Radar sequence image-based inland waterway traffic complexity evaluation method
By employing a four-step closed-loop method based on radar sequence images, the real-time and accuracy issues in assessing the complexity of inland waterway traffic were resolved. This method enabled stable tracking of vessel trajectories and precise quantification of interactive relationships, providing scientific decision support for maritime supervision and improving the real-time performance and accuracy of inland waterway supervision.
Patent Information
- Application Number
- CN202511459077.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-13
- Publication Date
- 2026-02-10
AI Technical Summary
Existing methods for assessing the complexity of inland waterway traffic suffer from problems such as insufficient real-time performance, high subjectivity, discontinuous analysis of ship motion trajectories, significant noise impact from radar images, and difficulty in quantifying dynamic interaction relationships. In particular, they are difficult to achieve accurate traffic situation awareness in densely populated areas.
A four-step closed-loop method based on radar sequence images is adopted, including ship detection with background subtraction, ship tracking based on cross-frame position mapping, dynamic information determination of ship interactions, and traffic complexity assessment based on graph theory. Through gray-scale region threshold filtering, Kalman filtering and Hungarian algorithm, Euclidean distance cost function, Euclidean distance optimization, and undirected weighted graph model, stable tracking of ship positions and accurate quantification of interaction relationships are achieved.
It significantly improved the real-time performance and accuracy of traffic situation awareness, enhanced the quantification of ship interactions, established a highly interpretable evaluation system, reduced the workload of maritime regulatory personnel, and promoted the development and application of intelligent maritime regulatory systems.
Smart Images

Figure CN121505924A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent shipping technology, and in particular to a method for evaluating the traffic complexity of inland waterways based on radar sequence images. Background Technology
[0002] As a crucial link in waterway transportation, the accurate assessment of inland waterway traffic complexity plays a vital role in ensuring navigation safety and improving transportation efficiency. However, current methods for assessing inland waterway traffic complexity have many shortcomings and fail to meet the urgent needs of maritime regulatory authorities for real-time and accurate situational awareness.
[0003] Traditionally, inland waterway traffic supervision has relied primarily on maritime inspectors manually observing radar images and Automatic Identification System (AIS) data. This method is highly subjective; different inspectors may make differing judgments about the same scenario, leading to inconsistent assessment results. Furthermore, manual interpretation is labor-intensive, especially in densely populated areas where inspectors struggle to monitor all vessel movements in real time and comprehensively, making it difficult to promptly identify potential collision risks. For example, at busy ports or waterway intersections, the large number of vessels and complex navigational patterns make manual interpretation prone to oversights, thus threatening navigational safety.
[0004] While existing automated assessment methods can reflect waterway congestion to some extent, they are mostly based on static traffic parameters such as ship density and speed distribution, lacking continuous analysis of ship trajectories and thus failing to accurately reflect the complex and ever-changing waterway situation. For example, ship density indicators only represent the number of ships in the waterway at a given moment, failing to reflect the relative motion relationships between ships and potential collision risks. Furthermore, some automated assessment methods based on computer vision technology are severely affected by radar image noise in practical applications. Noise such as sea clutter and rain / snow reflections in radar images can lead to unstable ship detection accuracy, affecting the accuracy of assessment results. Simultaneously, ship tracking algorithms are prone to identification number jumps in scenarios with target occlusion and dense intersections, making it impossible to construct reliable ship trajectories and limiting the application of automated assessment methods. Currently, although some research has been conducted on radar image denoising (such as using wavelet transform), it is still difficult to completely solve the problem of stable detection in complex noisy environments.
[0005] In terms of ship interaction assessment, existing methods, such as the field theory-based method for obtaining maritime traffic flow complexity described in CN116227780A, while considering ship interactions at a microscopic level, primarily rely on static ship attributes (such as position and speed) for modeling. They lack dynamic capture of real-time interactions between ships, making it difficult to comprehensively reflect the complex and ever-changing traffic environment. Similarly, the target waterway ship traffic flow complexity analysis method described in CN120279760A, although identifying interacting ship pairs through cluster analysis and reducing the computational burden of traversing all ships, may overlook dynamic interactions between ships when dealing with large-scale waterways or complex scenarios, leading to an incomplete and inaccurate assessment of complexity. For example, when multiple ships are running parallel or intersecting in a waterway, simple distance indicators alone are insufficient to comprehensively assess their interaction intensity, especially in complex encounter scenarios where traditional methods cannot accurately quantify it.
[0006] Most existing methods assess the risk of inland waterway traffic complexity solely based on vessel spacing, failing to consider dynamic interaction factors such as encounter angles and relative speeds. According to the International Regulations for Preventing Collisions at Sea (COLREGs), collision risk in complex encounter scenarios is not only related to vessel spacing but also closely linked to factors such as encounter angles and relative speeds. Therefore, assessment methods relying solely on vessel spacing cannot meet the needs for refined analysis of complex encounter scenarios.
[0007] In the field of traffic complexity modeling, existing technologies mostly employ statistical methods, such as entropy analysis and cluster analysis. These methods rely on historical data and are computationally complex, making real-time assessment difficult. For example, entropy analysis assesses complexity by calculating the degree of disorder in a system, but requires a large amount of historical data and is difficult to adapt to real-time changing environments. Meanwhile, graph theory, when describing traffic networks, often ignores the dynamic changes in ship interaction intensity, constructing topological graphs based solely on binary relationships (existence / non-existence of interaction). This leads to overly simplistic assessment results that fail to fully reflect the dynamic relationship network of waterways.
[0008] Furthermore, existing methods such as CN116227780A and CN120279760A are inefficient when dealing with large-scale waterways or complex scenarios because they require traversing all vessels for interactive analysis, resulting in a high computational burden. The inland waterway traffic complexity evaluation method based on radar sequence images proposed in this invention achieves matching of vessel detection results across consecutive frames through background subtraction and cross-frame position mapping. It also utilizes Kalman filtering and the Hungarian algorithm to match vessel detection results between adjacent frames, ensuring the correct allocation of identification numbers for each vessel and improving tracking accuracy. Simultaneously, a cost function is introduced, using Euclidean distance to optimize the matching between tracking results and predicted positions. The Euclidean distance cost function ensures accurate matching between detection and prediction results, improving the stability of vessel tracking.
[0009] In summary, existing methods for assessing the complexity of inland waterway traffic have significant shortcomings in terms of real-time performance, accuracy, and dynamic interactive quantification, making it difficult to meet the demands of modern intelligent shipping for efficient and precise traffic situation awareness. This invention achieves high-precision real-time assessment through background subtraction and cross-frame position mapping for ship tracking, filters and algorithms, dynamic information-based ship position determination, and the construction of a stable tracking system. This provides maritime regulators with a scientific and efficient decision support tool, contributing to improved real-time performance and accuracy of inland waterway supervision. Summary of the Invention
[0010] The technical problem this invention aims to solve is to provide a method for evaluating the traffic complexity of inland waterways based on radar sequence images, addressing the challenge of real-time traffic complexity assessment in the field of intelligent inland waterway monitoring. Specifically, traditional manual interpretation of radar images suffers from strong subjectivity and insufficient real-time performance, particularly in densely populated areas where dynamic interactions are difficult to quantify. Existing automated methods rely on static traffic parameters, lacking continuous analysis of vessel trajectories, and radar image noise leads to unstable vessel detection accuracy. Furthermore, vessel tracking is prone to identification number jumps in scenarios involving target occlusion and dense intersections. This invention overcomes the limitations of existing technologies, such as poor real-time performance, low detection accuracy, and insufficient quantification of dynamic interactions.
[0011] To address the aforementioned technical problems, this invention provides a method for evaluating the traffic complexity of inland waterways based on radar sequence images. This method achieves accurate and real-time traffic complexity assessment through a four-step closed loop of "detection-tracking-interaction-evaluation," with the specific steps as follows: Step 1: Ship inspection with background deduction Separate the target ship from the background components in radar images, using grayscale areas. The threshold (within the range of [30, 50], which can be dynamically adjusted according to the noise intensity of the radar image, with an adjustment range of ±10) removes some radar noise interference, providing a clear target basis for subsequent ship tracking.
[0012] Step 2: Ship tracking based on cross-frame position mapping Step 2.1, Introduction of matching algorithm: Kalman filtering and Hungarian algorithm are used to match the ship detection results of adjacent frames to solve the problem of identification number jump in target occlusion and dense intersection scenarios, connect discrete ship position coordinates into continuous trajectory, and quantitatively deduce ship dynamic information.
[0013] Step 2.2, Location Prediction and Update: No. The frame's ship position is represented as follows: (1) In the formula, , These are the horizontal and vertical coordinates of the ship's position in the radar image, respectively. The radar echo is marked by the smallest circumcircle, and the ship's position is represented by the coordinates of the center of the circle. The position of the ship in the current frame is predicted using the position of the ship in the previous frame. The predicted position and the probability distribution of uncertainty are shown in formulas (2) to (3): (2) (3) In the formula, Indicates the predicted position of the ship, representing the first Prior estimation of the ship's position in the frame; It is the covariance matrix of the prior prediction; It is achieved by integrating the detected ship positions with the predicted positions, in the... The optimal ship position obtained from the frame; It is the first The covariance matrix corresponding to the optimal ship position obtained in the frame; and Let these represent the state transition matrix and its transpose, respectively. It is the covariance matrix of the process noise; Indicates the number of frames; By distance threshold The matching detection and prediction results are expressed in Euclidean distance cost function as shown in formula (4); after a successful match, a unique identification number is assigned, and the ship position is updated using Kalman gain as shown in formula (5). The Kalman gain and the uncertainty probability distribution after the update are shown in formulas (6) to (8): (4) (5) (6) (7) (8) In the formula, This represents the Euclidean distance between the detection result and the prediction result. This indicates the location of the vessel being inspected. It is the posterior vector of the ship's position. Indicates the prefit residuals. Represents the inverse of the observation error covariance matrix; It is Kalman gain. To predict the covariance matrix of the posterior matrix, An observational model representing ship motion. It is the covariance matrix of the observation noise. The transpose matrix of the observation model representing the ship's motion.
[0014] Step 3: Use dynamic information to determine ship interactions Step 3.1, Dynamic Information Calculation: Ship dynamic information (relative distance, relative speed, relative heading) is calculated from the ship's position in continuous radar images. The ship's movement distance and relative distance (pixel distance) within adjacent time intervals are converted using a conversion factor. (0.0017, used to convert pixels to nautical miles) is mapped to the actual distance; the ship's distance traveled, speed, and heading are respectively as shown in formulas (9) to (11): (9) (10) (11) In the formula, This indicates the distance traveled by the same ship within adjacent time intervals. and Representing the first The ship's position coordinates in a frame radar image. and They represent the first The ship's position coordinates in the frame, Indicates speed, Indicates the course. Sampling interval; switch chairs This is a distance conversion factor that converts pixels to nautical miles. The value is affected by the selected radar image resolution and is set accordingly. =0.0017.
[0015] Step 3.2, Quantification of Interactions: Step 3.2.1, Relative Distance Influence: Ship interaction is inversely proportional to relative distance. The interaction forces affected by relative distance are as shown in formula (12): (12) In the formula, This represents the interaction forces between ships affected by relative distance. The relative distance between the two ships. This is the minimum safe distance under current traffic conditions; To correct the parameters, and .
[0016] Step 3.2.2, Encounter Angle Influence: According to the International Convention for Preventing Collisions at Sea (1972), ship interaction has a non-linear relationship with the encounter angle. When the relative distance is within the minimum safe distance range, ship interaction... Relative distance adjustment function Interactions affected by the encounter angle As in formulas (13) to (15): (13) (14) (15) In the formula, The encounter angle between the two ships indicates the angle of encounter between the ships. With ships The relative distance between them; It is the minimum distance at which ships must take collision avoidance measures. When the distance between ships is less than the minimum safe distance (i.e., At this determined distance, the interaction between ships is stronger; This is the maximum distance at which ship-to-ship interactions occur. When the relative distance between ships exceeds this distance (usually considered as 1 nautical mile in inland waterways), ship-to-ship interactions can be ignored.
[0017] Step 3.2.3, Encountering Trend Influence: Encountering trend divergence and convergence is determined by the cosine similarity between the relative velocity vector and the relative motion vector. During divergence, the interaction is ignored; the interaction influenced by the trend is considered. As in formula (16): (16) Step 3.2.4, Construction of Interaction Function: Construct the ship interaction function by combining relative distance, encounter angle, and encounter trend. The interaction intensity is normalized to the interval [0, 1] (the larger the value, the stronger the mutual influence), as shown in formulas (17) to (18): (17) (18) In the formula, Represents the relative velocity vector; Indicates the corresponding ship interaction strength; This indicates the final ship interaction intensity.
[0018] Step 4: Graph Theory-Based Traffic Complexity Assessment Step 4.1, Traffic Graph Modeling: Model the inland waterway traffic system as an undirected weighted graph. , is represented as: (19) In the formula, Represents the set of vertices. Represents the set of edges.
[0019] Traffic system modeling can also be represented using an adjacency matrix, which... The edges containing all possible pairs of vertices are expressed as follows: (20) In the formula, , The horizontal and vertical axes represent the number of vessels in the inland waterway transportation system; the interaction between vessel pairs is the same for both vessels, therefore the adjacency matrix is symmetric. = ,and It equals 0.
[0020] Step 4.2, Complexity Assessment: When ship interaction intensity When the threshold is exceeded, the corresponding ship pair is marked in the radar image; the complexity of the traffic flow is assessed using two indicators: "weighted map density" and "proportion of high-risk edges". Weighted graph density : ( )(twenty one) High-risk side proportion : (twenty two) In the formula, for The number of sides, (The number of combinations of choosing 2 vertices from l vertices).
[0021] The inland waterway traffic complexity evaluation method based on radar sequence images proposed in this invention has the following beneficial effects: 1. This invention significantly improves the real-time performance and accuracy of traffic situation awareness. Traditional manual interpretation of radar images suffers from strong subjectivity and insufficient real-time performance, especially in densely populated areas where it is difficult to quantify dynamic interactions. Existing automated methods rely on static traffic parameters, lack continuous analysis of ship trajectories, and radar image noise leads to unstable ship detection accuracy. In scenarios with target occlusion and dense intersections, ship tracking is prone to identification number jumps. This invention effectively suppresses radar noise interference through background subtraction and grayscale threshold filtering, and combines Kalman filtering and the Hungarian algorithm to achieve stable tracking of ship trajectories across consecutive frames, overcoming the above problems. It provides high-precision ship motion data for dynamic interaction analysis, ensuring timely and accurate understanding of ship dynamics in complex waterway scenarios.
[0022] 2. This invention enhances the quantitative ability of ship interactions. Most methods assess risk solely through ship spacing, failing to comprehensively consider dynamic factors such as encounter angle and relative speed, thus failing to meet the requirements of refined collision avoidance rules. This invention innovatively integrates three elements—relative distance, encounter angle, and encounter trend—to construct a nonlinear interaction function, accurately quantifying the complex interactions between ships. This aligns with the refined analysis requirements of the International Maritime Collision Prevention Regulations (IMDG) for encounter scenarios, overcoming the deficiency of traditional static parameter models in dynamically reflecting traffic conditions, and providing a more scientific and comprehensive basis for ship collision avoidance decisions.
[0023] 3. This invention establishes an interpretability assessment system based on graph theory. The traffic system is modeled as an undirected weighted graph, with ships as nodes and interaction strength as edge weights. Interaction reciprocity is expressed through the symmetry of the adjacency matrix, thus visualizing the topological structure of traffic complexity. This visualization method allows maritime regulators to intuitively understand the interaction relationships between ships within the waterway, assisting them in quickly locating high-risk ship pairs. The assessment results possess strong interpretability, contributing to improved regulatory efficiency and accuracy.
[0024] 4. This invention enhances the engineering practicality of inland waterway supervision. It forms a complete closed loop of "detection-tracking-interaction-evaluation," marking vessels exceeding threshold interaction pairs in real time and outputting dual-indicator evaluation results, significantly reducing the workload of manual interpretation. In practical applications, maritime regulators no longer need to spend a significant amount of time manually analyzing radar images; they can quickly make decisions based solely on the system's evaluation results. This provides data support for vessel collision avoidance decisions and waterway resource scheduling, promoting the implementation and application of intelligent maritime supervision systems.
[0025] 5. This invention effectively improves the accuracy of ship detection. In traditional methods, radar image noise leads to unstable ship detection accuracy, affecting subsequent traffic situation awareness and collision avoidance decisions. This invention introduces grayscale region thresholding and background subtraction techniques to effectively suppress noise interference such as sea clutter and rain / snow reflections in radar images. This enables more accurate separation of target ships from background components, thereby improving the accuracy of ship detection and reducing false positives and false negatives.
[0026] 6. This invention solves the problem of identification number jumps in scenarios involving target occlusion and dense intersections. In densely populated areas with ships, target occlusion and dense intersections occur frequently, making it difficult for existing automated methods to accurately track ships, leading to jumps in identification numbers. This invention combines Kalman filtering and the Hungarian algorithm, using the Euclidean distance cost function to allocate ship identification numbers and construct continuous motion trajectories. Even in scenarios involving target occlusion and dense intersections, stable ship tracking can be achieved, ensuring the accuracy and continuity of identification numbers.
[0027] 7. This invention achieves precise construction of a dynamic interaction quantification model. It innovatively integrates three elements—relative distance, encounter angle, and encounter trend—to construct a dynamic interaction quantification model that conforms to the International Maritime Collision Prevention Regulations (IMDG), normalizing the interaction intensity to the [0, 1] interval. This precise quantification method can more accurately reflect the complex interactions between ships, providing more accurate quantitative indicators for collision avoidance and helping ship operators and maritime regulators better assess collision risks between vessels.
[0028] 8. This invention achieves efficient assessment of traffic complexity through an undirected weighted graph model. An undirected weighted graph model is constructed, with ships as nodes and interaction strength as edge weights. Based on the symmetry of the adjacency matrix, a visual assessment of traffic complexity is achieved. This model comprehensively considers various interaction factors between ships, and through the calculation and analysis of the adjacency matrix, it quickly and accurately assesses the traffic complexity of waterways, providing an important basis for the rational scheduling and management of waterway resources.
[0029] 9. This invention reduces the workload of maritime regulatory personnel. Traditional methods of manually interpreting radar images suffer from high workload and insufficient real-time performance, especially in densely populated areas where maritime regulatory personnel need to spend a significant amount of time and effort on interpretation and analysis. The "detection-tracking-interaction-evaluation" closed-loop system formed by this invention can automatically complete tasks such as ship detection, tracking, interactive analysis, and evaluation, and output evaluation results in real time, significantly reducing the workload of maritime regulatory personnel and improving work efficiency.
[0030] 10. This invention promotes the development and application of intelligent maritime supervision systems. The technologies and methods employed in this invention provide key technical support for intelligent maritime supervision systems, enabling real-time, quantitative, and visual assessment of traffic complexity. By real-time marking of interacting vessels exceeding thresholds and outputting dual-indicator assessment results, accurate and timely data support is provided for vessel collision avoidance decisions and waterway resource scheduling. This helps promote the widespread application of intelligent maritime supervision systems in actual waterway management, improving waterway safety and operational efficiency.
[0031] 11. This invention improves the overall level of intelligent monitoring of inland waterways. It effectively solves the problem of real-time assessment of traffic complexity in the field of intelligent monitoring of inland waterways, overcoming the limitations of existing technologies such as poor real-time performance, low detection accuracy, and insufficient dynamic interaction quantification. Through a series of innovative technologies and methods, from ship detection and tracking to interactive analysis and evaluation, a complete intelligent monitoring system is formed, significantly improving the overall level of intelligent monitoring of inland waterways and providing strong support for the safe and efficient operation of inland waterways. Attached Figure Description
[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the overall method of the present invention; Figure 2 This diagram illustrates different encounter scenarios during the vessel's encounter process according to the present invention. Figure 3 This is a schematic diagram of the encounter angle during the encounter process of the ship according to the present invention. Detailed Implementation
[0033] The technical solutions of the present invention will be further described below with reference to the embodiments and accompanying drawings: Example 1 like Figure 1 As shown, this embodiment provides a method for evaluating the traffic complexity of inland waterways based on radar sequence images, including the following steps: Step 1: Ship Detection with Background Subtraction: First, in the crucial stage of radar image processing, it is essential to accurately separate the target ship from the complex background to ensure the accuracy of subsequent analysis. For this purpose, grayscale areas were used. The thresholding method effectively identifies and removes interference in radar images, significantly improving the accuracy and reliability of ship detection. This step not only clearly identifies the ship's outline but also lays a solid foundation for subsequent tracking and analysis.
[0034] Step 2: Ship Tracking Based on Cross-Frame Position Mapping: Match the previously obtained ship tracking results with the current detection results. To ensure that the identification number of each ship can be correctly assigned, a cost function is introduced to optimize the tracking effect by accurately calculating the distance between different ships. This method not only improves the accuracy of tracking but also maintains stable tracking performance in complex environments, ensuring that each ship can be identified and tracked in real time and accurately.
[0035] Step 3: Determine Ship Interactions Using Dynamic Information: To comprehensively assess ship-to-ship interactions, dynamic information such as ship speed and heading is fully utilized to calculate key attributes like relative distance, encounter angle, and encounter trend between ship pairs. By comprehensively analyzing these attributes, the state of the ships can be accurately determined, and potential risks and conflict points can be identified. This step not only provides crucial assurance for safe navigation but also offers strong data support for subsequent decision-making and management.
[0036] Step 4: Graph Theory-Based Traffic Complexity Assessment: In this step, a detailed traffic graph model incorporating ship interaction levels is constructed. When the interaction level between ships exceeds a preset threshold, these ship pairs are specifically marked in the radar image for focused monitoring and analysis. Furthermore, multiple assessment indicators are introduced, and by comprehensively analyzing these indicators, the complexity of traffic flow is fully evaluated. This method not only visually displays traffic conditions but also provides a scientific assessment basis for maritime traffic management, contributing to improved efficiency and safety.
[0037] Furthermore, the aforementioned baseband background subtraction for ship detection involves separating the target ship and background components in the radar image, using grayscale regions. Thresholding removes some radar noise interference: Detection and tracking are two distinct tasks in image processing. Ship detection results are based entirely on each radar image, resulting in independently obtained ship positions. This invention utilizes Kalman filtering and the Hungarian algorithm to match ship detection results from adjacent frames. By tracking ships, discrete ship position coordinates are connected into a continuous trajectory, thereby quantitatively deriving ship dynamic information from radar image sequences.
[0038] In the The detected ship position in the frame is represented as follows: (1) In the formula, , These are the horizontal and vertical coordinates of the ship's position in the radar image, respectively. The radar echo is marked by the smallest circumcircle, and the ship's position is represented by the coordinates of the center of the circle. The position of the ship in the current frame is predicted using the position of the ship in the previous frame. The predicted position and the probability distribution of uncertainty are shown in formulas (2) to (3): (2) (3) In the formula, Indicates the predicted position of the ship, representing the first Prior estimation of the ship's position in the frame; It is the covariance matrix of the prior prediction; It is achieved by integrating the detected ship positions with the predicted positions, in the... The optimal ship position obtained from the frame; It is the first The covariance matrix corresponding to the optimal ship position obtained in the frame; and Let these represent the state transition matrix and its transpose, respectively. It is the covariance matrix of the process noise; Indicates the number of frames; After deriving the predicted ship position, a distance threshold (i.e.) is used. The detection and prediction results of the current frame are matched, and the Euclidean distance cost function is used as shown in formula (4). This is because the position of the same radar echo is highly continuous in two consecutive frames. When the detection result and the prediction result are successfully matched, the radar echo will be assigned a unique ship identification number. Subsequently, the detection and prediction results are integrated using Kalman gain to update the ship position, and the ship position is corrected based on the ship position of the previous frame, as shown in formula (5). The Kalman gain is calculated by formulas (6) and (7), and the uncertainty probability distribution of the updated ship position is given by formula (8): (4) (5) (6) (7) (8) In the formula, This represents the Euclidean distance between the detection result and the prediction result. This indicates the location of the vessel being inspected. It is the posterior vector of the ship's position. Indicates the prefit residuals. Represents the inverse of the observation error covariance matrix; It is Kalman gain. To predict the covariance matrix of the posterior matrix, An observational model representing ship motion. It is the covariance matrix of the observation noise. The transpose matrix of the observation model representing the ship's motion.
[0039] Furthermore, ship interaction is determined based on dynamic information. After ship detection and tracking in the radar sequence images, the degree of interaction between ship pairs is determined using ship dynamic information. Ship dynamic information (i.e., relative distance, relative speed, and relative heading) is calculated from ship positions derived from continuous radar images. It is important to note that the distance traveled by the same ship within adjacent time intervals and the relative distance between ships are both pixel distances, requiring a conversion factor (i.e., ...) to determine the interaction. The distance is mapped to the actual distance. The ship's travel distance, speed, and heading are calculated using formulas (9) to (11), respectively: (9) (10) (11) In the formula, This indicates the distance traveled by the same ship within adjacent time intervals. and Representing the first The ship's position coordinates in a frame radar image. and They represent the first The ship's position coordinates in the frame, Indicates speed, Indicates the course. Sampling interval. Switch chairs. This is a distance conversion factor that converts pixels to nautical miles. The value is affected by the selected radar image resolution and is set accordingly. .
[0040] Formula (11) has four special cases: 1) When and At that time, the heading was 90°; 2) When and At that time, the heading was 270°; 3) When and At that time, the heading was 180°; 4) When and At that time, the heading was 0°.
[0041] Furthermore, the complexity of inland waterway traffic is closely related to environmental factors and traffic conditions, and environmental conditions are reflected through vessel dynamics. Therefore, this study quantitatively analyzes vessel interactions—used to assess the influence relationships between vessel pairs—by examining relative distances, encounter angles, and encounter trends. Under the same traffic density, different relative distances between vessels lead to varying degrees of traffic complexity. Figure 2 As shown, ship interactions are inversely proportional to relative distance: (12) In the formula, This represents the interaction forces between ships affected by relative distance. The relative distance between the two ships. This is the minimum safe distance under current traffic conditions; To correct the parameters, and .
[0042] Furthermore, ship encounters are governed by the International Convention for Preventing Collisions at Sea (1972). The minimum distance encounter time peaks when the encounter angle is approximately 120°, requiring ships to take longer to take collision avoidance measures. The minimum distance encounter time is inversely proportional to maritime traffic complexity, as more complex traffic conditions require a greater safe distance for collision avoidance. In this scenario, ships need to take longer to take collision avoidance measures. Because complex traffic conditions require a greater safe distance for collision avoidance, the minimum distance encounter time is inversely proportional to maritime traffic complexity. Based on the relationship between the encounter angle and the minimum distance encounter time, the trend of the encounter angle's influence on ship interactions can be determined. Specifically, when the encounter angle is greater than 20° or close to 180°, based on the correlation between the encounter angle and the minimum distance encounter time, the trend of ship interaction forces changing with the encounter angle can be determined: when the encounter angle is greater than 20° or close to 180°, the encounter situation is either overtaking or head-on, and the International Convention for Preventing Collisions at Sea details the division of responsibilities and collision avoidance measures between the ship and the target ship. The degree of ship interaction in the two scenarios described above is lower than that in a cross encounter. For overtaking or head-on encounter scenarios, the division of responsibility and collision avoidance measures between the vessel and the target vessel are detailed in the International Convention for Preventing Collisions at Sea (ICP-5). The intensity of ship interaction in these two scenarios is lower than in a cross encounter. Furthermore, ship interaction exhibits a non-linear relationship with the encounter angle.
[0043] like Figure 3 As shown, when the relative distance is within the minimum safe distance range and a sufficient collision avoidance distance is maintained, the vessel should perform relevant safe navigation operations to avoid a collision. Therefore, the vessel interaction in this process is represented by formula (13). The adjustment function related to the relative distance is given by formula (14). The vessel interaction affected by the encounter angle is represented by formula (15). The vessel should perform the corresponding safe navigation operations. The vessel interaction force in this process is represented by formula (13), the adjustment function related to the relative distance is given by formula (14), and the vessel interaction force affected by the encounter angle is given by formula (15): (13) (14) (15) In the formula, The encounter angle between the two ships indicates the angle of encounter between the ships. With ships The relative distance between them; It is the minimum distance at which ships must take collision avoidance measures. When the distance between ships is less than the minimum safe distance (i.e., At this determined distance, the interaction between ships is stronger; This is the maximum distance at which ship-to-ship interactions occur. When the relative distance between ships exceeds this distance (usually considered as 1 nautical mile in inland waterways), ship-to-ship interactions can be ignored.
[0044] Furthermore, when assessing ship interactions, the encounter trend is also a key factor. In reality, there are two encounter trend conditions: divergence and convergence. Both situations have the same encounter angle but exhibit divergent and convergent encounter trends, respectively. When the encounter trend between two ships is divergent, ship interactions are not considered because the ships move away from each other during this process. The encounter trend is expressed through the relative velocity vector. With relative motion vector The cosine similarity is determined. Ship interactions influenced by encounter trends can be expressed as: (16) Furthermore, a ship interaction function was constructed, which comprehensively considers the relative distance, encounter angle, and encounter trend between the ships, as shown in formulas (17) and (18). When the relative distance between the two ships is less than... At a threshold, ship interactions are only affected by relative distance, at which point ships must perform collision avoidance maneuvers. If the relative distance exceeds [a certain threshold], [further action is required]. If the threshold is reached, the impact of the encounter angle and the encounter trend on ship interaction can be ignored. As stated in formula (16), when the two ships are separated, the encounter trend will not affect ship interaction. It should be noted that the ship interaction intensity is normalized to a value in the range [0, 1], and the larger the value, the stronger the mutual influence between the ships. (17) (18) In the formula, Represents the relative velocity vector; Indicates the corresponding ship interaction strength; This indicates the final ship interaction intensity.
[0045] Furthermore, the complexity assessment of inland waterway transportation is based on graph theory, a mathematical method widely used in modeling the influences and processes of complex systems. It can be represented as a pair: (19) In the formula, Represents the set of vertices. The set of edges represents the set of unordered pairs of vertices. An unordered pair can be represented as... When two vertices and via edge When connecting edges, the edges do not have directionality; this type of structure is called an undirected graph.
[0046] In undirected graphs, edges between different vertices are considered equal, ignoring the differences in their mutual influence. To better describe complex systems, researchers have introduced the concept of undirected weighted graphs. In such graphs, each edge is assigned a weight value within the range of 0 to 1. This is used to quantify the degree of influence between adjacent vertices.
[0047] The inland waterway traffic system is modeled as an undirected weighted graph, with ships and their interactions forming the core elements of the graph: ships are extracted as nodes, and the interactions between ship pairs are represented as edges connecting the nodes, thus describing the relationships between them. The interaction strength is calculated based on parameters such as relative distance, meeting angle, and meeting trend, as mentioned earlier. Notably, due to the reciprocity of ship interactions, the traffic graph structure is constructed as an undirected graph.
[0048] By calculating the interaction strengths of all vessels and completing the transformation process, the inland waterway transportation system with vessel interaction characteristics can be fully represented in a graph structure. This system can also be represented using an adjacency matrix, which contains edges between all possible vertex pairs: (20) In the formula, , The horizontal and vertical axes represent the number of vessels in the inland waterway transportation system; the interaction between vessel pairs is the same for both vessels, therefore the adjacency matrix is symmetric. = ,and It equals 0.
[0049] Furthermore, when the intensity of ship interaction When the threshold is exceeded, the corresponding ship pair is marked in the radar image; the complexity of the traffic flow is assessed using two indicators: "weighted map density" and "proportion of high-risk edges". Weighted graph density : ( )(twenty one); High-risk side proportion : (twenty two); In the formula, for The number of sides, (The number of combinations of choosing 2 vertices from l vertices).
[0050] Example 2 In another preferred embodiment, based on Embodiment 1, this embodiment provides a method for evaluating the traffic complexity of inland waterways based on radar sequence images. This embodiment uses a certain inland waterway of the Yangtze River (longitude range 111°30′~111°40′, latitude range 30°20′~30°30′) as the test area. The average daily vessel traffic in this area is approximately 80~120 vessels, including cargo ships, passenger ships, and other types of vessels, resulting in complex traffic scenarios such as vessel intersections and overtaking. The radar equipment used is an X-band radar (resolution 0.01°×10m, sampling interval t=3s). The method of this invention achieves real-time assessment of traffic complexity. The specific implementation process is as follows: I. Experimental Preparation 1. Data Acquisition: Continuously acquire radar image sequences of the test area for 72 hours. Each frame of the image has a pixel size of 1024×1024. The actual distance corresponding to 1 pixel in the image is determined by a conversion factor. This is converted to 0.0017 nautical miles (approximately 3.15 meters); AIS data (vehicle position, speed, and heading) in this area are recorded simultaneously as a verification benchmark.
[0051] 2. Parameter Settings: Based on the test area's channel width (approximately 500 meters), vessel type, and navigation rules, set the key parameters: grayscale area. (Since the radar noise in this area is mainly slight sea clutter, no significant adjustments are needed); minimum safe distance (Approximately 185 meters); the minimum distance at which a vessel must take collision avoidance measures. (Approximately 92.5 meters); the maximum distance at which ship-to-ship interaction occurs. ; Correct parameters The threshold for ship interaction intensity exceeding the limit is set to 0.6 (i.e., ... (Time-marked as high-risk vessels).
[0052] II. Step 1: Implementation of Ship Inspection with Background Deduction Step 1.1 Background Separation: For each frame of radar image, the Gaussian background modeling method is used to separate the target ship from the background components. A background model is established by using 100 consecutive frames of radar images without ship interference. The mean and variance of grayscale of each pixel are calculated. The difference between the current frame image and the background model is calculated to obtain the foreground image containing the ship target.
[0053] Step 1.2, Noise Removal: Apply grayscale regions to the foreground image. Threshold ( The process involves classifying pixels with a grayscale value below 40 as noise and setting them to black, while retaining pixels with a grayscale value ≥ 40 as potential ship targets. After this step, the noise removal rate in the radar image reaches over 85%, and the outlines of ship targets are clearly discernible. For example, after processing, the echoes of five ships that were originally blurry all appear as complete minimum circumcircles, and their center coordinates can be accurately extracted (e.g., the center coordinates of a cargo ship are (320, 450), unit: pixels).
[0054] III. Step 2: Implementation of Ship Tracking Based on Cross-Frame Position Mapping Step 2.1, Adjacent Frame Matching: Using the first... frame( ) and the frame( For example, the first The frame detected 6 ships, with the following locations: p(999, 999) = [(210, 320), (320, 450), (480, 290), (550, 510), (630, 380), (720, 490)] (unit: pixels); Predicting the first using Kalman filtering The frame ship position, the state transition matrix H is set as follows: [1, 0, 3, 0], [0, 1, 0, 3], [0, 0, 1, 0], [0, 0, 0, 1]] (Considering uniform linear motion of the ship, sampling interval 3s) Process noise covariance matrix Let diag([0.1, 0.1, 0.05, 0.05]), and calculate the first value using formulas (2)-(3). Predicted ship position in frame: p'(1000,999)=[(212,323), (323,454), (482,292), (553,513), (632,383), (723,493)].
[0055] Step 2.2, Distance Matching and Number Assignment: The frame actually detected 6 ships, located as follows: p(1000,1000)=[(211,322), (324,455), (481,293), (552,512), (633,384), (722,494)]; By distance threshold (Unit: pixels) Matching detection and prediction results, using the Euclidean distance cost function (Formula 4), the distance between the detected and predicted positions of each ship is less than 2, and the matching is successful; a unique identification number is assigned to each ship (e.g., the cargo ship number at coordinates (320, 450) is Ship-002), and the number remains unchanged in subsequent frame tracking without any jumping phenomenon.
[0056] Steps 2.3 and 3: Position Update: Taking Ship-002 as an example, the observation model B is set to [[1, 0, 0, 0], [0, 1, 0, 0]], and the observation noise covariance matrix is... Let diag([0.05, 0.05]), calculate the Kalman gain G(1000) = [[0.82, 0.03], [0.03, 0.82], [0.15, 0.01], [0.01, 0.15]] using formulas (6)-(7), and then update Ship-002 at the th... Optimal position of the frame: p'(1000, 1000) = (323.6, 454.2), The covariance matrix w(1000, 1000) = diag([0.08, 0.08, 0.04, 0.04]) After the update, the position error with AIS data is less than 5 meters, and the tracking accuracy meets the requirements.
[0057] IV. Step 3: Use dynamic information to determine the implementation of ship interactions Step 3.1, Dynamic Information Calculation: Using Ship-002 (number) and Ship-005 (number, ... Taking frame position (633, 384) as an example, calculate the dynamic information of the two ships: (1) Travel distance, according to formula (9), Ship-002 in the first... to Frame movement distance: d(1000) = 0.0017 × √[(323-320)² + (454-450)²] = 0.0017 × 5 = 0.0085 nautical miles (approximately 15.7 meters); (2) Speed, according to formula (10): v(1000) = 0.0085 / 3 × 3600 / 1852 ≈ 5.1 knots (which is consistent with the normal speed of cargo ships in this area). (3) Course, Ship-002 Frame position (320, 450), the Frame position (323, 454). According to formula (11): h(1000)=arctan[(454-450) / (323-320)]×180 / π+90°=arctan(4 / 3)×180 / π+90°≈53.13°+90°=143.13°; Similarly, the heading of Ship-005 is calculated to be 85.2°.
[0058] (4) Relative distance: Two ships Frame relative distance = 0.0017 × √[(633-323)² + (384-454)²] = 0.0017 × √(310² + (-70)²) = 0.0017 × 318 ≈ 0.54 nautical miles.
[0059] Step 3.2, Quantification of Interactions: (1) The effect of relative distance, according to formula (12):
[0060] (2) Due to the encounter angle, the two ships are heading at 143.13° and 85.2° respectively, and the encounter angle is:
[0061] According to formula (14):
[0062] Because D=0.54 nautical miles is in (0.05 nautical miles) and Between (1 nautical mile), According to formula (13):
[0063] Then, according to formula (15):
[0064] (3) Calculate the relative velocity vector of the two ships when encountering the trend. With relative motion vector The included angle α = 120°, cos(α) = -0.5 < 0, according to formula (16): (The absolute value is then used in subsequent calculations). Step 3.3, Calculation of the interaction function: According to formula (17):
[0065] Because D=0.54 nautical miles is in and Between, according to formula (18):
[0066] The interaction between the two vessels was determined to be low, with no risk of collision.
[0067] Selecting Ship-003 (number, nth frame position (481, 293)) and Ship-004 (number, nth frame position (552, 512)), the calculation is as follows:
[0068] Vessels marked as high-risk require close monitoring.
[0069] V. Step 4: Implementation of Traffic Complexity Assessment Based on Graph Theory Step 4.1, Traffic Graph Construction: Taking frame n as an example, a total of 6 ships were detected in this frame. Construct an undirected weighted graph: (twenty three) The vertex set is as follows: The edge set E represents all ship pairs (a total of C(6,2) = 15 pairs), and the edge weight is the weight of each ship pair. Values (e.g., the edge weight between Ship-003 and Ship-004 is 0.72, and the edge weight between Ship-002 and Ship-005 is 0.34); the adjacency matrix A is a 6×6 matrix, satisfying A[3,4]=A[4,3]=0.72, A[2,5]=A[5,2]=0.34, and the diagonal element A[i,i]=0.
[0070] Step 4.2, Complexity Index Calculation: (1) Weighted graph density, according to the formula: ( )(twenty one); Calculate 15 pairs of ships The sum is 5.82, then ; (2) Proportion of high-risk sides: statistics If the number of edges k=2 (Ship-003 and Ship-004, Ship-001 and Ship-006 respectively), then ; Step 4.3, Result Output: High-risk vessel pairs (Ship-003 and Ship-004, Ship-001 and Ship-006) are marked with red boxes in the radar image. The weighted graph density (0.39) and the proportion of high-risk edges (0.13) are output simultaneously to indicate to maritime regulators that the current traffic complexity in this area is medium and that the navigation dynamics of the marked vessel pairs need to be closely monitored.
[0071] VI. Experimental Verification The evaluation results of the method of this invention are compared with those of manual interpretation and AIS data: 1. Ship detection accuracy: The accuracy rate of radar image ship detection reaches 92% within 72 hours, with a missed detection rate of less than 5% and a false detection rate of less than 3%, which is better than the traditional background subtraction method (accuracy rate of 85%). 2. Tracking stability: The identification number jump rate during ship tracking is 0, the continuous tracking success rate reaches 98%, and the trajectory can still be maintained even in scenarios with 5 ship intersections and 3 brief target obstructions; 3. Accuracy of interactive judgment: The consistency between the judgment and the collision risk reflected by AIS data for high-risk vessels reaches 90%, with no major risk misjudgments or omissions; 4. Real-time performance: The processing time for a single frame image is approximately 0.8 seconds, which meets the requirements for real-time monitoring of inland waterways (requirement ≤ 1 second).
[0072] In summary, the method of this invention demonstrates high accuracy and practicality in the testing of the Yangtze River inland waterway, and can effectively assist maritime regulatory personnel in perceiving traffic conditions and reducing their workload.
[0073] In the preferred embodiment, the grayscale area in step 1 The threshold value range is [30, 50], and in practical applications, it needs to be dynamically adjusted according to the noise intensity of the radar image, with an adjustment range not exceeding ±10. The above settings can ensure effective differentiation between target and background noise in most scenarios. When the noise intensity is low, the threshold can be appropriately lowered to the range of 20~30. If the noise intensity is high, the threshold can be raised to the range of 50~60 to ensure the radar image processing effect.
[0074] In the preferred embodiment, step 2 uses Kalman filtering and Hungarian algorithm to match the ship detection results of adjacent frames. This setting can effectively improve the accuracy and stability of multi-target ship tracking. Kalman filtering can accurately predict the ship's trajectory, and Hungarian algorithm can efficiently complete data association. The combination of the two can cope with situations such as ship occlusion and overlap in complex scenarios.
[0075] In a preferred embodiment, step 2 uses a distance threshold. The system matches detection and prediction results, assigning a unique identification number upon successful matching. These settings effectively prevent number confusion caused by duplicate identification, ensuring that each target corresponds to a unique identifier. At the same time, a reasonable setting of the distance threshold can improve matching accuracy, reduce the risk of misjudgment, provide a reliable basis for subsequent data processing, and ensure the stability and efficiency of the overall system operation.
[0076] In the preferred embodiment, the ship dynamic information (relative distance, speed, and heading) in step 3 is calculated from the ship's position in continuous radar images; the pixel distance is determined by a conversion factor. (0.0017 pixels converted to nautical miles) is mapped to the actual distance; the above settings can ensure the accuracy of ship dynamic information. The speed is calculated based on the position changes and time intervals between adjacent frames, and the heading is determined by connecting consecutive position points; in this way, the ship's dynamics can be grasped in real time and accurately, providing a reliable basis for subsequent decision-making.
[0077] In the preferred embodiment, the ship interaction in step 3 is inversely proportional to the relative distance. This setting ensures enhanced sensitivity of interaction monitoring when ships are close together, enabling timely detection of potential collision risks. When ships are far apart, the monitoring intensity is reduced to avoid unnecessary data processing and improve system operating efficiency and accuracy.
[0078] In the preferred embodiment, the encounter trend of the ships in step 3 is determined by the cosine similarity between the relative velocity vector and the relative motion vector. This setting can accurately quantify the motion correlation between ships and avoid the limitations of simply relying on distance judgment. By dynamically calculating the change of cosine value, the encounter situation can be predicted 0.5 to 1.2 times the ship length in advance, providing a critical reaction window of 8 to 15 seconds for collision avoidance decision-making.
[0079] In a preferred embodiment, step 4 models the ship traffic system as an undirected weighted graph, which can also be represented by an adjacency matrix. It includes edges between all possible vertex pairs; the above settings can clearly present the connection relationships and weight information of each node (such as ports and key points of waterways) in the shipping traffic system; through the values of the adjacency matrix elements, the tightness of traffic connections between different locations can be intuitively judged, providing a convenient data foundation for subsequent analysis.
[0080] In the preferred embodiment, the two indicators for assessing traffic complexity in step 4 are weighted graph density and the proportion of high-risk edges. This setting can more accurately characterize the complexity of the traffic network. Weighted graph density can reflect the tightness of connections between nodes, while the proportion of high-risk edges focuses on potential risks. The combination of the two provides a strong basis for comprehensively assessing traffic complexity and helps optimize subsequent decision-making and processing.
[0081] In summary, this invention provides a method for evaluating the traffic complexity of inland waterways based on radar sequence images, which effectively solves the various shortcomings of current inland waterway traffic complexity assessments and makes it difficult to meet the needs of maritime supervision for real-time and accurate situational awareness.
[0082] In terms of assessment methods, traditional and existing approaches have many limitations. On the one hand, manual interpretation has significant limitations. Traditional supervision relies on maritime inspectors manually observing radar images and AIS data, which is highly subjective, labor-intensive, and lacks real-time capability. Especially in densely populated areas, maritime inspectors struggle to identify potential collision risks in a timely manner, posing a significant threat to waterway safety. On the other hand, automated methods have many shortcomings. Most existing automated assessment methods are based on static traffic parameters such as vessel density and speed distribution, lacking continuous analysis of vessel trajectories and failing to reflect the complex and ever-changing waterway situation. In practical applications, some computer vision technologies are affected by radar image noise (such as sea clutter and rain / snow reflections), resulting in unstable vessel detection accuracy. Moreover, vessel tracking algorithms are prone to identification number jumps in scenarios with target occlusion and dense intersections, making it impossible to construct reliable trajectories. Furthermore, interactive assessment is incomplete. Most methods only assess risk based on vessel spacing, failing to comprehensively consider dynamic interactive factors such as encounter angles and relative speeds, thus failing to meet the refined analysis requirements of the International Regulations for Preventing Collisions at Sea (ICP-5) for complex encounter scenarios. In the field of traffic complexity modeling, existing technologies mostly rely on historical data and computationally complex statistical methods (such as entropy methods and cluster analysis), making it difficult to achieve real-time assessment. Furthermore, graph theory applications often ignore the dynamic changes in ship interaction intensity, constructing topological graphs based solely on binary relationships (existence / non-existence of interaction), resulting in overly simplistic assessment results.
[0083] To address the aforementioned problems, this invention achieves several technological innovations. Firstly, in ship detection, it is the first to apply grayscale region thresholding and background subtraction techniques to ship detection in radar sequence images. This breaks through the dependence of traditional threshold segmentation on fixed noise patterns, achieving stable target separation in dynamic noise environments and significantly improving ship detection accuracy, especially in complex noise environments, representing a technological breakthrough. Secondly, in ship tracking, it innovatively proposes a cross-frame ship position matching method based on Kalman filtering and the Hungarian algorithm. It solves the problem of identification number jumps in target occlusion and dense intersection scenarios through the Euclidean distance cost function, constructing continuous motion trajectories. The target tracking problem is transformed into a combinatorial optimization problem. By combining Kalman filter prediction with optimal allocation using the Hungarian algorithm, it creatively solves the problem of ship identification number jumps in dense scenes, providing a stable data foundation for real-time traffic situation awareness.
[0084] Furthermore, in terms of interaction quantification, this study is the first to integrate the three elements of relative distance, encounter angle, and encounter trend to establish a dynamic interaction quantification model that conforms to the International Maritime Collision Prevention Regulations (IMDG), breaking through the limitations of traditional methods that rely solely on a single distance parameter. It is the first to transform the dynamic interaction elements (relative distance, encounter angle, and encounter trend) in the IMDG GRD into a quantifiable mathematical model, achieving interaction intensity normalization through nonlinear functions, providing technical support for the analysis of complex encounter scenarios. Finally, regarding the evaluation system, a graph theory-based undirected weighted graph evaluation system is creatively constructed. It uses the symmetry of adjacency matrices to express the reciprocity of ship interactions, achieving visualization of the topological structure of traffic complexity, filling the gap in existing technologies for describing the dynamic relationship network of waterways. This innovative approach proposes using a graph theory model to describe waterway traffic complexity, visually displaying the interaction network between ships through an undirected weighted graph structure, and combining adjacency matrix symmetry to achieve dynamic complexity evaluation, forming a complete technical chain of "detection-tracking-interaction-evaluation," demonstrating significant technological advancement.
Claims
1. A method for evaluating the traffic complexity of inland waterways based on radar sequence images, characterized in that, Includes the following steps: Step 1, Ship Detection with Background Subtraction: Separate the target ship and background components in the radar image, and detect them by grayscale areas. Thresholding removes some radar noise interference; Step 2, Ship tracking based on cross-frame position mapping: Match the previous tracking results with the current detection results, and correctly assign the ship identification number as a cost function using Euclidean distance; Step 3: Determine ship interactions using dynamic information: Calculate attributes including the relative distance between ship pairs, encounter angle, and encounter trend using the ship's speed and heading, and then determine ship interactions by integrating these attributes; Step 4: Graph-based traffic complexity assessment: Construct a traffic graph with ship interaction degree. When ship interaction exceeds a threshold, mark ship pairs in the radar image and use two indicators to assess traffic complexity.
2. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 1, characterized in that: grayscale area in step 1 The threshold value ranges from [30, 50], and in practical applications, it needs to be dynamically adjusted according to the noise intensity of the radar image, with the adjustment range not exceeding ±10.
3. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 1, characterized in that: In step 2, Kalman filtering and the Hungarian algorithm are used to match the ship detection results of adjacent frames; The frame's ship position is represented as follows: (1); The current frame's predicted ship position and its uncertainty probability distribution are as follows: (2); (3); In the formula, Indicates the predicted position of the ship; It is the covariance matrix of the prior prediction; It is achieved by integrating the detected ship positions with the predicted positions, in the... The optimal ship position obtained from the frame; It is the first The covariance matrix corresponding to the optimal ship position obtained in the frame; and Let these represent the state transition matrix and its transpose, respectively. It is the covariance matrix of the process noise; Indicates the number of frames; , These are the horizontal and vertical coordinates of the ship's position in the radar image, respectively.
4. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 3, characterized in that: In step 2, the distance threshold is used. The matching detection and prediction results are expressed in Euclidean distance cost function as shown in formula (4); after a successful match, a unique identification number is assigned, and the ship position is updated using Kalman gain as shown in formula (5). The Kalman gain and the uncertainty probability distribution after the update are shown in formulas (6) to (8): (4); (5); (6); (7); (8); In the formula, This represents the Euclidean distance between the detection result and the prediction result. This indicates the location of the vessel being inspected. It is the posterior vector of the ship's position. Indicates the prefit residuals. Represents the inverse of the observation error covariance matrix; It is Kalman gain. To predict the covariance matrix of the posterior matrix, An observational model representing ship motion. It is the covariance matrix of the observation noise. The transpose matrix of the observation model representing the ship's motion.
5. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 1, characterized in that, In step 3, the ship dynamic information is calculated from the ship's position in continuous radar images; pixel distance is determined by a conversion factor. The actual distance is mapped to the ship's travel distance, speed, and heading, which are calculated using formulas (9) to (11): (9); (10); (11); In the formula, This indicates the distance traveled by the same ship within adjacent time intervals. and Representing the first The ship's position coordinates in a frame radar image. and They represent the first The ship's position coordinates in the frame, Indicates speed, Indicates the course. The sampling interval is denoted as .
6. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 5, characterized in that: In step 3, the ship interaction is inversely proportional to the relative distance, and the interaction forces affected by the relative distance are expressed as follows: (12); In the formula, This represents the interaction forces between ships affected by relative distance. The relative distance between the two ships. This is the minimum safe distance under current traffic conditions; To correct the parameters.
7. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 6, characterized in that, Ship interaction in step 3 Relative distance adjustment function Interactions affected by encounter angle They are represented as follows: (13); (14); (15); In the formula, The encounter angle between the two ships indicates the angle of encounter between the ships. With ships The relative distance between them; It is the minimum distance at which a ship must take collision avoidance measures; It is the maximum distance at which ship-to-ship interactions occur.
8. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 7, characterized in that, In step 3, the ship's encounter trend is determined by the cosine similarity between the relative velocity vector and the relative motion vector, and the interaction influenced by the trend... The ship interaction function is expressed by formula (17) and (18), using formula (16): (16); (17); (18); In the formula, Represents the relative velocity vector; Indicates the corresponding ship interaction strength; This indicates the final ship interaction intensity.
9. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 1, characterized in that, In step 4, the traffic system is modeled as an undirected weighted graph. , represented as: (19); In the formula, Represents the set of vertices. Represents the set of edges; Traffic system modeling can also be represented using an adjacency matrix, which... The edges containing all possible pairs of vertices are expressed as follows: (20); In the formula, , The horizontal and vertical coordinates represent the number of ships in the inland waterway transportation system.
10. The method for evaluating the traffic complexity of inland waterways based on radar sequence images according to claim 1, characterized in that, In step 4, the two metrics for evaluating traffic complexity are weighted graph density and the proportion of high-risk edges, which are expressed as follows: Weighted graph density : ( )(21); High-risk side proportion : (22); In the formula, for The number of sides; , which represents the number of combinations of choosing 2 vertices from l vertices.
Citation Information
Patent Citations
Sea traffic flow complexity acquisition method based on field theory
CN116227780A
Target water area ship traffic flow complexity analysis method, system and device and storage medium
CN120279760A
