the distance scaling factor. Let N be the distance between the center of the nth static point cloud data and the center of the ship's point cloud image, where N is the total number of static point cloud data before probability sampling. These are the gradient weight coefficients. Let n be the elevation gradient magnitude of the nth static point cloud data. This represents the maximum value of all elevation gradient magnitudes. Once the sampling probability is obtained, sampling can be performed uniformly across all static point cloud data according to this probability.

[0083] For static point cloud data, data points whose surface elevation gradient magnitude is greater than a preset minimum gradient threshold and whose determinant of the Hessian matrix of surface elevation is greater than a preset minimum curvature threshold can also be marked. When computing power is limited, the marked data points can be given priority for hypothesis testing, target detection or optimization of dynamic surface models.

[0084] S3: Determine the wake region, perform motion compensation using UAV motion data within the wake region to obtain compensated radar data, and obtain Poisson wake parameters by performing kernel density estimation based on surface water environment data and compensated radar data. Calculate Poisson wake model parameters using Poisson wake parameters and obtain Poisson term coefficients based on kernel density estimation.

[0085] Furthermore, the objective of this stage is to perform motion compensation and obtain the Poisson wake parameters, Poisson wake model parameters, and Poisson term coefficients. Specifically, step S3 further includes:

[0086] S31: Determine the ship speed and ship yaw rate, and determine the wake region based on the ship speed and ship yaw rate;

[0087] S32: Calculate the point cloud rotation term, point cloud translation term, and point cloud dynamic displacement term based on the motion data of the unmanned vehicle, and perform motion compensation on the static point cloud data in the wake region using the point cloud rotation term, point cloud translation term, and point cloud dynamic displacement term to obtain the compensated radar data;

[0088] S33: Calculate the wake characteristic wavelength based on the motion data of the unmanned vehicle, calculate the Kelvin wake mode term based on the wake characteristic wavelength and the compensation radar data, calculate the ship's draft based on the surface environment data, obtain the draft-dependent attenuation through the ship's draft, and perform kernel density estimation to obtain kernel density estimation parameters.

[0089] S34: Calculate the Poisson wake model parameters based on the Poisson wake parameters, including kernel density estimation parameters, draft-dependent decay, and Kelvin wake model terms, and calculate the Poisson term coefficients based on the kernel density estimation parameters.

[0090] The specific implementation method for the above steps in this embodiment is as follows:

[0091] First, the unmanned aerial vehicle (UAV) needs to determine the ship's speed and yaw rate while it is sailing. Based on the ship's speed and yaw rate, the wake radius of the ship can be determined. Then, the region in the static point cloud data where the distance difference between the data points and the center of the ship's point cloud image is less than the wake radius is taken as the wake region.

[0092] Subsequently, point cloud rotation and point cloud translation terms were calculated based on the motion data of the unmanned aerial vehicle. and point cloud dynamic displacement item :

[0093]

[0094] Where a is the acceleration of the unmanned aerial vehicle. For motion data noise, The noise intensity coefficient is the point cloud translation term, the point cloud rotation term is the displacement of the unmanned aerial vehicle on the sea level, and the point cloud rotation term is... Its form is a rotation matrix obtained from the values ​​of the unmanned aerial vehicle's rotation angle, pitch angle, and rotation angle. This allows for motion compensation of static point cloud data, yielding compensated radar data used to construct dynamic surface models. :

[0095]

[0096] Next, the characteristic wavelength of the wake is calculated based on the motion data of the unmanned aerial vehicle. :

[0097]

[0098] Where g is the acceleration due to gravity. The characteristic wavelength coefficient of the wake. This represents the ship's heading angle. The Kelvin wake mode term is then calculated based on the characteristic wavelength of the wake. :

[0099]

[0100] Let x be the x-axis coordinate of the center of the ship's image in the environment. Let y be the center coordinate of the ship's image in the environment, x be the x-axis coordinate of the compensated radar data, and y be the y-axis coordinate of the compensated radar data. Let be the wake diffusion coefficient along the x-axis. y is the wake diffusion coefficient along the y-axis.

[0101] Calculate ship draft based on surface environment data :

[0102]

[0103] in, The normal draft of a ship in the environment. To increase the draft of ships under load in the environment, The wave draft is then used as the basis for determining the draft-dependent decay. :

[0104]

[0105] in, Let be the draft-dependent attenuation coefficient, and max{} be the maximum value within the parentheses. This refers to the water surface elevation. To compensate for the elevation of radar data.

[0106] Next, kernel density estimation is performed to obtain the kernel density estimation parameter KDE:

[0107]

[0108] Where h is the bandwidth parameter, For the j-th static point cloud data, Let J be the current i-th static point cloud data in the wake region, and J be the total amount of data in the static point cloud data. This is the Gaussian kernel function.

[0109] Subsequently, Poisson wake model parameters are calculated in the wake region based on Poisson wake parameters, including kernel density estimation parameters, draft-dependent decay, and Kelvin wake model terms. :

[0110]

[0111] in, Based on the basic strength coefficient, Kelvin wake scaling factor, This is the draft effect coefficient. The value is an indicator coefficient; it is set to 1 when the static point cloud data is located in the wake region, and 0 otherwise.

[0112] The Poisson coefficient can also be calculated based on the kernel density estimation parameters. :

[0113]

[0114] in, Here are the kernel density estimation parameters for the j-th static point cloud data. The Poisson coefficients and Poisson wake model parameters can be used to compensate for the wake region in dynamic surface models.

[0115] S4: Calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of the static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. Construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0116] Furthermore, the objective of this stage is to calculate the total number of Gaussian components, the mean vector, and the weights of the Gaussian components, and to obtain the covariance matrix, thereby constructing a dynamic surface model. Specifically, step S4 further includes:

[0117] S41: Calculate the total number of Gaussian components, obtain the number of cluster points and sample ratio based on the total number of Gaussian components, calculate the mean vector based on the number of cluster points, and obtain the weight of the Gaussian components through the mean vector and the sample ratio;

[0118] S42: Perform principal component analysis to obtain the local surface principal component matrix, obtain the covariance matrix based on the local surface principal component matrix, and construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0119] The specific implementation method for the above steps in this embodiment is as follows:

[0120] First, we need to calculate the total number of Gaussian components, K:

[0121]

[0122] Where L is the pre-calculated value of the likelihood function. The adaptive penalty weights are d, where d represents the degrees of freedom. The default value for the prior sample size. This is the default value for regularization strength. β is the threshold for the number of samples that are fully activated according to the Bayesian information standard, and β is the steepness of the sigmoid function. Let be the weight of the Gaussian component in the k-th cluster. Here, for the weight of the Gaussian component and the total number of Gaussian components, the expectation-maximization algorithm is first used to estimate the total number of Gaussian components and the corresponding weight of the Gaussian component. Then, the estimated total number of Gaussian components and the weight of the Gaussian component are fine-tuned to make the equation hold, and the total number of Gaussian components can be obtained.

[0123] After obtaining the total number of Gaussian components, the static point cloud data can be divided into multiple clusters based on the total number of Gaussian components, thereby obtaining the number of data points in each cluster, which is the number of cluster points in the k-th cluster. And obtain the proportion of each cluster in the static point cloud data, that is, the sample proportion of the k-th cluster. Then, the mean vector of the k-th cluster is calculated based on the number of cluster points. :

[0124]

[0125] in, This represents the q-th static point cloud data in the k-th cluster.

[0126] The Gaussian component weights of the k-th cluster can then be obtained. :

[0127]

[0128] in, The empirical attenuation coefficient is... This is the mean vector of the previous time step.

[0129] Next, calculate the eigenvalues ​​of the k-th cluster. :

[0130]

[0131] Where T represents transpose. The wave disturbance intensity is 0.25 times the square of the effective wave height in the surface environment data.

[0132] Subsequently, principal component analysis was performed on the eigenvalues ​​to obtain the local surface principal component matrix of the k-th cluster. Thus, the covariance matrix of the k-th cluster is obtained based on the principal component matrix of the local surface. :

[0133]

[0134] in, Let be the eigenvalues ​​of the first principal component of the k-th cluster obtained from the eigenvalues. These are the eigenvalues ​​of the second principal component of the k-th cluster, obtained from the eigenvalues. The eigenvalues ​​are the eigenvalues ​​of the third principal component of the k-th cluster, obtained from the eigenvalues. Let I be the noise term and I be the identity matrix.

[0135] Finally, a dynamic surface model can be constructed based on the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters. :

[0136]

[0137] in, Let be the k-th Gaussian component of static point cloud data with a given covariance matrix and mean vector.

[0138] S5: Construct a parallel maximum likelihood function, estimate model parameters through the parallel maximum likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

[0139] Furthermore, the objective of this stage is to estimate model parameters and obtain target point cloud data, thereby completing the modeling of the point cloud data. Specifically, in step S5, the inlier probability density and outlier probability density of the static point cloud data are calculated, and the parallel maximized likelihood function is constructed based on the inlier probability density and outlier probability density.

[0140] In step S5, a backup model is constructed, the output of the dynamic surface model is used as the main target point cloud data, backup target point cloud data is obtained through the backup model, the health of the main target point cloud data is calculated, and a health threshold is determined. When the health is higher than the health threshold, the target point cloud data is obtained through the main target point cloud data and the backup target point cloud data; otherwise, the backup target point cloud data is used as the target point cloud data.

[0141] The specific implementation method for the above steps in this embodiment is as follows:

[0142] First, it is necessary to construct a parallel likelihood maximization function to estimate the model parameters of the dynamic surface model, and then adjust the various parameters in the dynamic surface model to obtain updated parameters. :

[0143]

[0144] in, To perform a product of the results obtained from each static point cloud dataset (within parentheses) and output the model parameters that maximize the product result. , This represents the inlier probability density, which is the probability that each static point cloud data point belongs to the dynamic surface model under these model parameters. The out-of-field probability density is the probability that each static point cloud data does not belong to the dynamic surface model under the model parameters. Each static point cloud data has its own in-field probability density and out-of-field probability density. The membership degree of the static point cloud data changes for each static point cloud data and is solved using the expectation-maximization algorithm. This adjusts the parameters of the dynamic surface model so that the dynamic surface model can cover as much static point cloud data as possible.

[0145] Finally, the updated parameters and static point cloud data are substituted into the dynamic surface model to output the target point cloud data, thus completing the point cloud data modeling. Furthermore, due to sudden environmental changes and other factors, the model may fail. In this case, a backup model needs to be constructed. The structure and configuration of the backup model are exactly the same as the lidar point cloud modeling method for dynamic sea states proposed in this invention. Subsequently, the output of the dynamic surface model is used as the primary target point cloud data, and the backup target point cloud data can be obtained through the backup model.

[0146] Next, the health of the main target point cloud data is calculated. :

[0147]

[0148] Here, 𝕀() is an indicator function that takes a value of 1 if the event within the parentheses is true, and 0 otherwise. This represents the position of a point obtained from the target point cloud data at time t. This indicates the actual position of the point at time t. This indicates calculating the distance between points, and TZ represents the total running time. This represents the distance threshold.

[0149] Then, a health threshold is determined. When the health is higher than the health threshold, which is set to 0.7 here, the target point cloud data is obtained by weighted summation of the main target point cloud data and the target point cloud data. Otherwise, the backup target point cloud data is used as the target point cloud data.

[0150] The effectiveness of the lidar point cloud modeling method under dynamic sea conditions was also verified here. A small unmanned surface vessel (USV) was used as the lidar test platform. This platform uses a single-hull external engine power system, which has good maneuverability. The USV carrying the lidar was used for sea trials, and the test was conducted on a vessel sailing at sea at a distance of 180 meters. Figure 2The diagram shows the obtained target point cloud data. It can be seen that the lidar can detect large ships sailing into the port at a distance and effectively eliminates sea clutter and ship wake interference. This illustrates the actual engineering effect of the algorithm and solves the problem of wake and wave interference when unmanned surface vessels detect ships at sea.

[0151] In addition, ablation experiments were conducted. The complete method provided by this invention achieved a true positive rate of 93.37%, a false positive rate of 2.70%, a multi-scale detection rate of 92.00%, and an average processing time of 67 ms. Removing step S4 resulted in a true positive rate of 79.36%, a false positive rate of 6.83%, a multi-scale detection rate of 76.42%, and an average processing time of 62 ms. Removing step S2 resulted in a true positive rate of 81.45%, a false positive rate of 5.72%, a multi-scale detection rate of 78.63%, and an average processing time of 59 ms. Removing step S3 resulted in a true positive rate of 88.74%, a false positive rate of 3.89%, a multi-scale detection rate of 86.27%, and an average processing time of 71 ms. It can be seen that each step in this invention is necessary.

[0152] The following describes the lidar point cloud modeling device under dynamic sea conditions provided by the present invention. The lidar point cloud modeling device under dynamic sea conditions described below and the lidar point cloud modeling method under dynamic sea conditions described above can be referred to in correspondence.

[0153] Figure 3 A schematic diagram of a lidar point cloud modeling system under dynamic sea conditions is shown, such as... Figure 3 As shown, the method for performing lidar point cloud modeling under dynamic sea conditions as described above includes:

[0154] Unmanned vehicle data module 100: used to identify the measured unmanned vehicle, acquire the unmanned vehicle radar data, unmanned vehicle motion data and water surface environment data of the measured unmanned vehicle;

[0155] Static point cloud data module 200: Used to perform inter-frame registration and dynamic point removal on unmanned vehicle radar data using unmanned vehicle motion data to obtain multi-frame point cloud data, and to perform smoothing and weighted fusion on multi-frame point cloud data to obtain static point cloud data;

[0156] Poisson parameter module 300: used to determine the wake region, perform motion compensation through UAV motion data in the wake region to obtain compensated radar data, obtain Poisson wake parameters based on water surface environment data and compensated radar data, perform kernel density estimation, calculate Poisson wake model parameters through Poisson wake parameters, and obtain Poisson term coefficients based on kernel density estimation.

[0157] Dynamic Surface Model Module 400: Used to calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. The dynamic surface model is constructed using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients and Poisson wake model parameters.

[0158] Target point cloud data module 500: Used to construct a parallel maximized likelihood function, estimate model parameters through the parallel maximized likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

[0159] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call a computer program in the memory 830 to execute a lidar point cloud modeling method under dynamic sea conditions, the method including:

[0160] S1: Identify the unmanned aerial vehicle (UAV) being measured, and acquire its radar data, motion data, and surface environment data.

[0161] S2: Perform inter-frame registration and dynamic point removal on the UAV radar data using UAV motion data to obtain multi-frame point cloud data. Smooth and weighted fusion of the multi-frame point cloud data to obtain static point cloud data.

[0162] S3: Determine the wake region, perform motion compensation using UAV motion data within the wake region to obtain compensated radar data, and obtain Poisson wake parameters by performing kernel density estimation based on surface water environment data and compensated radar data. Calculate Poisson wake model parameters using Poisson wake parameters and obtain Poisson term coefficients based on kernel density estimation.

[0163] S4: Calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of the static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. Construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0164] S5: Construct a parallel maximum likelihood function, estimate model parameters through the parallel maximum likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

[0165] Furthermore, when the computer program in the aforementioned memory 830 can be implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0166] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the aforementioned methods for modeling lidar point clouds under dynamic sea conditions, the method comprising:

[0167] S1: Identify the unmanned aerial vehicle (UAV) being measured, and acquire its radar data, motion data, and surface environment data.

[0168] S2: Perform inter-frame registration and dynamic point removal on the UAV radar data using UAV motion data to obtain multi-frame point cloud data. Smooth and weighted fusion of the multi-frame point cloud data to obtain static point cloud data.

[0169] S3: Determine the wake region, perform motion compensation using UAV motion data within the wake region to obtain compensated radar data, and obtain Poisson wake parameters by performing kernel density estimation based on surface water environment data and compensated radar data. Calculate Poisson wake model parameters using Poisson wake parameters and obtain Poisson term coefficients based on kernel density estimation.

[0170] S4: Calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of the static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. Construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

[0171] S5: Construct a parallel maximum likelihood function, estimate model parameters through the parallel maximum likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

[0172] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for modeling lidar point clouds under dynamic sea conditions, characterized in that, include: S1: Identify the unmanned aerial vehicle (UAV) being measured, and acquire its radar data, motion data, and surface environment data. S2: Perform inter-frame registration and dynamic point removal on the UAV radar data using UAV motion data to obtain multi-frame point cloud data. Smooth and weighted fusion of the multi-frame point cloud data to obtain static point cloud data. S3: Determine the wake region, perform motion compensation using UAV motion data within the wake region to obtain compensated radar data, and obtain Poisson wake parameters by performing kernel density estimation based on surface water environment data and compensated radar data. Calculate Poisson wake model parameters using Poisson wake parameters and obtain Poisson term coefficients based on kernel density estimation. When calculating the Poisson wake parameters, the draft-dependent attenuation is calculated based on the surface environment data, the wake characteristic wavelength is calculated, the Kelvin wake mode term is calculated based on the wake characteristic wavelength and the compensated radar data, and the kernel density estimation is performed to obtain the kernel density estimation parameters. The Poisson wake parameters include the kernel density estimation parameters, the draft-dependent attenuation, and the Kelvin wake mode term. S4: Calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of the static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. Construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters. S5: Construct a parallel maximum likelihood function, estimate model parameters through the parallel maximum likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

2. The lidar point cloud modeling method under dynamic sea conditions according to claim 1, characterized in that, Step S1 further includes: S11: Determine the measurement unmanned vehicle, and install motion acquisition sensors, lidar and water surface environment acquisition sensors on the measurement unmanned vehicle; S12: The radar data of the unmanned vehicle is collected by the lidar, the motion data of the unmanned vehicle is obtained by the motion acquisition sensor, and the water surface environment data is collected by the water surface environment acquisition sensor.

3. The lidar point cloud modeling method under dynamic sea conditions according to claim 1, characterized in that, Step S2 further includes: S21: The motion data of the unmanned vehicle includes the change of Euler angles of the unmanned vehicle and the displacement parameters of the unmanned vehicle. A rotation matrix is ​​constructed based on the change of Euler angles of the unmanned vehicle, and a displacement vector is obtained based on the displacement parameters of the unmanned vehicle. Initial multi-frame point cloud data is obtained using the rotation matrix, the radar data of the unmanned vehicle, and the displacement vector. S22: Calculate the point cloud moving speed using the displacement vector, calculate the wave period term, ship speed term, and transient decay term, calculate the dynamic threshold based on the wave period term, ship speed term, and transient decay term, and dynamically remove points from the initial multi-frame point cloud data using the dynamic threshold and the point cloud moving speed to obtain the multi-frame point cloud data; S23: The multi-frame point cloud data is smoothed using a Kalman filter and then weighted and fused to obtain the static point cloud data.

4. The lidar point cloud modeling method under dynamic sea conditions according to claim 1, characterized in that, Step S3 further includes: S31: Determine the ship speed and ship yaw rate, and determine the wake region based on the ship speed and ship yaw rate; S32: Calculate the point cloud rotation term, point cloud translation term, and point cloud dynamic displacement term based on the motion data of the unmanned vehicle, and perform motion compensation on the static point cloud data in the wake region using the point cloud rotation term, point cloud translation term, and point cloud dynamic displacement term to obtain the compensated radar data; S33: Calculate the wake characteristic wavelength based on the motion data of the unmanned vehicle, calculate the Kelvin wake mode term based on the wake characteristic wavelength and the compensation radar data, calculate the ship's draft based on the surface environment data, obtain the draft-dependent attenuation through the ship's draft, and perform kernel density estimation to obtain kernel density estimation parameters. S34: Calculate the Poisson wake model parameters based on the Poisson wake parameters, including kernel density estimation parameters, draft-dependent decay, and Kelvin wake model terms, and calculate the Poisson term coefficients based on the kernel density estimation parameters.

5. The lidar point cloud modeling method under dynamic sea conditions according to claim 1, characterized in that, Step S4 further includes: S41: Calculate the total number of Gaussian components, obtain the number of cluster points and sample ratio based on the total number of Gaussian components, calculate the mean vector based on the number of cluster points, and obtain the weight of the Gaussian components through the mean vector and the sample ratio; S42: Perform principal component analysis to obtain the local surface principal component matrix, obtain the covariance matrix based on the local surface principal component matrix, and construct a dynamic surface model using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters.

6. The lidar point cloud modeling method under dynamic sea conditions according to claim 1, characterized in that, In step S5, the inlier probability density and outlier probability density of the static point cloud data are calculated, and the parallel maximized likelihood function is constructed based on the inlier probability density and outlier probability density.

7. The lidar point cloud modeling method under dynamic sea conditions according to claim 1, characterized in that, In step S5, a backup model is constructed, the output of the dynamic surface model is used as the main target point cloud data, backup target point cloud data is obtained through the backup model, the health of the main target point cloud data is calculated, and a health threshold is determined. When the health is higher than the health threshold, the target point cloud data is obtained through the main target point cloud data and the backup target point cloud data; otherwise, the backup target point cloud data is used as the target point cloud data.

8. A lidar point cloud modeling system for dynamic sea conditions, used to execute the lidar point cloud modeling method for dynamic sea conditions as described in any one of claims 1 to 7, characterized in that, include: Unmanned Vehicle Data Module: Used to identify and measure unmanned vehicles, acquire radar data, motion data, and surface environment data of the unmanned vehicles being measured; Static point cloud data module: Used to perform inter-frame registration and dynamic point removal of UAV radar data using UAV motion data to obtain multi-frame point cloud data, and to perform smoothing and weighted fusion of multi-frame point cloud data to obtain static point cloud data; Poisson parameter module: used to determine the wake region, perform motion compensation using UAV motion data in the wake region to obtain compensated radar data, obtain Poisson wake parameters based on surface environment data and compensated radar data, perform kernel density estimation, calculate Poisson wake model parameters using Poisson wake parameters, and obtain Poisson term coefficients based on kernel density estimation. When calculating the Poisson wake parameters, the draft-dependent attenuation is calculated based on the surface environment data, the wake characteristic wavelength is calculated, the Kelvin wake mode term is calculated based on the wake characteristic wavelength and the compensated radar data, and the kernel density estimation is performed to obtain the kernel density estimation parameters. The Poisson wake parameters include the kernel density estimation parameters, the draft-dependent attenuation, and the Kelvin wake mode term. Dynamic Surface Model Module: Used to calculate the total number of Gaussian components, calculate the mean vector and Gaussian component weights of static point cloud data based on the total number of Gaussian components, and obtain the covariance matrix. The dynamic surface model is constructed using the mean vector, Gaussian component weights, covariance matrix, Poisson coefficients, and Poisson wake model parameters. Target point cloud data module: Used to construct a parallel maximize likelihood function, estimate model parameters through the parallel maximize likelihood function, obtain updated parameters, substitute the updated parameters and static point cloud data into the dynamic surface model to obtain target point cloud data, and complete the modeling of point cloud data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the lidar point cloud modeling method under dynamic sea conditions as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the lidar point cloud modeling method under dynamic sea conditions as described in any one of claims 1 to 7.

Images