Fusion filtering dynamic target tracking and compensating method for water surface three-dimensional point cloud

Abstract

The invention belongs to the technical field of water surface target perception, and particularly discloses a fusion filtering dynamic target tracking and compensation method for a water surface three-dimensional point cloud. The method comprises the following steps: firstly, constructing a state transition model of a water surface target based on the state of the water surface target; then predicting the state of the water surface target based on the state transition model by adopting dual-likelihood weighted particle filtering; and finally, introducing a sliding-mode observer, compensating the particle filtering prediction error, and carrying out weighted fusion on the prediction output of the particle filtering and the prediction output of the sliding-mode observer to obtain the final state of the water surface target. According to the invention, the extended particle filter and the sliding-mode observer are fused to track the position of the water surface dynamic target, and high-precision continuous and stable tracking of the water surface dynamic target in a complex environment is realized.

Description

Technical Field

[0001] This application belongs to the field of water surface target perception technology, and more specifically, relates to a dynamic target tracking and compensation method based on fusion filtering for three-dimensional point clouds on the water surface. Background Technology

[0002] The technology for detecting and tracking moving objects on the water surface has profound strategic significance and broad application value. This technology is not only a key support for the intelligent development of the ocean, but also an important guarantee for national security, the marine economy, and scientific research innovation.

[0003] In complex aquatic environments, such as those affected by dynamic waves, intense lighting changes, and target occlusion, the detection and continuous tracking of moving objects on the water surface, including USVs (Unmanned Surface Vehicles), face significant challenges. Especially when using radar or other sensors for target detection, multipath reflections, clutter interference, and environmental noise can lead to data distortion, loss, or false detections, making accurate target state perception even more difficult. Patent application CN202110329226.6 proposes a surface target detection and tracking control method based on a master-slave formation. This method uses Kalman filters and extended Kalman filters to predict the USV state. However, Kalman filters and extended Kalman filters rely on linear or weakly nonlinear system assumptions, making it difficult to effectively handle the complex nonlinear motion of USV targets and the drastic fluctuations in observation noise caused by environmental changes, easily leading to tracking drift or loss. Patent application CN202510009522.6 proposes a cooperative target encirclement method for unmanned vessels, which uses a particle filter algorithm to predict the target's trajectory and then encircle the target. However, under extreme water surface environment conditions, this method often suffers from data loss and lacks sufficient robustness, making it difficult to meet the real-time and stability requirements of practical engineering. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, this application provides a dynamic target tracking and compensation method for fusion filtering of three-dimensional point clouds on the water surface, which aims to solve the technical problem of unstable detection accuracy that is common in existing water surface USV target detection technology.

[0005] To achieve the above objectives, in a first aspect, this application provides a method for dynamic target tracking and compensation based on fusion filtering of three-dimensional point clouds over water surfaces, comprising: Based on the state of the water surface target, construct a state transition model for the water surface target; The state of the water surface target is predicted based on the state transition model using a double likelihood-weighted particle filter. A sliding mode observer is introduced to compensate for the prediction error of the particle filter, and the prediction output of the particle filter and the prediction output of the sliding mode observer are weighted and fused to obtain the final state of the water surface target.

[0006] Preferably, in the particle filtering, the weight of each particle is updated by combining the observation error of the water surface target with the particle state deviation.

[0007] Preferably, the weight of the particle is as follows:

[0008] in, yes Time of the first The weight of each particle, Let be the likelihood function. yes The observation status of the water surface target at any time. yes Time of the first The predicted state of each particle. This reflects the observation error of surface targets; yes Time of the first The weight of each particle reflects the particle's state deviation.

[0009] Preferably, after each prediction of the state of the water surface target using the particle filter, the number of particles is dynamically adjusted to ensure that the number of particles is proportional to the intensity of the water surface target's movement.

[0010] Preferably, the particle number is dynamically adjusted based on the effective particle number, wherein the effective particle number is:

[0011] in, For the effective number of particles, for Time of the first The weight of each particle, within a preset quantity threshold range. ,like If, then the number of ions decreases, if This increases the number of particles.

[0012] Preferably, a sliding mode observer is introduced to compensate for the particle filter prediction error, specifically as follows: Sliding surface in sliding mode observer Defined as:

[0013] in, The particle filter represents the predicted state component of the water surface target. For the sliding mode observer, this represents the predicted state component of the water surface target. It is a sparse matrix of sliding surfaces; The predicted state update method of the sliding mode observer is as follows:

[0014] in, It is a state transition model for surface targets. It represents the predicted state components of the sliding mode observer for the water surface target at the current moment, indicated by the superscript. Indicates the updated value. For control input of surface targets, It is the estimated output of the sliding mode observer. , For the observation function, It is a surface target The actual observed components at time, The observation gain matrix, It is a sliding mode control item.

[0015] Preferably, the predicted output of the particle filter is weighted and fused with the predicted output of the sliding mode observer to obtain the final state of the water surface target, specifically as follows:

[0016] in, For the final state components of the water surface target, This represents the predicted state components of the water surface target using particle filtering. For the predicted state components of the sliding mode observer for the water surface target, , for fusion weights.

[0017] Preferably, a sliding mode observer is introduced to correct the particle weights, and the corrected particle weights are:

[0018] in, For sliding mode observers to participate in the correction The first moment The weight of each particle, It is a proportional sign. yes The first moment The weight of each particle, It is an exponential function. It is the observation state noise covariance matrix. It is the prediction error covariance matrix of the sliding mode observer. For the surface of the water The actual observed components at time, It is the observation function. yes Time of the first The state of each particle It is the predicted state component of the sliding mode observer for the water surface target.

[0019] Preferably, the position, heading angle, velocity, and acceleration of the surface target are combined to form the state of the surface target, and a state transition model of the surface target is constructed based on the state, wherein the state is:

[0020] in, The state of a water surface target includes five state components: position, location, position ... heading angle ,speed and angular velocity ; This is the matrix transpose; the state transition model is:

[0021] in, For acceleration, Angular acceleration, Time step.

[0022] In a second aspect, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0023] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: (1) Based on traditional particle filtering, this application proposes a dual likelihood weight update mechanism that combines observation error and historical bias to improve the stability and accuracy of position tracking. At the same time, this application introduces an adaptive adjustment mechanism for the effective number of particles, which can dynamically optimize the number of particles according to the target motion state to achieve a balance between tracking performance and computational efficiency.

[0024] (2) In view of the problem that existing particle filtering for predicting the trajectory of water surface targets is prone to target loss in extreme water surface environments, in order to achieve stable target tracking, this application designs a sliding mode observer. By constructing the difference between the particle filter prediction and the observer prediction to form a sliding surface, a switching control law with an adaptive saturation interval is used to suppress system chattering. The sliding mode observer can continuously correct the state during the invalid observation stage of the target, prevent trajectory drift, and realize high-precision continuous and stable tracking of dynamic targets on the water surface in complex environments.

[0025] (3) In order to further integrate the advantages of particle filtering and sliding mode observer, this application designs a state weighted fusion method based on the dynamic adjustment of particle swarm dispersion (covariance matrix). At the same time, a sliding mode estimation bias penalty term is introduced in the particle weight update process. Thus, under the condition of long-term observation sparseness or failure, the continuous tracking time of water surface target can be effectively extended, and the continuity and stability of the final trajectory output of the target can be improved. Attached Figure Description

[0026] Figure 1 This is a flowchart of a dynamic target tracking and compensation method for fusion filtering of three-dimensional point clouds on water surface provided in an embodiment of this application.

[0027] Figure 2 This is a flowchart of the adaptive particle number adjustment provided in the embodiments of this application.

[0028] Figure 3 This is a comparison chart of the target tracking test results of the three filtering algorithms provided in the embodiments of this application.

[0029] Figure 4a This is a tracking point cloud map of a conventional water surface target tracking method provided in the embodiments of this application.

[0030] Figure 4b This is a tracking point cloud map of the water surface target tracking method provided in this application embodiment.

[0031] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0033] The terms "first" and "second," etc., used in the description and claims herein are used to distinguish different objects, not to describe a specific order of objects. For example, "first particle" and "second particle," etc., are used to distinguish different particles, not to describe a specific order of particles.

[0034] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0035] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple particles means two or more particles, etc.

[0036] Example 1: Example 1 of this application will be described below with reference to the accompanying drawings.

[0037] like Figure 1 As shown, Embodiment 1 of this application includes the following steps: S1. Construct a five-degree-of-freedom state vector containing the position, heading angle, velocity, and angular velocity of the surface target. Model the state changes of the surface target through a state transition model. Simultaneously, introduce process noise to simulate the actual disturbance environment, enhance the model's adaptability to different scale motion patterns, and provide accurate prior support for subsequent particle filtering and state prediction. In this embodiment, the five-degree-of-freedom state vector of the water surface target is set as follows:

[0038] in, Indicates matrix transpose. Represents the coordinates of the target's position on the water surface; The heading angle representing the surface target; The speed of the target on the water surface; Let be the angular velocity of the target on the water surface.

[0039] The state transition model for the water surface target is constructed as follows:

[0040] in, For time step, subscript and Indicates the time sequence.

[0041] The observation model is:

[0042] in, For the observation function, , To observe noise.

[0043] S2. The state of the water surface target is predicted using a particle filter algorithm. Based on the traditional particle filter algorithm, this application designs a dual-likelihood weight update mechanism that combines observation error and particle state deviation. The specific process is as follows: At the initial moment, generation One particle:

[0044] The weight of each particle is initialized as follows:

[0045] For each moment For each particle Prediction is made based on the state transition model of the water surface target. The first moment The state vector of each particle for:

[0046] in, It is a state transition model for surface targets. It is the first Particles The state vector at time t, For the first Particles Time-based control input, For the first Particles Process noise at any given moment.

[0047] Based on actual observation status Calculate the likelihood of each particle and update the weights:

[0048] Among them, the observed values It was derived from point cloud data of water surface targets collected by radar. For the first Particles The likelihood at any given time reflects the difference between the predicted state of a particle and the actual observed state. For the first Particles The weight of time step is the cumulative effect of particle state deviation in multiple iterations, reflecting the deviation of particle state from the state transition model; For the first Particles Weight of time.

[0049] The likelihood function typically uses a Gaussian distribution:

[0050] Where R is the observation noise covariance. d As an observation dimension, Pi It is an exponential function. It is the observation function.

[0051] In this application, particle weight updates are not only based on traditional observation error likelihood, but also incorporate the deviation between the predicted particle state and the expected observation value, forming a comprehensive double-likelihood weighting to improve the adaptability of the filter in complex noise environments. During the resampling stage, particle set regeneration is performed based on the weight distribution to prevent particle degradation.

[0052] After updating the particles, perform weight normalization:

[0053] S3. Considering that the number of particles has a direct impact on the performance of particle filtering and the consumption of computing resources, this application proposes to use the effective particle number (ESS) as the core indicator and dynamically adjust the total number of particles in real time in combination with the historical entropy change trend.

[0054] This mechanism can automatically increase the particle density when the target is moving violently and reduce the particle number during the stable phase of the movement, thus balancing tracking accuracy and real-time computation.

[0055] To optimize computational efficiency and prevent particle degradation, this embodiment uses the Effective Particle Count (ESS) to dynamically adjust the particle count. Effective Particle Count The calculation formula is:

[0056] Set threshold range ,like This increases the number of particles, i.e. ;like This reduces the number of particles, i.e. .

[0057] Each time a sample is resampled, particles are resampled according to normalized weights to ensure the representativeness of the particle distribution.

[0058] S4. To improve continuous tracking capability under abnormal conditions such as observation interruption and point cloud distortion, this application designs a sliding mode observer. A sliding surface is formed by constructing the difference between the particle filter estimate and the observer estimate. A switching control law with an adaptive saturation interval is used to suppress system chattering and improve robustness. The sliding mode observer can continuously perform state correction during invalid observation phases to prevent trajectory drift.

[0059] To improve the continuity and robustness of state estimation in cases of sparse observations or model uncertainty, a sliding mode observer is used to compensate for the particle filter output. Sliding mode surface Defined as the prediction component of particle filtering for water surface targets. With sliding mode observer predicting state components The difference:

[0060] It is a sparse matrix of the sliding surface, which can be taken as the identity matrix; The predicted state update method of the sliding mode observer is as follows:

[0061] in, It is a state transition model for surface targets. It represents the predicted state components of the sliding mode observer for the water surface target at the current moment, indicated by the superscript. Indicates the updated value. For control input of surface targets, It is the estimated output of the sliding mode observer. , For the observation function, It is a surface target The actual observed components at time, The observation gain matrix, This is a sliding mode control term used to enhance robustness, designed as follows:

[0062] in, This is the sliding mode gain, used to adjust the switching speed; The function is a sign function; to avoid chattering in sliding mode control, a saturation function is used. Perform a symbolic function instead of the conventional one.

[0063] The predicted states obtained by weighted fusion of the sliding mode observer and particle filter are used to obtain the final predicted state components:

[0064] in, As a fusion weight, it is usually adjusted according to the dynamic characteristics and noise level of the system.

[0065] S5. To further integrate the advantages of particle filtering and sliding mode observers, this application designs a state-weighted fusion method based on dynamic adjustment of particle swarm dispersion (e.g., covariance matrix trace), and introduces a sliding mode estimation bias penalty term during particle weight update. This strategy can effectively extend the continuous tracking time of the system and improve the continuity and stability of the final trajectory output under long-term observation sparsity or failure conditions. The predicted value from the sliding mode observer... The weight update formula for particle filtering is used to correct the weight update accuracy.

[0066] in, For sliding mode observers to participate in the correction The first moment The weight of each particle, It is a proportional sign. yes The first moment The weight of each particle, It is an exponential function. It is the observation state noise covariance matrix. It is the prediction error covariance matrix of the sliding mode observer. For the surface of the water The actual observed components at time, It is the observation function. yes Time of the first The state of each particle It is the predicted state component of the sliding mode observer for the water surface target.

[0067] Through the above-mentioned technological innovations, this invention significantly improves the continuous tracking performance of small targets based on lidar perception in dynamic water environments. It has high robustness, high real-time performance, and engineering deployment friendliness, and is applicable to fields such as intelligent unmanned surface vessels, water robots, and port automatic monitoring systems, with broad practical application prospects.

[0068] Example 2: Example 2 verifies the technical solution of this application through simulation experiments: Introduction to the test platform: Example 2 uses MATLAB / Simulink as the simulation platform to simulate the typical motion process of an unmanned surface vessel (USV). The target trajectory is generated based on a five-degree-of-freedom dynamic model, with process noise and observation noise from the actual environment superimposed. The observation data simulates the output of a shore-based lidar system. The sampling period is 0.1 seconds, and the initial particle count is... N 50 ， Particle number adjustment step size The minimum and maximum effective particle thresholds are 0.4 and 0.9, respectively.

[0069] Simulation parameter design: Sampling step size The time is 0.1 seconds, and the initial state of the target USV is... The initial position is at the origin, the initial velocity is 1 m / s, and the initial heading angle and angular velocity are both 0. To simulate the random motion of the USV on the water surface, the USV's realistic motion model uses uniform acceleration and periodic angular velocity changes to simulate the actual scene. The observed values ​​are constructed by adding Gaussian noise to generate noisy position information and heading angles. The linear velocity and angular velocity are as follows:

[0070] in, It is a uniform acceleration value.

[0071]

[0072] in, This is the maximum angular velocity value.

[0073] Therefore, the actual USV position and heading angle are updated as follows:

[0074] Covariance matrix of observation noise .

[0075] Performance metrics and evaluation: The evaluation criteria are based on three dimensions: tracking accuracy, stability, and operational efficiency. Tracking accuracy is assessed using mean squared error (MSE) and mean absolute error (MAE) to evaluate the filter's ability to fit the true USV trajectory.

[0076]

[0077] in, Represents the actual value. For predicted values, Describing the L2 norm, Represents the square of the L2 norm. This indicates the total number of samples collected.

[0078] Stability is calculated by measuring the standard deviation of the error. Measuring the filter's performance in dynamic scenes:

[0079] in, This represents the standard deviation function. The runtime can be used to evaluate the computational complexity of the algorithm.

[0080] Simulation Result Analysis: Number of cycles T When the values ​​are 20 and 200 respectively, the results are as follows: Figure 3As shown, the MSEs of the particle filter are 0.08 and 0.15, respectively, while the MSEs of the Kalman filter are as high as 2.21 and 15.72. This indicates that the error of the Kalman filter accumulates rapidly over time when dealing with nonlinear and non-Gaussian problems. The MSEs of the extended Kalman filter are 0.22 and 0.70, which are significantly lower than those of the Kalman filter, but still lower than those of the particle filter. The MAE results are also largely consistent, with the particle filter having errors of 0.21 and 0.34 in two sampling periods, respectively, which are much lower than those of the Kalman filter and the extended Kalman filter. This shows that the particle filter can more accurately fit the true trajectory of the USV in complex environments.

[0081] Regarding stability analysis, from From this perspective, particle filtering also outperforms the other two algorithms. When the number of iterations is 20 and 200, particle filtering... The values ​​are 0.13 and 0.18 respectively, significantly lower than the Kalman filter's 0.85 and 2.43 and the Extended Kalman filter's 0.19 and 0.54. This indicates that particle filtering has higher stability in dynamic scenes and can effectively suppress error fluctuations.

[0082] In terms of runtime, the Kalman filter exhibits the highest computational efficiency, completing calculations in just 0.01 seconds and 0.15 seconds, respectively. The Extended Kalman filter, due to the need to calculate the Jacobian matrix, has slightly longer runtimes, at 0.05 seconds and 0.46 seconds. The particle filter falls between the two, with runtimes of 0.01 seconds and 0.22 seconds, respectively. This demonstrates that the particle filter maintains high accuracy and stability while also achieving relatively high computational efficiency.

[0083] Experimental results show that the extended state particle filter exhibits high accuracy and stability in USV target tracking tasks, and can track the true trajectory of the USV well. In summary, the particle filter is suitable for complex nonlinear and non-Gaussian systems, and has broad application potential, especially in USV target tracking tasks.

[0084] Example 3: Example 3 also verifies the technical solution of this application through simulation experiments: Test conditions: The model was trained on a custom-constructed water surface point cloud dataset. Subsequently, a segment of raw point cloud data that was not used in the training was selected as the test input. This point cloud data was captured at a speed of approximately 7 knots, which resulted in some data distortion. Therefore, the detection model had a certain probability of missing detections.

[0085] Data results display: During the experiment, continuous point cloud data was captured frame by frame every 5 seconds and input into two different system architectures. The estimation algorithm began estimation after acquiring three valid historical trajectory data points. The comparison results are shown in the figure. Figure 4a For detection results without the introduction of a joint estimation algorithm, Figure 4b The results show the detection outcomes after incorporating the joint estimation algorithm. Each group contains 20 images, with different colored borders indicating the detection or estimation status: red indicates missed detection, green indicates successful compensation, yellow indicates prediction with some bias, especially in the z-direction, and black indicates normal detection.

[0086] Analysis of experimental results: from Figure 4a and Figure 4b As can be seen, the model exhibits significant missed detections on unknown data. In frames 2, 5, 9, 11, 14, 16, and 19, the point cloud can only scan part of the USV's tail information, making it impossible to determine whether it is a USV even with the naked eye. This represents a situation where severely distorted data leads to the complete failure to identify the target. Frame-by-frame statistics at a frequency of 1 second show that the missed detection rate for this point cloud data is approximately 30% at a speed of 7 knots. After introducing a joint estimation algorithm, most of the corresponding frames were successfully compensated, such as frames 4, 5, 14, and 16, achieving continuous target estimation even when detection is interrupted.

[0087] In summary, the joint estimation algorithm can effectively utilize historical trajectory data of the target to dynamically estimate and compensate for its two-dimensional position and heading, even when the detection model fails temporarily or misses a target. This strategy significantly improves the continuity and robustness of the detection and tracking system in complex water surface scenarios, and is particularly suitable for scenarios where point cloud data is distorted, occluded, or subject to noise interference.

[0088] Based on the methods in the above embodiments, this application provides an electronic device, such as... Figure 5 As shown, the electronic device may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions stored in the memory to execute the methods described in the above embodiments.

[0089] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 this application.

[0090] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0091] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0092] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0093] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0094] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0095] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0096] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for dynamic target tracking and compensation based on fusion filtering of 3D point clouds over water surfaces, characterized in that, include: Based on the state of the water surface target, construct a state transition model for the water surface target; The state of the water surface target is predicted based on the state transition model using a double likelihood-weighted particle filter. A sliding mode observer is introduced to compensate for the prediction error of the particle filter, and the prediction output of the particle filter and the prediction output of the sliding mode observer are weighted and fused to obtain the final state of the water surface target.

2. The fusion filtering dynamic target tracking and compensation method according to claim 1, characterized in that, In the particle filtering, the weight of each particle is updated by combining the observation error of the water surface target with the particle state deviation.

3. The fusion filtering dynamic target tracking and compensation method according to claim 1 or 2, characterized in that, The specific weights of the particles are as follows: in, yes Time of the first The weight of each particle, Let be the likelihood function. yes The observation status of the water surface target at any time. yes Time of the first The predicted state of each particle. This reflects the observation error of surface targets; yes Time of the first The weight of each particle reflects the particle's state deviation.

4. The fusion filtering dynamic target tracking and compensation method according to claim 1, characterized in that, Each time the particle filter is used to predict the state of the water surface target, the number of particles is dynamically adjusted to ensure that the number of particles is proportional to the intensity of the water surface target's movement.

5. The fusion filtering dynamic target tracking and compensation method according to claim 1 or 4, characterized in that, The particle count is dynamically adjusted based on the effective particle count, where the effective particle count is: in, For the effective number of particles, for Time of the first The weight of each particle, within a preset quantity threshold range. ,like If, then the number of ions decreases, if This increases the number of particles.

6. The fusion filtering dynamic target tracking and compensation method according to claim 1, characterized in that, A sliding mode observer is introduced to compensate for the particle filter prediction error, specifically as follows: Sliding surface in sliding mode observer Defined as: in, The particle filter represents the predicted state component of the water surface target. For the sliding mode observer, this represents the predicted state component of the water surface target. It is a sparse matrix of sliding surfaces; The predicted state update method of the sliding mode observer is as follows: in, It is a state transition model for surface targets. It represents the predicted state components of the sliding mode observer for the water surface target at the current moment, indicated by the superscript. Indicates the updated value. For control input of surface targets, It is the estimated output of the sliding mode observer. , For the observation function, It is a surface target The actual observed components at time, The observation gain matrix, It is a sliding mode control item.

7. The fusion filtering dynamic target tracking and compensation method according to claim 1 or 6, characterized in that, The final state of the water surface target is obtained by weighted fusion of the predicted output of the particle filter and the predicted output of the sliding mode observer. in, For the final state components of the water surface target, This represents the predicted state components of the water surface target using particle filtering. For the predicted state components of the sliding mode observer for the water surface target, , for fusion weights.

8. The fusion filtering dynamic target tracking and compensation method according to claim 1, characterized in that, A sliding mode observer is introduced to correct the particle weights. The corrected particle weights are as follows: in, For sliding mode observers to participate in the correction The first moment The weight of each particle, It is a proportional sign. yes The first moment The weight of each particle, It is an exponential function. It is the observation state noise covariance matrix. It is the prediction error covariance matrix of the sliding mode observer. For the surface of the water The actual observed components at time, It is the observation function. yes Time of the first The state of each particle It is the predicted state component of the sliding mode observer for the water surface target.

9. The fusion filtering dynamic target tracking and compensation method according to claim 1, characterized in that, The position, heading angle, velocity, and acceleration of the surface target are combined to form the state of the surface target, and a state transition model of the surface target is constructed based on the state, wherein the state is: in, The state of the water surface target includes 5 state components, namely position. heading angle ,speed and angular velocity ; This is the matrix transpose; the state transition model is: in, For acceleration, Angular acceleration, Time step.

10. An electronic device, characterized in that, include: At least one memory for storing computer programs; At least one processor is configured to execute a program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform the method as described in any one of claims 1-8.

Images