An automatic collision avoidance emergency takeover control method and system based on multi-source sensor fusion for a ship

Abstract

The application provides a ship automatic collision avoidance emergency takeover control method based on multi-source sensing fusion, comprising the following steps: S1, a multi-sensor space-time linkage trigger mechanism is established to realize cooperative detection and cross verification; S2, three-level progressive target type identification is completed; S3, a hierarchical collision avoidance strategy matrix is called to output a differentiated avoidance scheme; S4, a conflict resolution weight model is used to generate an optimal comprehensive control instruction set; S5, two types of emergency classification responses, i.e., self-ship failure and external threat, are distinguished; S6, a transfer coefficient is relied on to realize smooth control weight transfer from automatic to manual; and S7, under manual control, continuous monitoring and activation of bottom protection are realized. The application also simultaneously provides a matching hardware control system, which realizes the functions of sensing, research and judgment, decision-making, emergency and man-machine cooperation corresponding to each step, realizes safe and efficient control of ship automatic collision avoidance and emergency takeover, and greatly improves the ship navigation safety redundancy under complex navigation environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent ship navigation safety control technology, specifically to a ship automatic collision avoidance emergency takeover control method and system based on multi-source sensor fusion. Background Technology

[0002] With the advancement of intelligent upgrades in the shipping industry, the density of ship traffic continues to increase, and the navigation environment in nearshore areas, waterways, and ports is becoming increasingly complex. Safety accidents such as ship collisions, groundings, and encounters with obstacles are occurring frequently, leading to increasingly higher requirements for the intelligence and precision of ship autonomous collision avoidance and emergency safety redundancy control. Intelligent ship autopilot and assisted collision avoidance technologies are the core technological support for ensuring navigation safety in complex waters.

[0003] Existing intelligent collision avoidance technologies for ships suffer from numerous substantial technical defects, failing to meet the safety management requirements of complex navigation scenarios. A typical existing technical solution, such as the intelligent driving assistance system for ships (publication number CN202511571191.1), relies on multi-source information collection and unified data processing to assist decision-making. However, various sensors collect data independently and passively aggregate it, lacking an active spatiotemporal linkage triggering mechanism. When a single sensor detects an anomaly, it cannot link with other devices for cross-verification, making it highly susceptible to interference from complex maritime environments such as rain, fog, wind, waves, and light, resulting in insufficient detection accuracy and a high number of false targets. Furthermore, this type of solution has a limited emergency control mode, only supporting one-button automatic control cutoff via physical buttons. This is a switch-type abrupt takeover mode without a transition buffer mechanism, easily leading to control vacuums and sudden changes in ship attitude during control switching, posing significant navigational safety hazards.

[0004] The multi-scale collaborative target detection system with publication number CN202511120622.2 can realize multi-dimensional detection and identification of water targets. However, this solution only focuses on the perception of water targets and lacks the ability to fuse the detection of risks such as underwater terrain, obstacles, and shoals, resulting in a one-sided perception dimension. Furthermore, it only completes the target detection function and does not have the supporting logic of refined collision avoidance decision-making, external interaction, and emergency management throughout the entire process, thus failing to form a closed-loop safety control system.

[0005] In addition to the aforementioned existing technological deficiencies, current mainstream ship collision avoidance control systems generally suffer from three major shortcomings: First, the emergency takeover mechanism is crude, failing to distinguish between two fundamentally different emergency scenarios: shipboard equipment failure and external target threats. The risk causes, handling logic, and safety requirements of these two scenarios differ greatly, and a unified handling model cannot adapt to the needs of refined safety management. Second, the control switching mechanism is imperfect, mostly adopting a binary switching mode of fully automatic and fully manual operation without a gradual transition process. Conflicts between human and machine control commands can easily lead to ship instability. Third, there is no safety fallback after manual takeover. The system exits control directly after transferring control, making it impossible to monitor risk deviations from manual operation and losing the last line of safety redundancy.

[0006] In summary, existing ship collision avoidance technologies suffer from several technical shortcomings, including poor sensor fusion and coordination, low target recognition precision, simplistic collision avoidance strategies, lack of scenario-based emergency response logic, insufficient safety during human-machine interface switching, and lack of manual backup protection. These deficiencies make it difficult to meet the practical needs of intelligent collision avoidance and emergency safety management in complex waters. Therefore, this invention proposes a ship automatic collision avoidance emergency takeover control method and system based on multi-source sensor fusion. Summary of the Invention

[0007] The purpose of this invention is to solve the technical problems in the prior art, such as weak multi-source sensor collaborative detection capability, low target recognition accuracy, lack of differentiation in collision avoidance strategies, crude handling of emergency scenarios, safety hazards in human-machine control switching, and lack of manual operation as a backup protection. The invention provides a ship automatic collision avoidance emergency takeover control method and system based on multi-source sensor fusion.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for emergency takeover control of ship automatic collision avoidance based on multi-source sensor fusion, comprising the following steps: S1: The multi-source sensor acquisition module continuously collects information on the ship's navigation status, the dynamic and static environment of surrounding targets, and underwater topography. It performs spatiotemporal alignment, noise filtering, and outlier removal preprocessing on the multi-source heterogeneous sensor data. Based on the ship navigation perception scenario adaptation rules, a multi-sensor spatiotemporal correlation triggering mechanism is established. When any sensor's monitoring data shows a characteristic abnormal change, the associated sensors are linked to conduct collaborative detection and data cross-verification in the same spatiotemporal area, avoiding detection errors of a single sensor and environmental interference. S2: Based on the effective data and confidence assessment results after multi-sensor cross-validation, a hierarchical progressive identification logic is adopted to complete the three-level target identification operation in sequence: target dynamic and static attribute determination, target category identification, and target fine attribute and navigation status extraction. It distinguishes the feature differences of different types of targets on the water and underwater, and completes the fine classification of maritime targets. S3: Combining the refined target identification results with the relative distance between the ship and the target, retrieve the pre-constructed target type-risk distance hierarchical collision avoidance strategy matrix, and match differentiated and scenario-adaptive collision avoidance execution strategies according to different target types and different risk distance levels; S4: Construct a multi-dimensional conflict resolution weight model, select the optimal execution scheme from multiple candidate collision avoidance schemes, integrate multi-dimensional operation instructions of ship motion control, external communication, and audible and visual warnings, generate a comprehensive collision avoidance operation instruction set adapted to the current navigation scenario, and issue it for execution; S5: Throughout the entire process of collision avoidance decision-making and control execution, monitor the operating status of ship equipment and external navigation risks in real time, distinguish between two different emergency scenarios: ship equipment failure and external target threats, and implement a graded and progressive emergency response mechanism. S6: When the system determines that human crew intervention is required, it achieves a smooth integration of automatic control commands and manual control commands based on the dynamic control transfer coefficient, completing the gradual transfer of control from the control system to manual control and avoiding the control switching vacuum period. S7: After control is fully transferred to the human crew, the system continuously monitors the matching degree between the ship's navigation status and the safe navigation envelope. When the control deviation exceeds the limit and there is no manual correction action, the auxiliary correction and fallback safety protection mechanism is automatically triggered.

[0009] Furthermore, in step S1, the multi-sensor spatiotemporal correlation triggering mechanism adapts to the detection needs of ships in all scenarios above and below water. It includes five types of triggering logic: visual sensor linkage, radar linkage, AIS data linkage, lidar linkage, and underwater sonar linkage. Each linkage mechanism operates independently and complements the others, achieving a collaborative detection effect of "single device detection, full system verification," effectively solving the problem of false alarms and missed alarms from single sensors in complex maritime environments. The data timing synchronization error is controlled within the preferred range of 5-15ms to ensure the spatiotemporal consistency of multi-source sensor data, adapting to the real-time perception needs of ships at high speeds.

[0010] Furthermore, the three-level progressive target recognition logic in step S2 imposes progressive constraints and increases accuracy at each level. It first filters out false targets in the environment through dynamic and static attributes, then classifies them into broad categories, and finally refines the target-specific attributes. This adapts to the diverse types of targets, numerous interferences, and significant morphological differences encountered at sea, greatly improving target recognition accuracy. The target dynamic monitoring window is set to an adjustable range of 20–40 seconds, which can be adaptively adjusted according to the complexity of sea conditions.

[0011] Furthermore, in step S3, the target type-risk distance layered collision avoidance strategy matrix is ​​constructed by combining international maritime collision avoidance rules with actual ship operation experience. It matches exclusive strategies for different targets and different risk distances, breaking the traditional uniform collision avoidance mode and realizing refined and differentiated management of collision avoidance operations.

[0012] Furthermore, in step S4, the multi-dimensional conflict resolution weight model comprehensively considers four core dimensions: collision avoidance effect, control cost, response speed, and communication adaptability. It adapts to different navigation scenarios through dynamic weight coefficients, avoiding the control limitations caused by selecting the best from a single dimension and ensuring the scenario adaptability of the optimal solution. The sum of all weight coefficients is normalized to 1, allowing for flexible fine-tuning of the ratios according to different scenarios such as open waters, restricted channels, and near-shore operations.

[0013] Furthermore, in step S5, a dual-scenario differentiated emergency response mechanism is implemented to address two completely different types of navigation risks: ship malfunction and external threats. The malfunction scenario focuses on emergency support for ship loss of control, while the threat scenario focuses on proactive collision avoidance risk management. The tiered and progressive response avoids over-management or delayed management.

[0014] Furthermore, in step S6, the dynamic control transfer coefficient adaptively adjusts the switching rate according to the human-machine operation matching degree, solving the control vacuum problem of traditional binary switching, realizing a smooth transition of human-machine co-driving, and adapting to the motion characteristics of ships with large inertia and slow response.

[0015] Furthermore, in step S7, the manual control fallback protection mechanism continuously empowers the system's safety monitoring capabilities after manual takeover, filling the safety gap of the existing technology where the system is out of control after manual takeover, and constructing a full-process safety redundancy system of "automatic control - human-machine switching - manual control fallback".

[0016] This invention also provides a ship automatic collision avoidance emergency takeover control system based on multi-source sensor fusion, used to implement the above-mentioned control method. It includes: a multi-source sensor acquisition module, a data preprocessing and spatiotemporal synchronization module, a sensor spatiotemporal linkage triggering module, a multi-level fusion and comprehensive analysis module, a layered collision avoidance strategy module, a collision avoidance decision and comprehensive execution module, a dual-scenario emergency takeover module, a gradual transfer of control and intelligent feedback module, a human-machine interaction and comprehensive display module, and a data recording and playback module. The multi-source sensor acquisition module integrates various surface and underwater sensing devices to provide raw data for the entire process control. The data preprocessing and spatiotemporal synchronization module achieves unified calibration of multi-source heterogeneous data. The system includes: time-series alignment; a sensor spatiotemporal linkage triggering module that stores multiple linkage rules to achieve cross-sensor collaborative detection; a multi-level fusion and comprehensive judgment module that completes data verification and three-level target identification; a hierarchical collision avoidance strategy module that stores a differentiated collision avoidance strategy matrix; a collision avoidance decision and comprehensive execution module that completes optimal solution selection and multi-dimensional command output; a dual-scenario emergency takeover module that realizes dual-scenario hierarchical emergency response; a gradual transfer of control and intelligent feedback module that realizes smooth control transfer and manual backup protection; a human-computer interaction and comprehensive display module that realizes situation visualization and human-computer interaction; and a data recording and playback module that completes full-process data storage and traceability playback.

[0017] This invention has the following advantages:

[0018] 1. This invention constructs a multi-sensor spatiotemporal correlation active linkage triggering mechanism, breaking the existing technology of passive data aggregation of sensors, realizing closed-loop detection of "single device anomaly, whole system collaborative verification", which greatly improves the accuracy and stability of target detection in complex marine environments and solves the technical problem of single sensors being susceptible to environmental interference, false alarms and missed alarms.

[0019] 2. This invention adopts a three-level progressive refined target recognition logic, which filters interference and refines target attributes step by step. It can accurately distinguish multiple types of targets such as ships, small boats, marine life, and floating objects, and is specifically adapted to the collision avoidance needs of different targets, taking into account both navigation safety and marine ecological protection, and adapting to navigation needs in multiple scenarios.

[0020] 3. This invention constructs a two-dimensional hierarchical collision avoidance strategy matrix based on target type and risk distance, and combines scenario-differentiated collision avoidance schemes. It abandons the traditional unified collision avoidance mode and achieves hierarchical control of long-distance early warning, mid-distance assessment, and close-range handling. The collision avoidance strategy is more in line with actual navigation norms.

[0021] 4. This invention designs a multi-dimensional conflict resolution dynamic weight model, which adaptively adjusts the weight coefficients in combination with the navigation scenario, taking into account collision avoidance safety, maneuverability, response timeliness and communication compliance, ensuring the scenario adaptability of the optimal collision avoidance solution, and solving the limitations of the traditional single decision dimension.

[0022] 5. This invention innovatively distinguishes between two emergency scenarios: self-propelled equipment failure and external target threats, enabling differentiated and graded responses and accurately matching the handling logic for the two types of risks. This solves the problems of the existing emergency response models being crude and poorly adaptable.

[0023] 6. This invention achieves a smooth and gradual transfer of human-machine control through dynamic coefficients, adapting to the large inertial motion characteristics of ships, completely eliminating the control vacuum and attitude change risks of traditional binary switching, and improving the safety and stability of human-machine co-driving.

[0024] 7. This invention adds a manual control fallback protection mechanism, which continuously monitors and assists in correction after manual takeover, filling the safety gap of the existing technology where the system is out of control after manual takeover, and constructing a full-process, full-link safety redundancy system, which greatly improves the safety level of intelligent navigation of ships. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method described in this invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the technical solution of the present invention, the specific embodiments of the present invention will be described in detail and completely below in conjunction with different application scenarios.

[0027] Example 1 This embodiment is applied to the scenario of ocean-going cargo ships sailing in open waters across the ocean. The ship is equipped with a full set of multi-source sensing equipment, including X-band navigation radar, solid-state lidar, Class A AIS receiver, high-definition visible light and infrared cameras, underwater acoustic detection equipment, multiple sonar detection systems, GPS / IMU integrated navigation and hydro-meteorological monitoring equipment, which is suitable for the safety management needs of complex sea conditions and long-distance navigation in the open ocean.

[0028] In the S1 perception fusion phase, the system continuously collects data on surface navigation targets, underwater topography, hydrology and meteorology, and the ship's navigation status through multi-dimensional sensing devices. The data timing synchronization error is controlled within the range of 5-15ms. After noise filtering and abnormal data removal, a multi-sensor spatiotemporal linkage triggering mechanism is activated. When the visual sensor detects a distant target continuously approaching, it automatically links with radar and lidar for ranging verification and simultaneously retrieves AIS messages to match target information. When the underwater sonar detects shallow waters or obstacles, it links with underwater sensing devices to complete real-time verification of the water area, achieving collaborative perception both above and below water and effectively avoiding detection interference caused by ocean waves and water mist.

[0029] In the S2 target identification phase, the system collects target position change data through an adjustable dynamic monitoring window of 20-40 seconds, and accurately distinguishes between moving and stationary targets by combining water tidal drift and wind and wave interference parameters; then, it integrates multimodal sensor features to complete the target category determination. For identified ship targets, it further extracts ship size, navigation status, and identity information; for common ocean-going marine life such as whales and dolphins, it completes fine classification through body shape, out-of-water movement trajectory, and back features, providing accurate data support for eco-friendly collision avoidance.

[0030] In the S3 strategy matching and S4 decision selection stages, the ocean collision avoidance strategy is matched strictly according to the 3-nautical-mile, 1-3 nautical-mile, and 1-nautical-mile stratified standards defined in the claims. In medium-distance risk scenarios, the collision risk threshold is selected from the optimal value in the range of 0.3 to 0.5 to trigger the pre-collision coordination mechanism. When selecting the optimal value through the four-factor weight model, the value of λ1 is preferentially increased in open water navigation scenarios to ensure collision avoidance safety, and a comprehensive instruction set including course fine-tuning, VHF remote coordination, and light warning is generated.

[0031] During the S5 emergency response, S6 control handover, and S7 backup protection phases, the highest priority for emergency response to equipment failures during ocean voyages is given. When a failure of core equipment is detected, the highest level of emergency takeover is immediately triggered, and information about the ship's loss of control and distress is broadcast to the outside world. When manual intervention is required, the α coefficient is dynamically adjusted strictly according to the human-machine control matching degree. Control is transferred quickly when it is consistent and slow when there is a conflict, and a second confirmation is required. After manual takeover, the ship's safety status is monitored in real time at the preferred monitoring frequency of 10-15Hz. When parameters deviate from the limit and there is no manual correction action, a small auxiliary correction torque is added to ensure safety redundancy for long-distance ocean voyages.

[0032] Example 2 This embodiment is applied to scenarios involving narrow inland waterways and port navigation. Inland waterways have high vessel traffic density, narrow channels, limited water depth, and a large number of small vessels and floating objects. Therefore, the sensor configuration is simplified, retaining core equipment such as radar, AIS, visible light camera, and shallow water detection sonar, to meet the lightweight and highly adaptable management requirements of inland waterway scenarios.

[0033] This embodiment adapts to inland waterway scenarios by adjusting the risk distance levels, setting the long-range warning layer to greater than 2km, the medium-range assessment layer to 0.5-2km, and the close-range action layer to less than 0.5km, aligning with the characteristics of short-distance navigation on inland waterways. Target identification focuses on identifying small fishing boats, cargo ships, floating objects in the river channel, and obstacles in shallow waters, while filtering out false targets such as fixed facilities on the shore. The dynamic monitoring window defaults to a short duration of 20-30 seconds, adapting to the maneuverable and changeable navigation characteristics of inland waterway targets.

[0034] During the decision-making and optimization phase, the weight coefficients of λ2 and λ3 are prioritized and increased in restricted waterway scenarios to ensure maneuverability and response timeliness, and to prevent the risk of grounding or collision caused by large-angle or large-amplitude maneuvers. Emergency response focuses on high-frequency external threats such as sudden crossings by small vessels at close range and obstructions from floating debris in the waterway, rapidly triggering level two and three collision avoidance responses within the collision risk threshold range of 0.5–0.7. During the human-machine switching phase, in complex and dynamic inland waterway navigation scenarios, the growth rate of the α coefficient is slowed down, and a mandatory secondary confirmation mechanism for human-machine commands is added to mitigate the risk of human error. The fallback protection mechanism narrows the deviation thresholds for course and speed to meet the high-precision navigation requirements of narrow waterways, with the deviation judgment thresholds set within a stringent range of 12–15° and 2–3 knots.

[0035] Example 3 This embodiment is applied to nearshore fishing waters where marine life is dense, small vessels are mixed, and there are many underwater reefs and obstacles. The system focuses on enhancing underwater detection and marine life identification, ecological collision avoidance capabilities, and prioritizes the use of sonar linkage, biometric identification, and sound wave repelling functions.

[0036] Phase S1 focuses on underwater risk perception using underwater sonar and ultrasonic detection equipment, triggering collaborative verification of surface equipment. Phase S2 prioritizes the identification of marine life targets such as whales, dolphins, and sharks, distinguishing between species and movement status, and employing a 30-40 second long-term monitoring window to improve biometric identification accuracy. Phases S3 and S4 target marine life targets, prioritizing strategies such as slowing down, using sound waves to drive them away, and navigating around them at small angles to avoid harm to marine life from high-speed maneuvers, achieving eco-friendly collision avoidance. The risk assessment threshold is selected from the median of the range, balancing safety and ecological protection needs. Emergency scenarios focus on responding to the risks of sudden underwater obstacles and close-range biological clusters, quickly triggering collision avoidance responses. The manual control fallback mechanism focuses on monitoring the ship's low-speed navigation status, continuously monitoring at a low frequency of 8-12 Hz to ensure the safety of both the ship and marine life during fishing operations.

[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention; all equivalent substitutions, structural modifications and process optimizations made based on the concept and principle of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A method for emergency takeover control of automatic collision avoidance on ships based on multi-source sensor fusion, characterized in that, Includes the following steps: S1: The multi-source sensor acquisition module continuously collects information on the ship's navigation status, the dynamic and static environment of surrounding targets, and underwater topography. It performs spatiotemporal alignment, noise filtering, and outlier removal preprocessing on the multi-source heterogeneous sensor data. Based on the ship navigation perception scenario adaptation rules, a multi-sensor spatiotemporal correlation triggering mechanism is established. When any sensor's monitoring data shows a characteristic abnormal change, the associated sensors are linked to conduct collaborative detection and data cross-verification in the same spatiotemporal area, avoiding detection errors of a single sensor and environmental interference. S2: Based on the effective data and confidence assessment results after multi-sensor cross-validation, a hierarchical progressive identification logic is adopted to complete the three-level target identification operation in sequence: target dynamic and static attribute determination, target category identification, and target fine attribute and navigation status extraction. It distinguishes the feature differences of different types of targets on the water and underwater, and completes the fine classification of maritime targets. S3: Combining the refined target identification results with the relative distance between the ship and the target, retrieve the pre-constructed target type-risk distance hierarchical collision avoidance strategy matrix, and match differentiated and scenario-adaptive collision avoidance execution strategies according to different target types and different risk distance levels; S4: Construct a multi-dimensional conflict resolution weight model, select the optimal execution scheme from multiple candidate collision avoidance schemes, integrate multi-dimensional operation instructions of ship motion control, external communication, and audible and visual warnings, generate a comprehensive collision avoidance operation instruction set adapted to the current navigation scenario, and issue it for execution; S5: Throughout the entire process of collision avoidance decision-making and control execution, monitor the operating status of ship equipment and external navigation risks in real time, distinguish between two different emergency scenarios: ship equipment failure and external target threats, and implement a graded and progressive emergency response mechanism. S6: When the system determines that human crew intervention is required, it achieves a smooth integration of automatic control commands and manual control commands based on the dynamic control transfer coefficient, completing the gradual transfer of control from the control system to manual control and avoiding the control switching vacuum period. S7: After control is fully transferred to the human crew, the system continuously monitors the matching degree between the ship's navigation status and the safe navigation envelope. When the control deviation exceeds the limit and there is no manual correction action, the auxiliary correction and fallback safety protection mechanism is automatically triggered.

2. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 1, characterized in that, In step S1, the multi-sensor spatiotemporal correlation triggering mechanism adapts to the detection needs of ships in all scenarios above and below water. It includes five types of triggering logic: visual sensing linkage, radar linkage, AIS data linkage, lidar linkage and underwater sonar linkage. Each linkage mechanism operates independently and complements the others. Visual sensing linkage logic: When the visual sensor detects that the target image size is continuously increasing and the target position is regularly displaced for multiple consecutive frames, it will link with the radar and lidar to complete the accurate calculation of the target distance, and at the same time retrieve the message data of the AIS receiver to complete the target identity matching and verification. Radar linkage logic: When the radar detects a new echo, the echo intensity continues to increase, or the Doppler frequency shift changes abruptly, it links with vision and infrared sensing devices to complete target image acquisition and shape recognition, and works with lidar to complete ranging calibration. AIS data linkage logic: When the AIS receiver receives a new ship message or when the target's heading or speed changes drastically, the linkage radar completes the target echo verification and the linkage vision equipment completes the real-scene image confirmation. LiDAR linkage logic: When the LiDAR detects new target point cloud data at medium to close range, it links with the vision device to identify the target shape and works with the acoustic detection device to determine whether the target is a marine organism. Underwater sonar linkage logic: When the scanning sonar and multibeam echo sounder detect underwater obstacles or shallow water risks, the underwater sensing equipment is linked to complete the real-world verification of the water area. Combined with the water depth data of the electronic nautical chart, the risk cross-verification is completed, realizing the collaborative perception of risks above and below water.

3. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 2, characterized in that, The three-level progressive target recognition logic in step S2 is adapted to complex interference scenarios at sea, with hierarchical constraints and progressively improved accuracy. The first level is the dynamic and static attribute determination: based on the change in the target position within a continuous time window, combined with the ship's speed, heading and tidal environment parameters, it distinguishes between static fixed targets and dynamic moving targets, and filters out false targets caused by environmental interference such as waves and light and shadow. The second level is target category identification: by integrating visual image features, radar echo morphology, lidar point cloud contours, and AIS data label multimodal information, a multimodal feature fusion classifier trained to classify dynamic targets into four categories: ships, small boats, large marine life, and floating objects. The third level is refined attribute extraction: for ship targets, refined parameters such as ship name, ship size, navigation status, and call sign are extracted; for large marine organism targets, specific category determination is completed by combining body size, movement trajectory, and biological characteristics; for floating objects targets, artificial floating objects and natural floating objects are distinguished to provide data support for subsequent differentiated collision avoidance.

4. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 3, characterized in that, In step S3, the target type-risk distance hierarchical collision avoidance strategy matrix uses target type as the row dimension and risk distance level as the column dimension. It divides the risk range into three levels based on ship navigation safety regulations and actual operation experience, and executes differentiated control logic in each range. Long-range warning layer: when the distance is greater than 3 nautical miles, the system continuously tracks the target situation and updates navigation parameters, only performs situation monitoring and risk prediction, does not perform active collision avoidance operations, and maintains navigation watch status; Mid-range assessment layer: at a distance of 1 to 3 nautical miles, the system initiates a refined risk assessment, calculates the nearest encounter distance and the nearest encounter time, and generates a collision avoidance plan when the collision risk index reaches a preset threshold in the range of 0.3 to 0.5, and completes external coordination and early warning through wireless communication and light signals; Close-range action layer: When the distance is less than 1 nautical mile, the system automatically triggers active collision avoidance control, and performs control actions such as steering, deceleration, and stopping in combination with the target type, while continuously publicizing the collision avoidance intention to the outside world through communication and audio-visual equipment.

5. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 4, characterized in that, In step S4, the integrated collision avoidance operation command set realizes the integrated linkage of ship control, communication, and audible and visual warnings, specifically including ship motion control commands, external radio communication commands, optical warning commands, and acoustic warning commands; The vessel's motion control commands include refined control parameters such as target heading angle, target speed, turning rate, and deceleration gradient; external radio communication commands include VHF collision avoidance coordination communication, AIS navigation status broadcast, and digital selective emergency call; optical warning commands include compliant Morse code light signals and close-range searchlight warnings; acoustic warning commands include compliant shipboard sounding; and for marine life targets, an additional dedicated low-frequency acoustic repellency command is configured to achieve eco-friendly collision avoidance.

6. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 5, characterized in that, In step S4, the conflict resolution weight model adopts a multi-dimensional weighted fusion algorithm, and the formula is: W = λ1·R_resolve + λ2·E_maneuver + λ3·T_response + λ4·C_communication; Among them, R_resolve is the conflict resolution rate, representing the ability of the scheme to mitigate the risk of multi-target collisions; E_maneuver is the maneuverability score, representing the energy consumption of ship maneuvering and the cost of course deviation; T_response is the response time score, representing the execution response speed of the scheme; C_communication is the communication coordination score, representing the compliance and effectiveness of external interactions; λ1, λ2, λ3, and λ4 are dynamic weight coefficients, which are normalized and summed to 1, and the value range of each coefficient is 0.1 to 0.7 to avoid extreme proportions of single-dimensional weights; in open waters, the value of λ1 is increased first, and in restricted channels, the values ​​of λ2 and λ3 are increased first, dynamically and adaptively adjusted according to the actual navigation scenario. The scheme corresponding to the maximum value of weight W is selected as the optimal collision avoidance scheme. Each weight coefficient is adaptively matched according to the navigation scenario, and the sum of typical value ranges is normalized, which can be flexibly fine-tuned according to sea state, channel width, and ship tonnage.

7. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 6, characterized in that, In step S5, the judgment conditions and response logic for the two types of emergency scenarios are set differently to adapt to different ship fault and risk scenarios. Self-propelled equipment failure emergency: When critical faults such as abnormal steering response, main engine propulsion failure, loss of core sensor data, or abnormal power supply are detected, the highest priority emergency takeover is triggered, automatically executing steering and speed maintenance control, and simultaneously broadcasting the loss of control and distress status to the outside world, requiring the crew to complete manual takeover within a time limit. External target threat emergency response: When three conditions are met simultaneously, the target collision risk index reaches the preset range of 0.5 to 0.7, the remaining collision time is within the preset duration of 2 to 5 minutes, and the system continuously and stably tracks the target for no less than 8 to 12 seconds, a progressive three-level response is triggered. The first level response is situation warning and continuous tracking and monitoring; the second level response is pre-execution of the collision avoidance plan and public announcement of the collision avoidance intention. A Level 3 response involves emergency deceleration, maximum safe rudder angle deflection, and simultaneous triggering of a mandatory takeover alarm to prompt crew intervention.

8. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 7, characterized in that, In step S6, the formula for smoothing the transfer of control is: U_final=(1-α)·U_auto+α·U_manual; Wherein, U_auto is the system's automatic control command, U_manual is the crew's manual control command, and α is the control transfer coefficient, with a value range of 0 to 1. The rate of change of α is not a fixed constant and is adaptively adjusted based on the matching degree between the manual control command and the system's optimal decision, as well as the duration of the control. When the control direction and amplitude are consistent with the system's suggestions, α increases rapidly and linearly. When the control conflicts with the system's safety decision, α increases slowly and a pop-up window confirms the risk of mishandling. When α is 1, the system control is completely transferred to the crew, switching to a pure monitoring and assistance mode.

9. The ship automatic collision avoidance emergency takeover control method based on multi-source sensor fusion according to claim 8, characterized in that, In step S7, the fallback protection mechanism provides safety redundancy protection after manual operation. The system monitors the ship's course, speed and surrounding target status in real time at an adjustable frequency of 5-20Hz, and compares the real-time control status with the optimal safe navigation envelope. When the actual course of the vessel deviates from the optimal collision avoidance course by 12-18° within the preset deviation range, or the actual speed deviates from the optimal speed by 2-4 knots within the preset deviation range, and the remaining collision time is less than the scenario safety threshold and there is no manual correction for more than 8-12 seconds, the system triggers fallback protection. This mechanism strictly retains the priority of manual operation, only slightly adds auxiliary correction torque to correct the vessel's attitude deviation, does not forcibly overturn the manual operation command, and outputs dual audible and visual alarms to prompt the crew to review the control status.

10. A ship automatic collision avoidance emergency takeover control system based on multi-source sensor fusion, characterized in that, include: The system comprises a multi-source sensor acquisition module, a data preprocessing and spatiotemporal synchronization module, a sensor spatiotemporal linkage triggering module, a multi-level fusion and comprehensive analysis module, a layered collision avoidance strategy module, a collision avoidance decision and comprehensive execution module, a dual-scenario emergency takeover module, a gradual transfer of control and intelligent feedback module, a human-computer interaction and comprehensive display module, and a data recording and playback module. The multi-source sensor acquisition module integrates various surface and underwater sensing devices to provide raw data for the entire process control. The data preprocessing and spatiotemporal synchronization module achieves unified calibration and time-series alignment of multi-source heterogeneous data. The sensor spatiotemporal linkage triggering module stores multiple types of... The system employs a linkage rule to enable cross-sensor collaborative detection; a multi-level fusion and comprehensive analysis module to complete data verification and three-level target identification; a hierarchical collision avoidance strategy module to store a differentiated collision avoidance strategy matrix; a collision avoidance decision and comprehensive execution module to select the optimal solution and output multi-dimensional instructions; a dual-scenario emergency takeover module to achieve dual-scenario graded emergency response; a gradual transfer of control and intelligent feedback module to achieve smooth control transfer and manual backup protection; a human-computer interaction and comprehensive display module to achieve situational visualization and human-computer interaction; and a data recording and playback module to complete full-process data storage and traceability playback.

Images