An unmanned ship path optimization method based on visual detection

Abstract

The application discloses an unmanned ship path optimization method based on visual detection, and belongs to the technical field of unmanned ships, and comprises the following steps: receiving first image information collected by a visual sensor carried by an unmanned ship and navigation parameter information of the unmanned ship itself, and constructing an environment perception model; performing motion trend prediction on a dynamic target and comprehensive evaluation of collision risk of the unmanned ship; constructing a multi-dimensional threat evaluation function, quantitatively sorting all known collision risk sources, defining a threat evaluation index for each risk source, and optimizing multi-target conflict logic; and based on the output threat evaluation result, dynamically adjusting the heading and speed of the unmanned ship. In the technical scheme implementation process of the application, the surrounding environment is perceived in real time through visual detection, parameters such as obstacle type, distance, relative speed and motion direction are input into the multi-dimensional threat evaluation function, dynamic weighted sorting of different risk sources is realized, and thus the navigation path of the unmanned ship is optimized.

Description

Technical Field

[0001] This application relates to the field of unmanned surface vessel (USV) technology, specifically a path optimization method for USVs based on vision detection. Background Technology

[0002] Unmanned surface vessels (USVs) are intelligent vessels that can navigate autonomously on water without human intervention. They are widely used in fields such as environmental monitoring, water patrol, and surveying and exploration.

[0003] With the rapid development of artificial intelligence and computer vision technology, intelligence has become one of the core directions of unmanned surface vessels (USVs). Visual detection technology provides key support for the autonomous navigation of USVs in complex water environments, enabling them to identify obstacles, channel boundaries and dynamic targets on the water surface in real time. This provides USVs with accurate environmental perception capabilities and provides them with accurate path planning and obstacle avoidance decision-making basis.

[0004] However, there are still some problems in the existing visual navigation process of unmanned vessels. For example, when multiple static or dynamic obstacles are detected, such as dense floating objects and high-speed vessels on the side, existing methods usually adopt a simple strategy of prioritizing the nearest obstacle or ranking by fixed weights, without considering other factors such as target type and degree of motion threat. This leads to rigid obstacle avoidance decisions and non-optimal navigation planning results, making it difficult to adapt to the real-time and safety requirements of complex dynamic water environments.

[0005] Therefore, it is necessary to provide a vision-based unmanned surface vessel path optimization method to solve the above problems.

[0006] It should be noted that the information disclosed in this background section is only for understanding the background technology of this application concept, and therefore may include information that does not constitute prior art. Summary of the Invention

[0007] Based on the aforementioned problems in the existing technology, the problem to be solved by this application is to provide a path optimization method for unmanned vessels based on visual detection, which optimizes the navigation route of unmanned vessels to make their navigation process more accurate and effective.

[0008] The technical solution adopted by this application to solve its technical problem is: a path optimization method for unmanned surface vessels based on visual detection, comprising:

[0009] The data processing center receives the first image information collected by the visual sensor on the unmanned vessel, preprocesses the first image information, and receives the navigation parameter information from the unmanned vessel itself, calibrates the navigation parameter information, and then constructs an environmental perception model.

[0010] Based on the data output by the aforementioned environmental perception model, the motion trend of dynamic targets is predicted and the collision risk of unmanned vessels is comprehensively assessed.

[0011] Construct a multi-dimensional threat assessment function to quantify and rank all known collision risk sources, define threat assessment indicators for each risk source, and optimize multi-objective conflict logic;

[0012] Based on the output threat assessment results, the course and speed of the unmanned vessel are dynamically adjusted, and a smooth obstacle avoidance trajectory is generated by combining environmental constraints and dynamic models.

[0013] In the implementation of the technical solution of this application, the surrounding environment is perceived in real time through visual detection, and parameters such as obstacle type, distance, relative speed and direction of movement are input into a multi-dimensional threat assessment function to achieve dynamic weighted ranking of different risk sources, thereby optimizing the navigation path of the unmanned vessel.

[0014] Furthermore, the data processing center performs spatiotemporal alignment and feature-level fusion of the preprocessed image information and calibrated navigation parameters to construct a unified environmental perception model, providing input for subsequent path planning.

[0015] Furthermore, spatiotemporal alignment synchronizes image frames and navigation data through timestamp matching and motion compensation algorithms, and uses the kinematic model of the unmanned vessel to predict the trajectory of dynamic targets in the image. Feature-level fusion extracts edges, corners and semantic information from the image, and combines them with velocity and acceleration vectors from the navigation parameters to construct a comprehensive environmental representation that includes static obstacles, dynamic target motion trends and navigable areas.

[0016] Furthermore, the prediction of motion trends of dynamic targets and the comprehensive assessment of collision risks of unmanned vessels include:

[0017] Extract the motion parameters of each dynamic target from the dynamic target elements output by the environmental perception model, including the current position, velocity vector, and acceleration vector;

[0018] Establish a motion trend prediction model to predict the motion process of dynamic targets. The prediction process includes short-term prediction and long-term prediction, where short-term prediction is for linear motion processes and long-term prediction is for nonlinear motion processes.

[0019] Based on the predicted motion of the dynamic target, combined with the unmanned vessel's own motion parameters, the dynamic relative distance between the target and the unmanned vessel, as well as the collision risk, are obtained.

[0020] Furthermore, the current position is obtained by back-projecting the pixel coordinates of the geometric center point of the target detection box in the image to the geographic coordinate system through camera calibration. The velocity vector is calculated based on the position change of the center point of the target detection box in k consecutive frames of images, combined with the time interval. The acceleration vector is obtained by dividing the velocity change when there are enough consecutive frames by the corresponding time interval difference.

[0021] Furthermore, when the target's trajectory is approximately uniform and straight in a short period of time, a short-term prediction method based on constant speed is adopted. When the target's motion exhibits nonlinear characteristics such as acceleration changes or heading adjustments, a nonlinear motion pattern of the target based on a long short-term memory network is adopted. By utilizing its ability to model long-term dependence on time series data, the trajectory evolution law of dynamic targets in complex navigation environments can be captured.

[0022] Furthermore, threat assessment indicators include: relative distance threat, which reflects the current and future minimum proximity of the target to the unmanned vessel; the closer the distance, the higher the threat value, exhibiting a negative exponential decay relationship; relative speed threat, which represents the approach rate of the target to the unmanned vessel; and the faster the approach speed, the higher the collision risk; and mission conflict threat, which reflects the degree to which the target's trajectory will cause the unmanned vessel's mission execution to be interrupted or significantly deviate from its original path. This mission conflict threat is related to the deviation angle of the unmanned vessel's original route, the energy consumption required for path correction, and the time loss.

[0023] Furthermore, the relative velocity threat value is quantified based on the projection components of the relative velocities of the two objects in the collision direction. The larger the projection value, the higher the threat. The sigmoid function is used to map the values ​​to a normalized range.

[0024] Furthermore, the deviation angle, energy consumption increment, and time delay are normalized and weighted summed, and then a threshold comparison method is used to determine whether a task conflict threat has occurred.

[0025] Furthermore, the optimization process of the multi-target conflict logic is as follows: when the spatial avoidance areas of multiple high-threat targets overlap, the target with the highest threat assessment index is selected for active avoidance, and secondary avoidance measures are taken for the second highest threat target; when the avoidance directions of all high-threat targets conflict, the minimum risk path strategy is activated: calculate the total threat exposure area corresponding to each possible avoidance direction, and select the path with the smallest exposure area as the optimal obstacle avoidance scheme.

[0026] The beneficial effects of this application are: This application provides a path optimization method for unmanned vessels based on visual detection, which perceives the surrounding environment in real time through visual detection, inputs parameters such as obstacle type, distance, relative speed and direction of movement into a multi-dimensional threat assessment function, realizes dynamic weighted ranking of different risk sources, and thus optimizes the navigation path of the unmanned vessel.

[0027] In addition to the purposes, features, and advantages described above, this application has other purposes, features, and advantages. A further detailed description of this application will be provided below with reference to the figures. Attached Figure Description

[0028] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0029] Figure 1 This is a schematic diagram of the overall process of a vision-based unmanned vessel path optimization method in this application. Detailed Implementation

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0032] like Figure 1 As shown, this application provides a vision-based unmanned surface vessel (USV) path optimization method. This method provides safe and efficient navigation for USVs in complex aquatic environments and can offer better navigation and obstacle avoidance strategies when multiple obstacles are identified. It adjusts the navigation trajectory in real time to avoid dynamic and static obstacles, ensuring navigation safety. The method includes the following steps:

[0033] Step A: The data processing center receives the first image information collected by the visual sensor on the unmanned vessel, preprocesses the first image information, and receives the navigation parameter information from the unmanned vessel itself. After calibrating the navigation parameter information, it constructs an environmental perception model.

[0034] To optimize the path of the unmanned vessel, it is necessary to acquire its surrounding environmental information and its own state data. The environmental information is the first image information, which is acquired by the visual sensor on the unmanned vessel. The visual sensor is a sensor that can acquire enough raw images for the visual processing system to process. It is a common intelligent sensor and is widely used in navigation systems and visual processing systems. The first image information includes visual data of obstacles in the water, channel boundaries and other dynamic targets. The preprocessing process includes steps such as image denoising, grayscale conversion and contrast enhancement to improve image quality and reduce the computational burden of subsequent processing.

[0035] When optimizing the path of an unmanned surface vessel (USV), in addition to collecting and analyzing the external environment, it is also necessary to consider the USV's own navigation status. Therefore, it is necessary to simultaneously receive its own navigation parameters such as speed, heading, position coordinates, and attitude angles to ensure that the path planning can be based on real-time dynamic characteristics for comprehensive judgment. The USV's own navigation parameters are provided in real time by sensors such as inertial measurement unit, GPS module, and compass, and are calibrated through data fusion algorithm. This data fusion algorithm includes a combination of Kalman filtering and complementary filtering to improve the accuracy and stability of navigation parameters and effectively reduce the measurement error caused by a single sensor in a dynamic environment.

[0036] Kalman filtering is a recursive algorithm that uses the state equation of a linear system to make the optimal estimate of the state through the statistical characteristics of system noise and observation noise. It can effectively fuse multi-source sensor data and suppress noise interference, and is especially suitable for unmanned vessels to solve attitude jitter and positioning drift problems caused by wave disturbance in complex waters. Complementary filtering, on the other hand, uses frequency domain characteristics to weight and fuse high-frequency dynamic response and low-frequency steady-state accuracy, further improving the real-time performance and accuracy of attitude angle estimation.

[0037] Based on this, the data processing center performs spatiotemporal alignment and feature-level fusion of the preprocessed image information and calibrated navigation parameters to construct a unified environmental perception model, providing input for subsequent path planning and supporting the unmanned surface vessel (USV) to achieve high-precision environmental understanding and autonomous decision-making in dynamic waters. Spatiotemporal alignment refers to synchronizing and unifying the coordinates of data collected by different sensors in time and space, ensuring that the position of obstacles in the image information is consistent with the current position and heading of the USV, and eliminating perception deviations caused by sampling delays or coordinate system differences. Specifically, image frames and navigation data are synchronized through timestamp matching and motion compensation algorithms, and the dynamic target in the image is predicted using the USV's kinematic model, thereby achieving dynamic perception.

[0038] Feature-level fusion extracts edges, corners, and semantic information from images and combines them with velocity and acceleration vectors from navigation parameters to construct a comprehensive environmental representation that includes static obstacles, dynamic target motion trends, and navigable areas. This provides a basis for subsequent path planning. For example, feature-level fusion can extract the contour edges (such as HOG features) and semantic categories (such as the "ship" and "floating object" labels detected by YOLOv8) of obstacles in images. Combined with the current velocity vector and heading angle of the unmanned vessel in navigation parameters, the relative position and relative motion direction of the obstacle relative to the unmanned vessel can be calculated. The relative position is mapped from the image pixel coordinate system to the geographic coordinate system through coordinate transformation. The relative motion direction is calculated by the temporal displacement difference of the center point of the target detection box. The temporal displacement difference refers to the change in the pixel coordinates of the center point of the target detection box in two or more consecutive frames of images. Combined with camera calibration parameters and unmanned vessel pose information, the actual movement direction and velocity components in the geographic coordinate system are inferred, thereby determining the motion trend of the dynamic target.

[0039] The environmental perception model outputs information including, but not limited to: static environmental elements, such as channel boundaries, fixed obstacles, and their positions and dimensions; dynamic target elements, such as the real-time position, velocity vector, and predicted movement trend of moving obstacles; and navigable areas, which are the water areas in which unmanned vessels can safely pass, calculated based on obstacle distribution and water depth data. The calculation method can be to divide the remaining water surface space into regular grids after excluding the static obstacle projection area and the safe distance buffer of dynamic targets, and to assign passage costs by combining the water depth layer and dynamic obstacle avoidance radius, thereby generating a two-dimensional rasterized navigable area map, which provides data for subsequent path planning.

[0040] It should be noted that there are significant differences between the paths of water navigation and land navigation. This is mainly reflected in the fact that water navigation is more dynamic, the reference points are unstable, such as changes in water flow speed and wind direction, which have a significant impact on the trajectory, and there is a lack of fixed road network constraints. Therefore, existing road-based path planning methods are difficult to apply directly to the water environment.

[0041] Step B: Based on the data output by the aforementioned environmental perception model, predict the motion trend of dynamic targets and conduct a comprehensive assessment of the collision risk of unmanned vessels;

[0042] After constructing the environmental perception model, the various directly collected data can be output as structured environmental information including the location of static obstacles, the real-time pose of dynamic targets, and navigable areas. This supports trajectory prediction and collision risk quantification analysis of moving targets. Specifically, the above process includes:

[0043] Step B1: Extract the motion parameters of each dynamic target from the dynamic target elements output by the environmental perception model, including the current position, velocity vector, and acceleration vector;

[0044] The motion parameters of dynamic targets are important inputs for unmanned surface vessel (USV) path planning, used to construct a perception model of surrounding traffic trends. The motion parameters of dynamic targets include their current position, velocity vector, and acceleration vector. The current position is obtained by back-projecting the pixel coordinates of the geometric center point of the target detection box in the image to the geographic coordinate system after camera calibration. The velocity vector is calculated based on the position change of the center point of the target detection box in k consecutive frames, combined with the time interval. The acceleration vector is obtained by dividing the velocity change by the corresponding time interval difference when there are enough consecutive frames. In this embodiment, "sufficient consecutive frames" means k is greater than or equal to 10 frames. In addition to calculating by dividing the velocity change by the corresponding time interval difference, the acceleration estimate can also be obtained by smoothing the velocity sequence based on polynomial fitting or Kalman filtering and then taking the derivative. This embodiment does not limit the calculation, as long as the acceleration vector of the dynamic target can be obtained.

[0045] Step B2: Establish a motion trend prediction model to predict the motion process of a dynamic target. The prediction process includes short-term prediction and long-term prediction, where short-term prediction is for linear motion processes and long-term prediction is for nonlinear motion processes.

[0046] Since the motion of dynamic targets in the water environment is often affected by wind, waves, water flow and control commands, it exhibits significant nonlinearity and uncertainty. This nonlinear motion is difficult to accurately obtain through traditional linear models, and it also includes conventional linear motion targets. Therefore, it is necessary to establish a motion trend prediction model. This motion trend prediction model can predict the motion trend of linear dynamic targets through short-term prediction and predict the motion trend of nonlinear dynamic targets through long-term prediction.

[0047] When the target's trajectory is approximately uniform and linear within a short period, a short-term prediction method based on constant velocity is used. For example, position and velocity are recursively estimated using a Kalman filter, and the trajectory is smoothly predicted through state vectors, state transition equations, and observation updates. When the target's motion exhibits nonlinear characteristics such as acceleration changes or heading adjustments, a method based on a long short-term memory network is used to learn the target's nonlinear motion pattern. This method leverages the target's ability to model long-term dependencies on time series data to capture the trajectory evolution patterns of dynamic targets in complex navigation environments. For example, after inputting the target's position and velocity time series data within a fixed time period into the model, the model can automatically learn its motion pattern and output position prediction sequences for future times. Connecting these future position prediction sequences forms a complete prediction result for the future trajectory of the dynamic target.

[0048] Step B3: Based on the predicted motion process of the dynamic target, combined with the motion parameters of the unmanned vessel itself, obtain the dynamic relative distance between the target and the unmanned vessel and the collision risk;

[0049] By spatiotemporally aligning the predicted dynamic target trajectory with the current heading, speed, and position information of the unmanned surface vessel (USV), the minimum relative distance, nearest encounter distance, and encounter time between the two in subsequent time series are calculated. Combined with the size of the USV and safety thresholds, it is determined whether there is a collision risk. If the predicted nearest encounter distance is less than the preset safety distance, a collision threat is determined, triggering the obstacle avoidance decision mechanism and entering the path planning adjustment stage. The nearest encounter distance (DCPA) and encounter time (TCPA) are calculated using relative motion vectors, and can be referred to in the prior art for details, which will not be elaborated in this embodiment.

[0050] Step C: Construct a multi-dimensional threat assessment function, quantify and rank all known collision risk sources, define threat assessment indicators for each risk source, and optimize the multi-objective conflict logic;

[0051] When multiple targets encounter each other, it is necessary to quantify and rank the collision risk levels of the collision risk sources to optimize the subsequent path planning process. In this embodiment, a multi-dimensional threat assessment function is constructed, and a threat assessment index is defined for each risk source. Specifically, the threat assessment index includes:

[0052] The relative distance threat reflects the current and future minimum proximity of the target to the unmanned vessel. The closer the distance, the higher the threat value, exhibiting a negative exponential decay relationship. Specifically, it can be calculated by combining the minimum relative distance and the safe distance threshold in step B3. For example, when the minimum relative distance of the target in the future prediction period is known to be d_min and the safe distance threshold is d_safe, the relative distance threat value can be defined as 1-d_min / d_safe. When d_min≤d_safe, the threat value approaches 1.

[0053] Relative speed threat represents the approach rate between the target and the unmanned vessel. The faster the approach speed, the higher the collision risk. The relative speed threat value can be quantified based on the projection component of the relative speed of the two objects in the collision direction. The larger the projection value, the higher the threat. It can be mapped to a normalized range using the sigmoid function.

[0054] Mission conflict threat reflects the degree to which the target's trajectory will cause the unmanned vessel's mission execution to be interrupted or to deviate significantly from its original path. This mission conflict threat is related to the deviation angle of the unmanned vessel's original route, the energy consumption and time loss required for path correction. The deviation angle, energy consumption increment and time delay can be normalized and weighted summed separately, and then a threshold comparison method can be used to determine whether a mission conflict threat has occurred.

[0055] Based on the above three threat indicators, a multi-dimensional threat analysis is conducted on obstacles, considering dimensions including but not limited to: collision risk threat based on phase distance, dynamic threat based on relative velocity, and mission conflict threat. The core comprehensive threat score used for path optimization decision-making is calculated primarily through a collaborative evaluation model that integrates the two key parameters of relative distance threat and relative velocity threat. For mission conflict threat, it serves as an independent criterion, used to identify high-priority conflict scenarios that may lead to mission disruption, thereby ensuring that the unmanned surface vessel can accurately identify the most critical high-threat targets that need to be avoided. The specific formula is: T_S=(α / TTC +β / d) )×(1+W×T), where T_S is the comprehensive threat score, TTC is the estimated collision time, calculated from the projection components of relative distance d and relative velocity in the collision direction. The smaller the TTC, the higher the collision urgency. d is the real-time relative distance between the unmanned vessel and the risk source, W is the weighting coefficient of the mission conflict threat, and T is the mission conflict threat value. The mission conflict threat value can only be 0 or 1. When there is a significant mission deviation or execution interruption, T is 1, otherwise it is 0. α and β are normalized weight coefficients, satisfying α+β=1, used to balance the contribution of collision urgency and spatial proximity. W is the weight adjustment factor of mission conflict, usually set to a positive value to improve the obstacle avoidance response sensitivity in high-priority mission scenarios. When T=1, the comprehensive threat score will be significantly amplified, thus triggering the obstacle avoidance mechanism first. When T=0, the comprehensive threat score is determined only by the collision urgency and spatial proximity, ensuring that the obstacle avoidance priority can still be accurately assessed in the absence of mission interference, while avoiding over-responding to low-risk targets.

[0056] The mission conflict threat value T is calculated as follows: T = α·Δθ / θ_max + β·ΔE / E_base + γ·Δt / t_plan, where Δθ is the path deviation angle, θ_max is the maximum allowable deviation angle, ΔE is the energy consumption increment due to path correction, E_base is the baseline energy consumption, Δt is the time delay, t_plan is the original planned flight time, and α, β, and γ are normalized weighting coefficients, satisfying α + β + γ = 1. In the formula, Δθ / θ_max refers to the relative deviation. The closer this value is to 1, the more serious the course deviation and the greater the impact on the original mission. ΔE / E_base refers to the relative energy consumption ratio, which is the ratio of the energy consumption increment caused by avoidance to the baseline energy consumption, reflecting the relative increase in energy consumption. The larger this ratio is, the heavier the burden on the energy system of the avoidance action. Δt / t_plan is the relative time delay rate, reflecting the time efficiency loss of mission execution. The larger this value is, the more significant the flight delay. After weighted summation, the mission conflict threat value T is obtained. When T≥0.5, it is determined that there is a significant mission conflict threat, and T is 1; otherwise, it is 0.

[0057] The normalization process is achieved by scaling various indicators to the [0,1] interval. Before performing the normalization operation, the original data needs to be optimized to meet the normalization requirements. Input optimization includes obtaining the statistical distribution characteristics of the original data, such as maximum and minimum values, mean, and variance, and filtering or smoothing outliers accordingly. Then, the normalization operation is performed. In the normalization process of relative speed threats, the min-max normalization method is used to map the relative speed projection components to the [0,1] interval. The sigmoid function is combined to enhance nonlinearity to highlight the threat sensitivity under high proximity rates. In the normalization process of task conflict threats, Z-score standardization combined with threshold truncation strategy is used to eliminate dimensional differences while suppressing extreme value disturbances, thereby ensuring that each threat component is comparable when fused.

[0058] The optimization process of the multi-target conflict logic is as follows: when the spatial avoidance areas of multiple high-threat targets overlap (for example, two ships are located to the left and right front of the unmanned ship respectively, and both need to avoid the middle), the target with the highest threat assessment index is selected first for active avoidance, and secondary avoidance measures are taken for the second highest threat target.

[0059] When the avoidance directions of all high-threat targets conflict, the minimum risk path strategy is activated: calculate the total threat exposure area corresponding to each possible avoidance direction, and select the path with the smallest exposure area as the optimal obstacle avoidance scheme.

[0060] Step D: Based on the output threat assessment results, dynamically adjust the course and speed of the unmanned vessel, and generate a smooth obstacle avoidance trajectory by combining environmental constraints and dynamic models.

[0061] Based on the threat assessment results in step C, the threat assessment indicators for each risk source are updated in real time and input into the obstacle avoidance decision center of the unmanned vessel. The obstacle avoidance decision center triggers corresponding response strategies according to the threat level: for high-threat targets, an emergency avoidance path is immediately planned, the course is adjusted to increase the relative distance, and the speed is controlled to shorten the exposure time; for medium- and low-threat targets, a gradual course fine-tuning is adopted to ensure that the path is smooth and conforms to dynamic constraints, and environmental information such as wind, waves, current disturbances, channel boundaries, and static obstacles are combined to achieve unmanned vessel navigation path optimization in dynamic environments.

[0062] In order to make the obstacle avoidance trajectory smooth and prevent the unmanned vessel from violently shaking due to frequent turns or sudden speed changes, it is also necessary to combine environmental constraints and dynamic models to generate a smooth obstacle avoidance trajectory, use spline interpolation or Bézier curves to smooth the heading angle, and limit the rate of change of acceleration through velocity profile planning to ensure smooth motion. For details, refer to the existing technology, which will not be described in detail in this embodiment.

[0063] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A path optimization method for unmanned surface vessels based on visual detection, characterized in that: include: The data processing center receives the first image information collected by the visual sensor on the unmanned vessel, preprocesses the first image information, and receives the navigation parameter information from the unmanned vessel itself, calibrates the navigation parameter information, and then constructs an environmental perception model. Based on the data output by the aforementioned environmental perception model, the motion trend of dynamic targets is predicted and the collision risk of unmanned vessels is comprehensively assessed. Predicting the motion trend of dynamic targets and comprehensively assessing the collision risk of unmanned vessels includes: Extract the motion parameters of each dynamic target from the dynamic target elements output by the environmental perception model, including the current position, velocity vector, and acceleration vector; Establish a motion trend prediction model to predict the motion process of dynamic targets. The prediction process includes short-term prediction and long-term prediction, where short-term prediction is for linear motion processes and long-term prediction is for nonlinear motion processes. Based on the predicted motion process of the dynamic target, combined with the motion parameters of the unmanned vessel itself, the dynamic relative distance between the target and the unmanned vessel and the collision risk are obtained. The current position is obtained by back-projecting the pixel coordinates of the geometric center point of the target detection box in the image to the geographic coordinate system through camera calibration. The velocity vector is calculated based on the position change of the center point of the target detection box in k consecutive frames of images, combined with the time interval. The acceleration vector is obtained by dividing the velocity change when there are enough consecutive frames by the corresponding time interval difference. A multi-dimensional threat assessment function is constructed to quantify and rank all known collision risk sources. Threat assessment indicators are defined for each risk source, and the multi-target conflict logic is optimized. The threat assessment indicators include: relative distance threat, reflecting the current and future minimum proximity of the target and the unmanned vessel; the closer the distance, the higher the threat value, exhibiting a negative exponential decay relationship; relative speed threat, representing the approach rate between the target and the unmanned vessel; and mission conflict threat, reflecting the degree to which the target's trajectory will cause the unmanned vessel's mission execution to be interrupted or significantly deviate from its original path. This mission conflict threat is related to the deviation angle of the unmanned vessel's original route, the energy consumption required for path correction, and the time loss. The relative speed threat value is quantified based on the projection component of the relative speeds of the two objects in the collision direction; the larger the projection value, the higher the threat, and a sigmoid function is used to map it to a normalized interval. Based on the output threat assessment results, the course and speed of the unmanned vessel are dynamically adjusted, and a smooth obstacle avoidance trajectory is generated by combining environmental constraints and dynamic models.

2. The unmanned surface vessel path optimization method based on vision detection according to claim 1, characterized in that: The data processing center performs spatiotemporal alignment and feature-level fusion of the preprocessed image information and calibrated navigation parameters to construct a unified environmental perception model, providing input for subsequent path planning.

3. The unmanned surface vessel path optimization method based on vision detection according to claim 2, characterized in that: Spatiotemporal alignment synchronizes image frames and navigation data through timestamp matching and motion compensation algorithms, and uses the kinematic model of the unmanned vessel to predict the trajectory of dynamic targets in the image. Feature-level fusion extracts edges, corners and semantic information from the image, and combines them with velocity and acceleration vectors from the navigation parameters to construct a comprehensive environmental representation that includes static obstacles, dynamic target motion trends and navigable areas.

4. The unmanned surface vessel path optimization method based on vision detection according to claim 1, characterized in that: When the target's trajectory is approximately uniform and straight for a short period of time, a short-term prediction method based on constant velocity is used. When the target's motion exhibits nonlinear characteristics such as acceleration changes or heading adjustments, a nonlinear motion pattern of the target based on a long short-term memory network is used. This network's ability to model long-term dependencies on time series data is utilized to capture the trajectory evolution patterns of dynamic targets in complex navigation environments.

5. The unmanned surface vessel path optimization method based on vision detection according to claim 1, characterized in that: The deviation angle, energy consumption increment, and time delay are normalized and weighted summed respectively, and then a threshold comparison method is used to determine whether a task conflict threat has occurred.

6. The unmanned surface vessel path optimization method based on vision detection according to claim 1, characterized in that: The optimization process of multi-target conflict logic is as follows: when the spatial avoidance areas of multiple high-threat targets overlap, the target with the highest threat assessment index is selected for active avoidance, and secondary avoidance measures are taken for the second highest threat target; when the avoidance directions of all high-threat targets conflict, the minimum risk path strategy is activated: calculate the total threat exposure area corresponding to each possible avoidance direction, and select the path with the smallest exposure area as the optimal obstacle avoidance scheme.

Images