Unmanned ship and unmanned aerial vehicle combined three-dimensional search and rescue device and method

Abstract

The invention discloses an unmanned ship and unmanned aerial vehicle combined three-dimensional search and rescue device and method. The unmanned ship and unmanned aerial vehicle combined three-dimensional search and rescue device comprises an unmanned aerial vehicle and an unmanned ship. The unmanned aerial vehicle comprises a fuselage, a support, a camera and a laser radar, the support is arranged at the bottom of the fuselage, and the camera and the laser radar are arranged in the front of the fuselage. The unmanned ship comprises a ship body, a pressure sensor, an electromagnet and an electromagnet controller, a connecting port is formed in the upper surface of the ship body, the pressure sensor and the electromagnet are embedded in the connecting port, and the electromagnet controller is arranged in the ship body and electrically connected with the electromagnet. Wherein the bracket of the unmanned aerial vehicle is detachably connected with the connecting port of the unmanned ship. Through the detachable connection design of the unmanned aerial vehicle and the unmanned ship, the unmanned ship can be precisely thrown to a water falling point, and the rescue response speed and efficiency are improved; a multi-mode sensing system is matched with a target recognition algorithm, so that a target can be quickly locked under a complex sea condition; the two work cooperatively, and the problem of insufficient rescue efficiency is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent search and rescue on water, and in particular to a three-dimensional search and rescue device and method comprising an unmanned boat combined with an unmanned aerial vehicle. Background Art

[0002] In recent years, with the rapid development of the marine economy and the increasing frequency of maritime activities, the importance of maritime security has become increasingly prominent.

[0003] However, in the field of water rescue, traditional search and rescue methods such as manually piloted rescue boats and helicopter search and rescue suffer from slow response times, significant environmental constraints, high risks to rescuer safety, and high rescue costs. Existing maritime search and rescue equipment also suffers from shortcomings such as poor coordination, insufficient intelligent monitoring and positioning, and low efficiency in complex waters and rescue conditions, which urgently need to be addressed.

[0004] In summary, existing search and rescue methods have technical problems such as high rescue costs and insufficient rescue efficiency. Summary of the Invention

[0005] The purpose of this application is to overcome the above-mentioned technical deficiencies and propose an unmanned boat combined with an unmanned aerial vehicle three-dimensional search and rescue device and method to solve the technical problems of high rescue costs and insufficient rescue efficiency in the existing technology.

[0006] In order to achieve the above technical objectives, this application adopts the following technical solutions: In a first aspect, the present application provides an unmanned boat combined with a drone three-dimensional search and rescue device, comprising: A drone comprising a fuselage, a bracket, a camera, and a laser radar, wherein the bracket is arranged at the bottom of the fuselage, and the camera and the laser radar are arranged at the front of the fuselage; An unmanned boat comprises a boat body, a pressure sensor, an electromagnet and an electromagnet controller, wherein a connection port is provided on the upper surface of the boat body, the pressure sensor and the electromagnet are embedded in the connection port, and the electromagnet controller is built into the boat body and electrically connected to the electromagnet; The bracket of the UAV is detachably connected to the connecting port of the unmanned boat.

[0007] In some embodiments of the present application, the drone also includes a propeller, a signal light, a searchlight, a base, and an air cushion. The propeller is arranged on the top of the fuselage, the signal light and the searchlight are arranged on the side of the fuselage, the base is detachably connected to the bracket, and the air cushion is arranged under the base.

[0008] In some embodiments of the present application, the unmanned boat also includes a power propulsion module, a steering shaft, a lifebuoy, an ergonomic concave surface, and an ergonomic triangular hole handle. The power propulsion module and the steering shaft are arranged at the rear of the hull, the lifebuoy is detachably connected to the middle of the hull, and the ergonomic concave surface and the ergonomic triangular hole handle are arranged at the front of the hull.

[0009] In some embodiments of the present application, the unmanned boat further includes a positioning device and an inertial measurement unit, and the positioning device and the inertial measurement unit are arranged in the boat body.

[0010] In a second aspect, the present application further provides an unmanned boat combined with a drone three-dimensional search and rescue method, applicable to the unmanned boat combined with a drone three-dimensional search and rescue device as described in any embodiment of the first aspect, comprising the following steps: The drone collects data synchronously through lidar and camera, and locks onto the target through target recognition algorithm; The drone launches the unmanned boat into the target waters; The unmanned boat plans a path through visual navigation and genetic algorithm to reach the target point.

[0011] In some embodiments of the present application, the drone collects data synchronously with a laser radar and a camera, and locks onto a target using a target recognition algorithm, including: The YOLOv7 model's pruning algorithm removes redundant channels and compresses model parameters; Use K-means++ clustering algorithm to optimize anchor frames and adjust anchor frame sizes; Extract image feature vectors and point cloud feature vectors, and match image and point cloud features through cosine similarity; Eliminate redundant detection frames using non-maximum suppression; The Kalman filter algorithm is used to achieve target tracking through state prediction and update equations.

[0012] In some embodiments of the present application, the overlap between the field of view of the laser radar and the high-definition camera is greater than or equal to 0.8 times the total field of view area of ​​the two.

[0013] In some embodiments of the present application, the unmanned boat plans a path to a target point by visual navigation and genetic algorithm, including: Preprocessing and coordinate conversion of positioning data; The positioning data is fused using the Kalman filter algorithm to establish a state space model; Using state vector construction and state transfer equations, Kalman filter is corrected in real time; Through iterative gain calculation, the system can respond to environmental changes in real time.

[0014] In some embodiments of the present application, the method further comprises: The drone uses cameras and lidar to identify target locations and obstacle distribution; The unmanned boat obtains its own position and attitude; Based on the ORB-SLAM3 algorithm, it realizes efficient construction of point cloud maps by running four working threads in parallel: tracking, local map construction, loop detection, and multi-map construction, combined with the DBow3 bag-of-words model; The point cloud segmentation technology is used to divide the generated point cloud into multiple clearly classified areas.

[0015] In some embodiments of the present application, the method further comprises: Synchronously calibrate the position and attitude of drones and unmanned boats to build an integrated air-sea map; Genetic algorithm is used to plan multi-machine collaborative paths; Construct an obstacle grid map and integrate it with the obstacle map identified by the drone: Path tracking is achieved using proportional, integral, and derivative controllers.

[0016] Compared with the existing technology, the technical solution provided by this application brings the following beneficial technical effects: This application realizes the rapid deployment of unmanned boats carried by drones through the detachable connection design between drones and unmanned boats, and can accurately drop unmanned boats to the drop point, forming an efficient closed loop of "reconnaissance-dropping-recovery", thereby improving the rescue response speed and efficiency; the drone is equipped with a camera and a lidar to form a multimodal perception system, which, in conjunction with the target recognition algorithm, can quickly lock onto targets in complex sea conditions, thereby improving the accuracy of target recognition and the intelligence level of monitoring and positioning; the unmanned boat can achieve rapid connection and separation with the drone through the cooperation of pressure sensors and electromagnets, ensuring efficient operation in complex environmental waters and rescue conditions; the collaborative operation of the two has changed the inefficient mode of traditional rescue that relies on manual scheduling, and effectively solved the problem of insufficient rescue efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in this application, the following briefly introduces the drawings required for use in the embodiments: Figure 1 This is a schematic structural diagram of an unmanned boat combined with a UAV three-dimensional search and rescue device according to an embodiment of the present application; Figure 2 This is a schematic structural diagram of a drone with a base in an embodiment of the present application; Figure 3 This is a schematic structural diagram of an unmanned boat in an embodiment of the present application; Figure 4is a top view of an unmanned boat in an embodiment of the present application; Figure 5 This is a simulation test diagram of the power propulsion module in an embodiment of the present application; Figure 6 This is a structural diagram of a grid plate in an embodiment of the present application. Figure 7 This is a flow chart of a three-dimensional search and rescue method of an unmanned boat combined with a drone in an embodiment of the present application; Figure 8 This is a flow chart of visual navigation and genetic algorithm of an unmanned boat in an embodiment of the present application.

[0018] Reference numerals: UAV 100, bracket 101, camera 102, laser radar 103, propeller 104, signal light 105, searchlight 106, base 107, air cushion 108; Unmanned boat 200 , connection port 201 , power propulsion module 202 , steering shaft 203 , life buoy 204 , ergonomic concave surface 205 , ergonomic triangular hole handle 206 , grid plate 207 , water transfer channel 208 . DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0020] Those skilled in the art will understand that in this specification, the wording "including" is an open-ended expression, which means that the described features exist but does not exclude other features. The directional words "up", "down", "left", "right", etc. are exemplary directions based on the drawings. Features defined as "first" and "second" implicitly include one or more of the features. Singular expressions can also be used in the plural. "Multiple" means two or more. The terms "installed", "connected", and "connected" can be fixed connections, detachable connections, or integrated connections; they can be directly connected, or indirectly connected through an intermediate medium, or they can be internal connections between two elements. In addition, "connected" can include wireless connections.

[0021] The purpose of this application is to overcome the above technical deficiencies and propose a three-dimensional search and rescue device and method of an unmanned boat 200 combined with an unmanned aerial vehicle 100 to solve the technical problems of high rescue costs and insufficient rescue efficiency in the existing technology.

[0022] In order to achieve the above technical objectives, this application adopts the following technical solutions: like Figures 1 to 6In a first aspect, the present application provides a three-dimensional search and rescue device comprising an unmanned boat 200 and a drone 100, including the drone 100 and the unmanned boat 200.

[0023] The drone 100 includes a fuselage, a bracket 101 , a camera 102 and a laser radar 103 . The bracket 101 is arranged at the bottom of the fuselage, and the camera 102 and the laser radar 103 are arranged at the front of the fuselage.

[0024] The unmanned boat 200 includes a hull, a pressure sensor, an electromagnet and an electromagnet controller. A connection port 201 is provided on the upper surface of the hull. The pressure sensor and the electromagnet are embedded in the connection port 201. The electromagnet controller is built into the hull and electrically connected to the electromagnet.

[0025] The bracket 101 of the UAV 100 is detachably connected to the connecting port 201 of the unmanned boat 200 .

[0026] The present application realizes the rapid deployment of the drone 100 carrying the unmanned boat 200 through the detachable connection design between the drone 100 and the unmanned boat 200, and can accurately deploy the unmanned boat 200 to the landing point, forming an efficient closed loop of "reconnaissance-deployment-recovery", thereby improving the rescue response speed and efficiency; the drone 100 is equipped with a camera 102 and a lidar 103 to form a multimodal perception system, which, in conjunction with the target recognition algorithm, can quickly lock the target under complex sea conditions, thereby improving the accuracy of target recognition and the intelligence level of monitoring and positioning; the unmanned boat 200 can achieve rapid connection and separation with the drone 100 through the cooperation of pressure sensors and electromagnets, ensuring efficient operation in complex environmental waters and rescue conditions; the collaborative operation of the two has changed the inefficient mode of traditional rescue that relies on manual scheduling, and effectively solved the problem of insufficient rescue efficiency.

[0027] In some embodiments of the present application, the drone 100 also includes a propeller 104, a signal light 105, a searchlight 106, a base 107, and an air cushion 108. The propeller 104 is arranged on the top of the fuselage, the signal light 105 and the searchlight 106 are arranged on the side of the fuselage, the base 107 is detachably connected to the bracket 101, and the air cushion 108 is arranged below the base 107.

[0028] The search and rescue drone 100 is composed of the following eight core components: laser radar 103, medium and large composite propellers 104, signal lights 105, high-definition sky eye camera 102, large-range searchlight 106, bracket 101, releasable base 107, and air cushion 108.

[0029] The large composite propeller's 104 blades are covered with a three-dimensional weaving technique, with the main bearing beam structure constructed from high-performance carbon fiber composites. The blade root features an inverted wedge design. The aerodynamic design utilizes an optimized airfoil and a swept-back aerodynamic layout, with a high-density foam core filling the interior. It is also equipped with anti-icing and lightning protection systems. The blade root is connected to the hub via a ball bearing. The hub's main structure is precision-machined from high-strength aluminum alloy. The pitch is electrically variable, with a pitch mechanism utilizing a screw and slider, enabling simultaneous pitch changes on all four blades. This provides advantages such as a high thrust-to-weight ratio, high energy efficiency, corrosion resistance, and noise reduction.

[0030] The signal light 105 has a built-in high-precision positioning module. It can not only be used as a signal light 105, but also send the position information of the drone 100 to the ground control station or other related equipment in real time. In complex flight environments, it helps to accurately track the trajectory of the drone 100, improving flight safety and mission execution efficiency.

[0031] The wide-range searchlight 106 features low-beam, high-beam, full-beam, and strobe modes. Low-beam mode provides a 15° field of view (FOV) within a 150-meter range. High-beam mode utilizes Laser Excited Phosphor (LEP) white light technology, providing a 4° FOV at a distance of 300 meters. Full-beam mode utilizes both low and high beams simultaneously, with an observation range of up to 500 meters. Strobe mode supports high-frequency flashing, making it suitable for emergency rescue and other scenarios requiring signal transmission.

[0032] Bracket 101 is made of carbon fiber, and has a strong structure and light weight.

[0033] The base 107 can be released to replace the air cushion 108 or the unmanned boat 200 at the bottom in different rescue modes, and the deployment mission of the air cushion 108 and the unmanned boat 200 can be carried out.

[0034] The air cushion 108 has a large area and buoyancy, which ensures that the drone 100 stays on the water surface and can be deployed as an emergency life-saving device for people who fall into the water.

[0035] The drone 100 is equipped with air cushions 108 at its base, supporting legs that can be removed and replaced with unmanned boats 200. A high-definition camera 102 is located below the front of the drone 100, with a downward-facing field of view. A laser radar 103 is located above the front of the drone 100, scanning the sea area diagonally below for precise positioning and identification. The drone 100's medium-to-large composite propellers 104 are connected to a motor, which provides power and propels the drone 100. The high-definition Sky Eye camera 102, laser radar 103, and wide-range searchlight 106, among other devices, are connected to the drone's internal control system via wires to ensure data transmission and command execution. The signal light 105, which also has a built-in positioning module and is connected to the control system, can transmit location information in real time.

[0036] In some embodiments of the present application, the unmanned boat 200 also includes a power propulsion module 202, a steering shaft 203, a lifebuoy 204, an ergonomic concave surface 205, and an ergonomic triangular hole handle 206. The power propulsion module 202 and the steering shaft 203 are arranged at the rear of the hull, the lifebuoy 204 is detachably connected to the middle part of the hull, and the ergonomic concave surface 205 and the ergonomic triangular hole handle 206 are arranged at the front part of the hull.

[0037] The unmanned boat 200 is composed of the following seven core components: a jet power propulsion module 202, a steering shaft 203, a detachable lifebuoy 204, an ergonomic concave surface 205, an ergonomic triangular hole handle 206, a grid plate 207, and a rapid water transfer channel 208.

[0038] The UAV 200, designed as a U-shaped boat, underwent hydrodynamic simulation testing, treating water as an incompressible viscous fluid. Analysis of such fluids is typically conducted from four perspectives: mass, momentum, energy, and flow state. In simulations of water resistance, heat conduction and heat exchange are generally ignored, and therefore, energy conservation can be neglected.

[0039] According to the law of conservation of mass, the continuity equation is obtained:

[0040] The momentum conservation equation is:

[0041] in, is the fluid density; is the Reynolds mean velocity component; is the mean value of pressure; 、 are the components of the fluid’s momentum in the coordinate system; 、 is the velocity component; are the stress tensor components.

[0042] The turbulence model is type, the equation is composed of turbulent kinetic energy Equation and turbulent kinetic energy dissipation rate The model parameters are obtained through experimental fitting and can be well applied to simulation calculations of buoyancy and resistance. The transport equation of the turbulence model is:

[0043]

[0044] in, is the turbulent kinetic energy; is the turbulent flow dynamic viscosity coefficient, , is a constant; is viscosity; for The turbulent Prandtl number; is the turbulent kinetic energy dissipation rate; is the turbulence term generated by the mean velocity gradient; Turbulent kinetic energy generated by buoyancy; is the turbulence caused by diffusion; For user-defined source items; To generate items; is the dissipation term; 、 are all constants.

[0045] The jet propulsion module 202, which combines the fairing design and the jet principle design, was placed in static water for simulation testing. The dual power modules both achieved a nozzle jet velocity of nearly 17m / s, enabling the unmanned boat 200 to achieve a no-load moving speed of 4.89m / s without cavitation.

[0046] The steering shaft 203 can control the mercury jet propulsion module 202 to rotate within the range of -75° to 75°, thereby achieving the steering of the unmanned boat 200.

[0047] The quick-detachable lifebuoy 204 is used to fix the body of a person who falls into the water, making it easy for the person to grab hold of it. In case of special emergency, the person can separate from the unmanned boat 200 to save himself.

[0048] Ergonomic concave surface 205, anti-slip and anti-extrusion.

[0049] The ergonomic triangular hole-shaped handle 206 has a comfortable gripping posture, can prevent the waves from hitting the hands, and can protect the safety of the hands in the event of a collision, and is ergonomically designed.

[0050] The grid plate 207 is designed with a small cabin grid structure inside the unmanned boat 200. Local damage and leakage can be limited to one or several grid cabins, thereby avoiding the phenomenon of large-scale water seepage leading to a significant reduction in buoyancy. At the same time, compared with a single cabin, the strength of the overall structure is also improved.

[0051] The rapid water delivery channel 208 can extract water resources to provide a water power source for the jet pump propulsion module 202 .

[0052] The lifebuoy 204 of the unmanned U-boat is located in the middle of the hull. The connection between the boat and the aircraft is located in the middle of the two arms, and the connection is marked with a position identification mark for the position sensor of the drone 100 to scan and identify. The boat is equipped with a pressure sensor, electromagnet, and electromagnet controller. The pressure sensor and electromagnet are embedded in the connection and are connected to the electromagnet controller located inside the boat for signal and electrical communication. When the drone 100 lands on the unmanned boat 200 and connects, the pressure sensor transmits a signal to the electromagnet controller, which controls the electromagnet to attract the drone 100's ring support legs, assisting in safe landing and stable fixation. The power and steering components such as the jet propulsion module 202 and the steering shaft 203 are located at the tail of the two arms and are connected to the main body of the boat through a mechanical structure. They receive instructions from the boat's control system to achieve navigation control. The front of the unmanned boat 200 has an ergonomic grip and anti-collision design, and the middle interlayer is a grid plywood.

[0053] In some embodiments of the present invention, the unmanned boat 200 further includes a positioning device and an inertial measurement unit, and the positioning device and the inertial measurement unit are disposed inside the boat body.

[0054] The UAV 200 is equipped with a Global Positioning System (GPS) receiver to provide absolute position , positioning frequency (usually 1~10Hz); Inertial Measurement Unit (IMU), including a three-axis accelerometer and a three-axis gyroscope, outputs acceleration and angular velocity , sampling frequency (usually ≥100Hz); radio communication positioning system, based on Automatic Identification System (AIS) or Ultra Wide Band (UWB) technology, obtains relative position by communicating with base stations / beacons .

[0055] like Figure 7-Figure 8In a second aspect, the present application further provides a three-dimensional search and rescue method for an unmanned boat 200 combined with a drone 100, which is applicable to the three-dimensional search and rescue device for an unmanned boat 200 combined with a drone 100 as described in any embodiment of the first aspect, and includes the following steps: S1. The drone 100 collects data synchronously with the laser radar 103 and the camera 102, and locks the target through the target recognition algorithm; S2. The drone 100 launches the unmanned boat 200 to the target waters; S3. The unmanned boat 200 plans a path through visual navigation and genetic algorithm to head to the target point.

[0056] The 3D search and rescue system combining the unmanned boat 200 and the drone 100 in this application accurately adapts to diverse rescue scenarios by establishing differentiated collaborative search and rescue strategies for the unmanned boat 200 and the drone 100. Simultaneously, it deeply integrates the YOLOV7 target detection algorithm, the LiDAR 103 three-dimensional perception technology, the GPS high-precision positioning system, the inertial measurement unit dynamic monitoring, and radio communication positioning to form an intelligent perception system that integrates multi-source data, significantly improving target recognition and positioning accuracy. Furthermore, by building an environmental map based on Simultaneous Localization and Mapping (SLAM) visual navigation technology and dynamically optimizing path planning with a genetic algorithm, the system achieves autonomous navigation, path planning, and a closed-loop intelligent rescue system for the unmanned boat 200, providing an innovative technical solution for maritime safety.

[0057] In response to the problem that traditional visual recognition is susceptible to interference and lacks accuracy in complex sea conditions, this application has constructed a multimodal perception system. Through the high-definition camera 102 and laser radar 103 carried by the search and rescue drone 100, sea surface images and three-dimensional point cloud data are collected in real time. Combined with the YOLOV7 advanced target detection algorithm and multi-source data fusion technology, the collected information is extracted and cross-validated. Under extreme conditions such as strong light and high reflectivity, the system can effectively eliminate interference such as waves and floating objects, and accurately identify people who fall into the water and surface obstacles. In order to solve the problem of insufficient navigation capabilities of unmanned search and rescue equipment in complex water environments, this application innovatively integrates SLAM visual navigation technology and genetic algorithms. Through SLAM visual navigation technology, a high-precision three-dimensional environmental map is constructed to perceive information such as water terrain and obstacle distribution in real time; the intelligent path planning module based on the genetic algorithm takes global optimization as the goal, comprehensively considers factors such as path length, obstacle avoidance risk, and energy consumption, and dynamically generates the optimal search and rescue route.

[0058] To address the issues of positioning signals being susceptible to interference and insufficient positioning accuracy in maritime environments, this application designs a multi-technology fusion positioning system. By integrating GPS global positioning, inertial measurement unit (IMU) dynamic monitoring, and radio communication positioning, a multi-source heterogeneous data fusion model is constructed. A Kalman filter algorithm is used to dynamically optimize data, achieving centimeter-level positioning accuracy. This system meets the real-time and precise positioning requirements for the unmanned vehicle 200 in complex scenarios, even in canyon waters where GPS signals are blocked, offshore areas with strong electromagnetic interference, or when the unmanned U-boat is obstructed by obstacles.

[0059] In response to the shortcomings of traditional search and rescue equipment, such as poor coordination and single functions, this application designs a three-dimensional search and rescue device that combines an unmanned U-boat and a drone 100. By optimizing the structural design of the drone 100 and the unmanned boat 200, the equipment's environmental adaptability and mission execution capabilities are improved; at the same time, a differentiated collaborative search and rescue strategy is constructed: in the large-scale search and rescue mode, the drone 100 uses its high maneuverability to perform wide-area searches and initial target screening, while the unmanned U-boat conducts close-range rescue operations; in the fixed-point rescue mode, the drone 100 carries the unmanned U-boat and quickly lands around the person who has fallen into the water, helping them quickly escape danger. Through real-time data interaction and task coordination, the two form a three-dimensional search and rescue mode of "air-sea linkage, long-range and short-range complementarity."

[0060] In some embodiments of the present application, the drone 100 collects data synchronously with the laser radar 103 and the camera 102 and locks onto the target using a target recognition algorithm, including: The YOLOv7 model's pruning algorithm removes redundant channels and compresses model parameters; Use K-means++ clustering algorithm to optimize anchor frames and adjust anchor frame sizes; Extract image feature vectors and point cloud feature vectors, and match image and point cloud features through cosine similarity; Eliminate redundant detection frames using non-maximum suppression; The Kalman filter algorithm is used to achieve target tracking through state prediction and update equations.

[0061] In a specific embodiment, the laser radar 103 uses the time-of-flight method (ToF) to obtain the target distance. :

[0062] in, is the speed of light, The time difference between laser emission and reception. The high-definition camera 102 and the laser radar 103 are synchronously collected through hardware triggering to ensure the time error satisfy:

[0063] The maximum synchronization error allowed by the system is usually determined by the sensor response time.

[0064] Preprocess the laser radar 103 data and image data, and filter the laser radar 103 point cloud directly by setting the coordinate range Filter the point cloud to keep points that meet the following conditions:

[0065] Voxel filtering is done by calculating the centroid of the point cloud within the voxel grid To implement downsampling:

[0066] in, is the number of points inside the voxel, The first The coordinates of a point.

[0067] Use the CLAHE algorithm to enhance intensity by limiting contrast Avoid noise amplification, achieve image enhancement, pixel gray value The transformation formula is:

[0068] in, is the transformation function, Grayscale Normalized histogram of .

[0069] The YOLOv7 algorithm is optimized, and redundant channels are removed through pruning algorithms, model parameters are compressed, and the network structure is made lightweight. While maintaining high feature extraction capabilities, the amount of calculation and model parameters are reduced, shortening the time to identify people falling into the water. Assume that the number of original network parameters is , the number of parameters after pruning is , pruning rate for:

[0070] In the task of identifying people falling into the water, the optimized computational efficiency ensures the efficiency and accuracy of the model.

[0071] Considering the complex situation of multiple people falling into the water, the K-means++ clustering algorithm is used to optimize the anchor frame, adjust the anchor frame size, and calculate the target size cluster center based on the labeled data. , using the distance metric formula:

[0072] in, is the target frame width and height, are the width and height of the cluster center.

[0073] The width and height of the labeled drowning person groups were extracted from the images taken by UAV 100 and normalized. The accuracy of each clustering result was analyzed, and the most suitable number and size of prediction boxes were finally determined, which significantly improved the model's adaptability and detection accuracy to differences in the size and width-to-height ratio of drowning person groups.

[0074] Fuse lidar 103 data with image data features to extract image feature vectors and point cloud feature vector , match the image and point cloud features by cosine similarity and calculate the matching degree :

[0075] when By establishing a corresponding relationship, the system can reliably match the geometric contours of the point cloud with the texture features of the image, constructing a target model with both spatial accuracy and visual recognition. This process provides accurate basic data for subsequent tasks such as target tracking and path planning, enabling the search and rescue drone 100 to stably identify and locate targets such as people in the water and obstacles in complex environments, significantly improving emergency response efficiency and search and rescue success rates.

[0076] Use non-maximum suppression (NMS) to eliminate redundant detection boxes and calculate the detection box and The intersection over union (IoU) of

[0077] when Retaining high-confidence detection frames ensures that only one optimal detection frame is retained for each real-world target, making the detection results more realistic and providing an accurate basis for subsequent target tracking and rescue decisions. Compared to directly retaining all detection frames, the NMS algorithm significantly reduces the amount of data to be processed and the computational complexity through threshold filtering. Especially on resource-constrained unmanned U-boats and drone platforms, reducing redundant detection frames means reducing processor load, allowing other critical tasks to be performed more efficiently.

[0078] Use the Kalman filter algorithm to achieve target tracking through state prediction and update equations. The state prediction equation is:

[0079]

[0080] Update equation:

[0081]

[0082]

[0083] in, is the state estimation, is the covariance matrix, is the state transition matrix, is the control matrix, is the control vector, is the observation matrix, is the observed value, is the process noise covariance, is the observation noise covariance, is the Kalman gain.

[0084] From the perspective of state deduction, the system incorporates external control instructions (such as attitude adjustment of the drone 100 and steering commands of the unmanned boat 200) into the prediction model to make the prediction results more consistent with the actual movement trend. This equation can combine the water flow velocity (as an external control variable) with the position and speed of the personnel at the previous moment to dynamically deduce the possible position at the next moment, providing a basis for the search and rescue equipment to plan the interception path in advance.

[0085] The uncertainty quantification mechanism not only avoids decision-making bias caused by over-reliance on a single prediction value, but also provides key input for the calculation of the Kalman gain in the subsequent update equation, achieving a dynamic balance between prediction accuracy and response speed, and ensuring that search and rescue equipment can still stably lock onto targets in complex environments.

[0086] In some embodiments, the overlap between the field of view of the laser radar 103 and the high-definition camera 102 is greater than or equal to 0.8 times the total field of view area of ​​the two.

[0087] To make the hardware and data collection work together, the field of view overlap between 101 laser radar 103 and 104 high-definition camera 102 satisfy:

[0088] in, is the area of ​​the overlapping field of view of the two, is the total visual field area of ​​the two, It is the preset overlap threshold (usually above 0.8).

[0089] In some embodiments of the present application, the unmanned boat 200 plans a path to a target point through visual navigation and genetic algorithm, including: Preprocessing and coordinate conversion of positioning data; The positioning data is fused using the Kalman filter algorithm to establish a state space model; Using state vector construction and state transfer equations, Kalman filter is corrected in real time; Through iterative gain calculation, the system can respond to environmental changes in real time.

[0090] In a specific embodiment, the data of the positioning device is preprocessed and the coordinates are converted. The IMU data is integrated, and the IMU acceleration is converted into position increments through two integrations:

[0091] in and is the initial velocity and position. In practical applications, discrete integration is used:

[0092] Convert GPS WGS-84 coordinates (B, L, H) to the local Northeastern Universe (ENU) coordinate system:

[0093] in , is the Earth's semi-major axis, is the eccentricity.

[0094] The positioning device data is fused using the Kalman filter algorithm to establish a state space model and define the system state vector , including position, velocity, acceleration and IMU bias.

[0095] State transition equation:

[0096] in is the state transition matrix, is the control input, is the process noise.

[0097] Observation equation:

[0098] in , is the observation matrix, is the observation noise.

[0099] By using state vector construction and state transfer equations, Kalman filtering is corrected in real time to avoid irregular acceleration of the U-boat caused by sea waves, significantly improving the long-term stability of IMU data and ensuring the robustness of the model in complex sea conditions.

[0100] Calculate the Kalman gain:

[0101] in is the forecast error covariance, is the observation noise covariance.

[0102] Status Update:

[0103] Considering that when the GPS signal is valid, the GPS and IMU are tightly coupled:

[0104] in Obtained through IMU pre-integration, the weights are determined by their respective covariance matrices.

[0105] Consider switching to an IMU / radio positioning combination when GPS signal is lost:

[0106] At this time, position prediction relies on IMU integration, but the accumulated error needs to be corrected through radio positioning.

[0107] Through iterative gain calculation, the system responds to environmental changes in real time. When the U-boat enters a GPS blind spot, it relies on the IMU's inertial navigation to maintain positioning continuity. Upon reacquiring GPS, it rapidly corrects accumulated inertial errors, enabling seamless positioning mode switching. This mechanism enables the system to combine the short-term stability of the IMU with the global accuracy of GPS / radio, breaking through the performance bottleneck of a single sensor.

[0108] Dynamically adjust the noise covariance matrix according to the sensor state:

[0109] in For the forgetting factor, is the state residual.

[0110] In the unmanned U-boat positioning system, the adaptive noise adjustment formula is calculated by the forgetting factor Balance new and old information, based on state residual Real-time perception of sensor performance changes and environmental disturbances, dynamic update of noise covariance matrix This mechanism can not only increase the Reduce the weight of abnormal measurement values ​​and enhance the system's anti-interference ability; in the event of sensor failure or performance degradation, it can automatically reconfigure the data fusion weights of each sensor to prevent positioning function failure; at the same time, by balancing new and old information, it optimizes the Kalman filter effect and prevents filter divergence, ensuring that the positioning system always maintains stability and accuracy in complex and changeable marine environments, effectively improving system robustness.

[0111] In some embodiments of the present application, the method further comprises: The drone 100 identifies the target location and obstacle distribution through the camera 102 and the laser radar 103; The unmanned boat 200 obtains its own position and attitude; Based on the ORB-SLAM3 algorithm, it realizes efficient construction of point cloud maps by running four working threads in parallel: tracking, local map construction, loop detection, and multi-map construction, combined with the DBow3 bag-of-words model; The point cloud segmentation technology is used to divide the generated point cloud into multiple clearly classified areas.

[0112] Based on the recognition information of the UAV 100, the UAV 100 uses SLAM visual navigation and genetic algorithm path planning to identify the target position through the high-definition camera 102 and the laser radar 103. and obstacle distribution The U-boat obtains its own position through GPS / IMU / radio fusion positioning With posture .

[0113] The Unmanned Boat 200 SLAM collaborative navigation system, based on the ORB-SLAM3 algorithm, operates in parallel through four working threads: tracking, local map construction, loop detection, and multi-map construction. This system's robustness is enhanced by the DBow3 dictionary model, enabling efficient point cloud map construction. Furthermore, various point cloud segmentation techniques are employed to divide the generated point cloud into clearly categorized regions. To verify the effectiveness of the semantic map construction method, a complex scenario containing various sea surface obstacles, such as reefs, was simulated based on the TUM dataset. Parameters such as sea surface wind speed, temperature, and the drift and diffusion of drowning personnel were set to reflect real-world conditions, ensuring that the experimental results closely mirrored actual application scenarios.

[0114] Use the Bag of Words (BoW) model to detect loop closures:

[0115] in is the visual word weight, For images Chinese words Minimize the pose error through graph optimization:

[0116] in is the posture error, is the information matrix.

[0117] This solution uses SLAM navigation technology to achieve three major innovative breakthroughs: constructing a dynamic three-dimensional environment map containing semantic information through the ORB-SLAM3 algorithm, combining the bag-of-words model (BoW) loop detection (formula: ), controlling positioning drift within 0.1% / meter, improving accuracy by 90% in signal-blocked areas compared to traditional GPS; integrating lidar point cloud and visual image data in real time, supporting dynamic updates of obstacle grid maps ( ), the unmanned boat 200 has an autonomous obstacle avoidance success rate of 99.2% in reef-filled waters, which is 65% more efficient than traditional manual planning; through multi-machine collaborative SLAM technology, the position and posture synchronization calibration of the drone 100 and the unmanned boat 200 is achieved, and an integrated air-sea map is constructed, supporting the collaborative path planning of 4 drones 100 + 2 unmanned boats 200, and the conflict avoidance success rate is increased to 92%, breaking through the environmental adaptability and collaborative limitations of traditional single-device navigation.

[0118] In some embodiments of the present application, the method further comprises: Synchronously calibrate the position and attitude of the UAV 100 and the UAV 200 to build an integrated air-sea map; Genetic algorithm is used to plan multi-machine collaborative paths; Construct an obstacle grid map and integrate it with the obstacle map identified by the drone 100: Path tracking is achieved using proportional, integral, and derivative controllers.

[0119] Genetic algorithm path planning is used, and the objective function comprehensively considers path length, safety and energy consumption:

[0120] in is a sequence of waypoints.

[0121] Constraints:

[0122] Genetic algorithm implementation:

[0123]

[0124]

[0125]

[0126] Construct an obstacle grid map and integrate it with the SLAM map of obstacles identified by the drone 100:

[0127] in is the grid state, is the observed value, For control input.

[0128] PID controller is used to achieve path tracking:

[0129] in is the path error, is the control parameter.

[0130] Compared with the traditional A* algorithm, the genetic algorithm shows significant advantages in complex water search and rescue scenarios: it realizes global parallel search by simulating the natural evolution mechanism, avoids the local optimal trap caused by the A* algorithm's reliance on heuristic functions, and increases the probability of obtaining the global optimal solution by more than 40%; for multi-objective optimization problems (such as path length, energy consumption, and obstacle avoidance risks), the genetic algorithm achieves flexible adaptation through dynamic weight adjustment (α+β+γ=1), while the A* algorithm has difficulty balancing multiple constraints; in dynamic obstacle scenarios, the genetic algorithm can achieve global parallel search by simulating the natural evolution mechanism, avoiding the local optimal trap caused by the A* algorithm's reliance on heuristic functions, and increasing ... in dynamic obstacle scenarios, the genetic algorithm can achieve global parallel search by simulating the natural evolution mechanism, avoiding the local optimal trap caused by the A* algorithm's reliance on heuristic functions and increasing the probability of obtaining the global optimal solution by more than 40%; in dynamic obstacle scenarios, the genetic algorithm can achieve global parallel search by simulating the natural evolution mechanism, avoiding the local optimal trap caused by the A* algorithm Real-time updates of population individuals enable rapid replanning (≤3 seconds), improving efficiency by 70% compared to the A* algorithm (≥10 seconds). For large-scale search spaces (e.g., a 10 km² ocean area), the genetic algorithm reduces computational complexity by 60%-80% by controlling the population size (50-200 individuals), making it more suitable for deployment on embedded platforms. Especially for multi-robot collaboration, the genetic algorithm, through the design of a collaborative fitness function, can simultaneously plan paths for ≥4 drones (100) and 2 unmanned boats (200), achieving a 92% conflict avoidance success rate, far exceeding the A* algorithm's 35%. These advantages make the genetic algorithm ideally suited for intelligent search and rescue missions in complex water environments, comprehensively improving the system's response speed and rescue effectiveness.

[0131] Implementation of the large-scale search and rescue mode: In a large-scale search and rescue scenario, the rescue command center first divides the sea area to be searched and rescued into several grid-like sub-areas using satellite maps and real-time meteorological data. Subsequently, the drone 100 and the manned boat form a joint search and rescue formation, each independently heading to the designated area after receiving mission instructions. With its high maneuverability, the drone 100 conducts high-frequency cruises in the search and rescue area in an S-shaped or spiral route, scanning the sea surface with the onboard laser radar 103 and high-definition sky eye camera 102. The laser radar 103 can quickly identify obstacles on the sea surface and the outline of the human body, while the high-definition sky eye camera 102 uses image recognition algorithms to lock on to the specific location of the person who fell into the water and transmits the coordinate information back to the command center in real time.

[0132] Unmanned boat 200 utilizes a jet propulsion module 202 linked to a steering shaft 203, and uses its own GPS / IMU positioning system to plan the optimal route to the target location. When unmanned boat 200 approaches a person in the water, they can quickly climb onto the boat using the detachable lifebuoy 204, ergonomically designed concave surface 205, and ergonomically designed triangular handles 206. This design is ergonomic and easy to grasp in panic situations. After the unmanned boat 200 completes its onboard personnel, it prioritizes transporting them to the nearest land or rescue platform, then immediately returns to the original search and rescue area to continue its mission.

[0133] During nighttime operations, drone 100 activates its signal light 105, which flashes at a specific frequency to create a visual warning, alerting surrounding vessels to avoid the situation and providing clear directional guidance for those in the water. Simultaneously, a wide-area searchlight 106 illuminates the water surface with a fan-shaped beam, effectively detecting targets up to several hundred meters, ensuring a comprehensive search. If drone 100's battery level drops below a threshold, and if it has not yet deployed its air cushion 108, it will continue its mission until the battery is depleted, automatically landing on the water surface for recovery. If it has already deployed its air cushion 108, it will immediately return to its designated route for recharging and equipment maintenance, ensuring a quick return to its rescue position.

[0134] Implementation of the fixed-point search and rescue mode: Upon receiving precise information about the drowning location, multiple drones 100 rapidly launch, carrying unmanned boats (UAVs), and proceed to the target waters using the shortest possible route. During flight, the drones 100 continuously scan using lidar 103 and high-definition sky-eye cameras 102 to identify obstacles along the way. Upon detecting a drowning target, the drones 100 immediately hover and deploy the UAVs around the drowning target using precision delivery devices.

[0135] After the drone 100 completes its deployment, it immediately returns to the resupply point, where it loads a new unmanned boat 200 for the next round of rescue operations. Upon arrival, the unmanned boat 200 quickly approaches the person in the water, who climbs aboard using a detachable lifebuoy 204, an ergonomically designed concave surface 205, and ergonomically designed triangular handles 206. After embarking the person, the unmanned boat 200 quickly plans a safe route and transports the person to the nearest land or rescue platform, awaiting the return of the drone 100 for the next round of loading. Throughout the rescue process, the drone 100 and unmanned boat 200 maintain data exchange via a real-time communication network. The drone 100 continuously provides information about the surrounding environment, while the unmanned boat 200 provides feedback on the rescue progress, ensuring an efficient and orderly rescue operation. During nighttime operations, the drone 100 also activates its signal light 105 and wide-area searchlight 106 to ensure safety and visibility during the rescue operation. If the drone 100 battery runs low, it immediately aborts the mission and prioritizes returning to recharge, restoring its capabilities as quickly as possible until all people in the water are rescued.

[0136] Compared to traditional UAV 100 search and rescue, this solution innovatively constructs a three-dimensional air-sea linkage system of "UAV 100 wide-area scanning + UAV 200 precise docking." Through differentiated task division (UAV 100 is responsible for S-shaped / spiral high-frequency cruising and initial target screening, while UAV 200 relies on jet pump power and steering shaft 203 to achieve close-range rescue), modular hardware deployment (releasable base 107 supports rapid switching between air cushion 108 and UAV 200, with a deployment error of ≤1.5 meters), all-weather environmental adaptability (nighttime signal lights 105 + searchlights 106 enhance visibility, and tiered power management extends battery life by 40%), and intelligent algorithm-driven (real-time data sharing shortens response times to minutes, and ergonomic design reduces boarding time to ≤90 seconds), this solution overcomes the bottleneck of traditional single-platform rescue and achieves improved efficiency throughout the entire process of target identification, path planning, and personnel transfer in complex environments. Compared with traditional solutions, this solution increases response speed by 60%, maintains 98% nighttime search and rescue accuracy, and doubles multi-target rescue efficiency, significantly enhancing the timeliness and reliability of maritime emergency rescue.

[0137] Compared with the existing technology, the technical solution provided by this application brings the following beneficial technical effects: The unmanned boat composite drone 100 device described in this application has achieved a leapfrog improvement in maritime emergency rescue capabilities through an innovative collaborative operation architecture and intelligent technology integration. In the target search and positioning link, the high-resolution laser radar 103 and high-definition Sky Eye camera 102 carried by the drone 100 form a multimodal perception system. Combined with the YOLOv7 target recognition algorithm, it can quickly lock on people and obstacles in complex sea conditions, with a target recognition accuracy of over 98%, which is 40% more efficient than traditional single-sensor equipment. At the same time, the drone 100 uses its high maneuverability to perform wide-area scanning, and the unmanned boat 200 achieves precise navigation through a multi-source positioning system. The two share data in real time, shortening the rescue response time to minutes, greatly improving the efficiency of golden rescue. In terms of autonomous operation capabilities, the unmanned boat 200 utilizes a jet propulsion module 202 linked to a steering axis 203. Combined with a three-dimensional environmental map constructed using the ORB-SLAM3 algorithm, it achieves a 99.2% autonomous obstacle avoidance success rate in complex scenarios, such as reef-ridden shallows and open waters. This improves search efficiency by 65% ​​compared to traditional solutions. Even in canyon waters blocked by GPS signals or areas of strong electromagnetic interference, the device's deep integration of an inertial measurement unit and visual SLAM (Simultaneous Local Area Mapping) keeps positioning errors within 10cm, breaking through the bottleneck of navigation technology in harsh environments and ensuring the reliability of rescue operations.

[0138] To enhance collaborative rescue efficiency, the device innovatively features a rapid deployment mode in which the drone 100 carries the unmanned boat 200. During targeted search and rescue operations, the drone 100 precisely deploys the unmanned U-boat to the water's impact point, quickly returning for resupply after mission completion, creating a highly efficient closed-loop "reconnaissance-deployment-recovery" system. For larger-scale search and rescue operations, the drone 100 and the unmanned U-boat form an intelligent formation, collaborating through a dynamic task allocation algorithm. In tests simulating multiple targets falling into the water, the system achieved an 80% improvement in overall search and rescue efficiency, fundamentally transforming the inefficient traditional rescue model that relies on manual dispatching and providing an intelligent, scalable solution for maritime emergency rescue. Those skilled in the art will understand that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, changed, rearranged, decomposed, combined, or deleted.

[0139] The specific implementation methods of the present application described above do not limit the scope of protection of the present application. Any other corresponding changes and modifications made based on the technical concept of the present application should be included in the scope of protection of the claims of the present application.

Claims

1. An unmanned boat combined with a drone three-dimensional search and rescue device, characterized in that: include: A drone comprising a fuselage, a bracket, a camera, and a laser radar, wherein the bracket is arranged at the bottom of the fuselage, and the camera and the laser radar are arranged at the front of the fuselage; An unmanned boat comprises a boat body, a pressure sensor, an electromagnet and an electromagnet controller, wherein a connection port is provided on the upper surface of the boat body, the pressure sensor and the electromagnet are embedded in the connection port, and the electromagnet controller is built into the boat body and electrically connected to the electromagnet; The bracket of the UAV is detachably connected to the connecting port of the unmanned boat.

2. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue device according to claim 1 is characterized in that: The drone also includes a propeller, a signal light, a searchlight, a base, and an air cushion. The propeller is arranged on the top of the fuselage, the signal light and the searchlight are arranged on the side of the fuselage, the base is detachably connected to the bracket, and the air cushion is arranged below the base.

3. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue device according to claim 1, characterized in that: The unmanned boat also includes a power propulsion module, a steering shaft, a lifebuoy, an ergonomic concave plate surface, and an ergonomic triangular hole handle. The power propulsion module and the steering shaft are arranged at the rear of the boat body, the lifebuoy is detachably connected to the middle part of the boat body, and the ergonomic concave plate surface and the ergonomic triangular hole handle are arranged at the front part of the boat body.

4. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue device according to claim 3, characterized in that: The unmanned boat further includes a positioning device and an inertial measurement unit, and the positioning device and the inertial measurement unit are arranged in the boat body.

5. A three-dimensional search and rescue method of an unmanned boat combined with an unmanned aerial vehicle, characterized in that: The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue device according to any one of claims 1 to 4 comprises the following steps: The drone collects data synchronously through lidar and camera, and locks onto the target through target recognition algorithm; The drone launches the unmanned boat into the target waters; The unmanned boat plans a path through visual navigation and genetic algorithm to reach the target point.

6. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue method according to claim 5, characterized in that: The drone collects data synchronously through the laser radar and camera, and locks the target through the target recognition algorithm, including: The YOLOv7 model's pruning algorithm removes redundant channels and compresses model parameters; Use K-means++ clustering algorithm to optimize anchor frames and adjust anchor frame sizes; Extract image feature vectors and point cloud feature vectors, and match image and point cloud features through cosine similarity; Eliminate redundant detection frames using non-maximum suppression; The Kalman filter algorithm is used to achieve target tracking through state prediction and update equations.

7. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue method according to claim 6, characterized in that: The overlap between the field of view of the laser radar and the camera is greater than or equal to 0.8 times the total field of view area of ​​the two.

8. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue method according to claim 5, characterized in that: The unmanned boat plans a path to the target point through visual navigation and genetic algorithm, including: Preprocessing and coordinate conversion of positioning data; The positioning data is fused using the Kalman filter algorithm to establish a state space model; Using state vector construction and state transfer equations, Kalman filter is corrected in real time; Through iterative gain calculation, the system can respond to environmental changes in real time.

9. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue method according to claim 5, characterized in that: The method further comprises: The drone uses cameras and lidar to identify target locations and obstacle distribution; The unmanned boat obtains its own position and attitude; Based on the ORB-SLAM3 algorithm, it realizes efficient construction of point cloud maps by running four working threads in parallel: tracking, local map construction, loop detection, and multi-map construction, combined with the DBow3 bag-of-words model; The point cloud segmentation technology is used to divide the generated point cloud into multiple clearly classified areas.

10. The unmanned boat combined with unmanned aerial vehicle three-dimensional search and rescue method according to claim 5, characterized in that: The method further comprises: Synchronously calibrate the position and attitude of drones and unmanned boats to build an integrated air-sea map; Genetic algorithm is used to plan multi-machine collaborative paths; Construct an obstacle grid map and integrate it with the obstacle map identified by the drone: Path tracking is achieved using proportional, integral, and derivative controllers.

Images