Water surface garbage intercepting and cleaning ship based on yolo visual prediction and motion cooperative decision-making

Abstract

The application discloses a water surface garbage intercepting and cleaning ship based on YOLO visual prediction and motion cooperative decision, which comprises a visual perception subsystem, an electric control decision subsystem, a power execution subsystem, a garbage collection module and a remote monitoring unit which are carried on the water surface garbage intercepting and cleaning ship; through visual motion prediction function, ship-body-environment cooperative modeling and intercepting type decision, it is ensured that visual perception information and motion prediction data can be converted into accurate intercepting execution actions in real time and reliably, so that the whole process automation from "seeing" to "accurate intercepting and cleaning" is realized, the operation efficiency and success rate of flowing water area are improved, the energy consumption is reduced, and the real-time performance and reliability of the system are ensured.

Description

Technical Field

[0001] This invention relates to the field of intelligent robot control and machine vision technology, and is particularly applicable to surface debris interception and cleaning vessels based on YOLO visual prediction and motion cooperative decision-making. Background Technology

[0002] The current technical bottlenecks of surface garbage cleaning equipment are mainly reflected in three aspects: First, "perception" is limited to current state recognition. The vision systems of existing equipment can only obtain the current position of the garbage and cannot predict its trajectory. As a result, dynamic drifting garbage can only be tracked by following, which is inefficient and has a low success rate. Second, "decision-making" is disconnected from "environment-hull characteristics". The control system does not fully integrate the kinematic model of the hull and environmental factors such as water flow, making it difficult to generate accurate and forward-looking control commands. Third, the control system architecture is outdated. Most equipment uses single and coupled software logic, which is difficult to handle the concurrent needs of multiple tasks such as visual recognition, motion prediction, interception decision-making, and remote communication. The system has poor scalability and maintainability and cannot support complex collaborative algorithms.

[0003] Furthermore, existing solutions largely focus on the visual algorithms themselves or improvements to the mechanical structure, neglecting the collaborative design of perception prediction and kinematic models, as well as the optimization of the top-level architecture of the electronic control system, which is the core of the intelligent carrier. An intelligent system capable of predicting the movement of waste, integrating environmental and hull characteristics to make interception decisions, and efficiently allocating computing resources is key to solving the problem of waste cleanup efficiency in flowing waters, which is currently lacking in technology. Summary of the Invention

[0004] The purpose of this invention is to provide a surface garbage interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making, which can solve the problem of garbage cleaning efficiency in flowing water.

[0005] To achieve the above objectives, the YOLO-based visual prediction and motion cooperative decision-making surface garbage interception and cleaning vessel of the present invention includes a visual perception subsystem, an electronic control decision-making subsystem, a power execution subsystem, a garbage collection module, and a remote monitoring unit mounted on the surface garbage interception and cleaning vessel. The visual perception subsystem, based on the MaixCam embedded AI vision development kit, is equipped with a YOLOv5n or YOLOv8n target detection model specifically trained on a water surface debris dataset. It utilizes the Kendryte K510 AI chip built into the vision development kit for real-time accelerated inference of image data, accurately identifying debris categories and outputting the location coordinates and confidence level of the target debris in the image. Simultaneously, it calls a target tracking algorithm to analyze continuous image frames and predict the short-term motion vector of the target debris; the motion vector includes the direction and velocity of motion. The target debris identification results are sent to the electronic control decision subsystem via serial communication. The electronic control decision-making subsystem uses an STM32H723 microcontroller as the main control chip and runs a layered software architecture based on FreeRTOS. It integrates a waste movement prediction module, a hull-environment collaborative modeling module, an interception point calculation module, and a model prediction control module. This subsystem receives and processes data from the visual perception subsystem and environmental water flow information. Combined with the hull-environment collaborative modeling module and the interception point calculation module, it calculates the interception point where the hull and the target waste will meet at a future moment, generating forward-looking control commands. The electronic control decision-making subsystem has two operating modes: autonomous cruise and interception tracking, switching modes based on visual signals. When no target waste data is received, the cleaning vessel operates in autonomous cruise mode along a preset path. When it receives target waste data and the target waste's motion vector from the visual perception subsystem, it switches to interception tracking mode, driving the hull towards the interception point. The power execution subsystem includes two brushless DC motors, two electronic speed controllers (ESCs) that are paired with the brushless DC motors, and a propeller. The electronic control decision subsystem outputs two independent PWM signals to the ESCs to control the rotational speed and direction of the two brushless DC motors, thereby enabling the ship to propel, turn, and move differentially to perform cruise and interception missions. The remote monitoring unit is connected to the electronic control decision subsystem via a wireless module, receives status information from the electronic control decision subsystem, and sends working mode switching commands and autonomous cruise parameters to the electronic control decision subsystem.

[0006] Furthermore, the visual perception subsystem employs the MaixCam embedded AI vision development kit, wherein the image sensor is an OV9732 or a model with equivalent or higher performance, supporting image acquisition at a resolution of at least 1920x1080; when running the target detection model, the processing frame rate is no less than 25fps; the target tracking algorithm is KCF, SORT algorithm, or optical flow method; the serial communication adopts the standard UART protocol, with a baud rate of no less than 115200; the target garbage identification result includes target category, bounding box coordinates, confidence score, and motion vector information.

[0007] Furthermore, the working mode switching mechanism of the electronic control decision subsystem is as follows: In autonomous cruise mode, the main control chip generates a PWM control signal through a preset cruise algorithm; when the target garbage identification result is received from the visual perception subsystem via the serial port, the cruise task is interrupted; the interception point calculation module is called, combined with the hull-environment collaborative modeling module, to calculate the future intersection interception point between the cleaning vessel and the target garbage, and the PWM control quantity of the two brushless DC motors is calculated through the model prediction control module to drive the hull to move towards the interception point.

[0008] Furthermore, the electronic control decision subsystem also integrates an SBUS protocol parsing module for receiving control signals from the remote controller. The remote controller's control signals have the highest priority. When a valid remote controller control signal is received, the cleaning vessel will directly generate a corresponding PWM signal based on the remote controller control signal and output it to the power execution subsystem.

[0009] Furthermore, the software architecture of the electronic control decision subsystem is clearly divided into three layers: the application layer, the functional module layer, and the board-level support package. In the application layer, multiple concurrent tasks are created using FreeRTOS, including the cmd.c task for communication, the boat.c task for motor control, and the intercept.c task for intercept point calculation. The tasks interact with each other and transmit instructions through a message center designed in the style of ROS, thereby achieving decoupling and efficient communication.

[0010] Furthermore, the garbage collection module is controlled by the electronic control decision subsystem; when the cleaning vessel is in the interception and tracking mode and the distance between the vessel and the target garbage is determined to be less than a set threshold, the electronic control decision subsystem will activate the garbage collection device to complete the garbage collection.

[0011] Furthermore, the waste motion prediction module in the electronic control decision subsystem obtains the position sequence of waste in continuous images based on visual tracking, filters out noise using a state estimation algorithm, establishes a motion model, and predicts the position and speed of waste in the next few seconds.

[0012] Furthermore, the hull-environment collaborative modeling module in the electronic control decision subsystem includes a hull kinematic model and a water flow environment model; the hull kinematic model establishes the relationship between the hull's linear velocity, angular velocity and the rotational speeds of the left and right wheels through geometric relationships, describing the hull speed, steering angle and motor PWM control quantity; the water flow environment model is used to estimate the flow velocity and direction of the water flow, and correct the hull and garbage movement trajectory.

[0013] The advantages of this invention are that by using visual motion prediction function, ship-environment collaborative modeling and interception decision-making, it ensures that visual perception information and motion prediction data can be reliably and in real time converted into precise interception execution actions, thereby realizing full-process autonomy from "seeing" to "precise interception and clearing", improving the efficiency and success rate of operations in flowing water, reducing energy consumption, and ensuring the real-time performance and reliability of the system. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the functional modules in the cleaning vessel described in this invention. Detailed Implementation

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] like Figure 1 As shown, the surface garbage interception and cleaning vessel based on YOLO visual prediction and motion collaborative decision-making of the present invention includes a visual perception subsystem, an electronic control decision-making subsystem, a power execution subsystem, a garbage collection module, and a remote monitoring unit mounted on the surface garbage interception and cleaning vessel.

[0017] The visual perception subsystem is based on the MaixCam embedded AI vision kit, equipped with an optimized YOLOv5n or YOLOv8n target detection model. It utilizes the Kendryte K510 AI chip built into the vision development kit for real-time accelerated inference of image data. It is responsible for real-time capture of water surface images and identification and localization of debris targets, achieving accurate identification of debris categories and outputting the position coordinates and confidence level of the target debris in the image. The MaixCam image sensor is an OV9732 or equivalent or higher performance model, supporting image acquisition at least 1920x1080 resolution. Simultaneously, the visual perception subsystem calls a target tracking algorithm to analyze continuous image frames and predict the short-term motion vectors of the debris targets. The target tracking algorithm processes frames at a rate of no less than 25fps. Based on side-scan and bottom-scan images from LowranceHook (a consumer-grade fish finder / chartist), a reasonable simplified model of debris drift is established using fluid / flow direction and corresponding water level changes at the target location. Initial image data from SonarWiz7 (a professional-grade underwater sonar data processing and mapping software) and target tracking algorithms such as KCF (Kernelized Correlation Filters), SORT (Simple Online and Realtime Tracking), or optical flow are used to analyze the positional changes of target debris in consecutive image frames, predicting the short-term motion vector of the target debris, including its direction and velocity. A Kalman filter algorithm is introduced to smooth the motion direction and velocity, reducing prediction errors caused by water flow disturbances and image noise. Through the visual perception subsystem's recognition and object motion prediction mechanism, the target debris identification results, including debris type, location coordinates, confidence level, and motion vector, are transmitted in real-time via serial communication using a specific data protocol to the electronic control decision subsystem, providing a basis for path planning. The serial communication uses a standard UART-TTL serial port to communicate with the main controller, with a baud rate of no less than 115200.

[0018] The electronic control decision-making subsystem uses an STM32H723 microcontroller as its main control chip and runs a layered software architecture based on FreeRTOS. It integrates a waste movement prediction module, a hull-environment collaborative modeling module, an interception point calculation module, and a model prediction control module. This subsystem receives and processes data from the visual perception subsystem and environmental water flow information. Combined with the hull-environment collaborative modeling module and the interception point calculation module, it calculates the interception point where the hull and the target waste will meet at a future moment, generating forward-looking control commands. The electronic control decision-making subsystem has two operating modes: autonomous cruise and interception tracking, switching modes based on visual signals. When no target waste data is received, the system operates in autonomous cruise mode, navigating along a preset path. Upon receiving target waste data and its motion vector from the visual perception subsystem, it immediately switches to interception tracking mode, driving the hull towards the interception point.

[0019] The garbage motion prediction module uses visual tracking to obtain the position sequence of garbage in continuous images, a state estimation algorithm to filter out noise, establish a motion model, and predict the position and speed of garbage in the next few seconds.

[0020] The hull-environment co-modeling module includes a hull kinematics model and a water flow environment model. The hull kinematics model establishes the relationship between the hull's linear velocity, angular velocity, and the rotational speeds of the left and right wheels through geometric relationships, describing the hull speed, steering angle, and motor PWM control quantities. The water flow environment model is used to estimate the flow velocity and direction of the water flow and correct the hull and garbage movement trajectory.

[0021] The working mode switching mechanism of the electronic control decision subsystem is as follows: In autonomous cruise mode, the main control chip generates PWM control signals through a preset cruise algorithm; when the target garbage identification result is received from the visual perception subsystem via the serial port, the cruise task is interrupted; the interception point calculation module is called, and the future intersection interception point between the cleaning vessel and the target garbage is calculated by combining the hull kinematic model and water flow environment information; the PWM control quantity of the two brushless DC motors is calculated through the model predictive control algorithm, driving the hull to move towards the interception point.

[0022] The core of the interception point calculation module is to solve a spatiotemporal intersection problem. It is determined by the ship-environment collaborative modeling module based on the current location and predicted trajectory of the target waste, the current location, mobility, and water flow information of the cleaning vessel. Through an optimization algorithm, with the shortest intersection time as the objective, it calculates the spatial coordinates at a future point in time where the cleaning vessel can reach the waste, i.e., the interception point.

[0023] In each control cycle, the model predictive control algorithm predicts the ship's trajectory under different control inputs over a future period (prediction time domain) based on the current state (ship position, speed, etc.) and the ship-environment cooperative model (determined by the ship-environment cooperative modeling module). Then, by solving an optimization problem—typically including rapid approach to the interception point, control smoothing, and energy saving—the optimal set of future control sequences is selected from all possible sequences, and only the first control input (the next PWM instruction) is output to the actuators (two brushless DC motors). This process is repeated in the next cycle.

[0024] The software architecture of the electronic control decision subsystem is clearly divided into three layers: APP (application layer), Module (functional module layer), and BSP (board support package). The BSP layer has completed the driver encapsulation of all the underlying hardware of STM32H723 (such as PWM timer, UART, I / O, etc.), providing a unified and stable hardware operation interface for the upper layer.

[0025] The Module layer serves as a reusable functional library, containing independent functional modules such as SBUS protocol parsing, PID control algorithms, model predictive control algorithms, and communication protocol encapsulation. These modules exist in the form of .c and .h files, facilitating management and integration.

[0026] The application layer uses FreeRTOS to create multiple parallel tasks, including cmd.c for communication, boat.c for motor control, and intercept.c for intercept point calculation, enabling concurrent execution of functions. To achieve efficient and decoupled communication between tasks, the application layer designs a ROS-like Message_Center. The Message_Center acts as an information bus, responsible for message subscription, publishing, and routing. Tasks interact with each other and transmit commands through the ROS-like Message_Center, achieving decoupling and efficient communication.

[0027] The cmd.c task for communication acts as the command input hub, responsible for receiving serial port commands from MaixCam and SBUS commands from the model aircraft remote controller, parsing them, and then publishing them to Message_Center.

[0028] The intercept.c task, used for intercept point calculation, subscribes to visual prediction data and environmental water flow information, calls the hull-environment collaborative modeling module (integrating the hull kinematics model and the water flow model), calculates the future intercept point, and generates optimized control commands through the model predictive control algorithm, which are then published to Message_Center.

[0029] The boat.c task for motor control serves as a motion control task. It subscribes to messages related to interception control instructions, integrates internal control algorithms, calculates the final PWM control quantities acting on the left and right brushless motors, and outputs them to the electronic speed controller by calling the BSP layer interface, thereby enabling the hull to move precisely towards the interception point.

[0030] The power execution subsystem includes two brushless DC motors, two electronic speed controllers配套 with the brushless DC motors, and propellers; the electronic control decision-making subsystem controls the rotational speed and direction of the two brushless DC motors by outputting two independent PWM signals to the electronic speed controllers, realizes the propulsion, steering, and differential motion of the hull, drives the hull to move along the optimized path towards the interception point, and realizes the interception-type recovery of garbage to execute the cruise and interception tasks.

[0031] The electronic control decision-making subsystem is also integrated with an SBUS protocol parsing module for receiving control signals from the remote control; the control signals of the remote control have the highest priority. When a valid remote control signal is received, the cleaning ship will directly generate corresponding PWM signals according to the remote control signal and output them to the power execution subsystem.

[0032] The garbage collection module is controlled by the electronic control decision-making subsystem; when the cleaning ship is in the interception tracking working mode and determines that the distance between the hull and the target garbage is less than the set threshold, the electronic control decision-making subsystem will start the garbage collection device to complete garbage collection. The garbage collection module also includes anti-interference design.

[0033] The remote monitoring unit is connected to the electronic control decision-making subsystem through a wireless module (such as a 4G module or a WiFi module), receives the status information of the electronic control decision-making subsystem, and sends a working mode switching instruction and autonomous cruise parameters to the electronic control decision-making subsystem.

[0034] In the development of the electronic control decision-making subsystem in this invention: Use STM32CubeMX to initialize the H723 clock and peripherals to generate a FreeRTOS project. Complete the PWM and UART driver encapsulation at the BSP layer. Write modules such as SBUS parsing and message center at the Module layer. Create tasks at the APP layer and complete the core application logics such as cmd.c and boat.c, and perform inter-task communication through Message_Center.

[0035] Visual perception subsystem: In the MaixCam development environment, load the trained YOLO model in the.kmodel format onto the device, and write application programs for image acquisition, inference, motion prediction, and sending results through the serial port to ensure the accuracy and real-time performance of motion vector prediction.

[0036] Inter-system integration: First, ensure that MaixCam can stably send recognition results and motion prediction data to STM32; then debug Message_Center to ensure accurate message transmission between tasks; next, debug the interception point calculation and model prediction control modules and optimize algorithm parameters; finally, test the complete process of mode switching, garbage interception and collection in actual water areas (including flowing water areas), and optimize the control algorithm and system parameters based on the test results.

[0037] The actual work process is as follows: System power-on: All hardware initializes, FreeRTOS tasks begin scheduling, the system enters autonomous cruise mode by default, and conducts water area coverage patrols according to the preset path (combined with the three-dimensional terrain map generated by sonar).

[0038] Intelligent Sensing and Prediction: Based on the MaixCam-based OV9732 image sensor and the Lowrance sonar device used to assist in transmitting information about the ship's surroundings, the system continuously scans the water surface and acquires images in real time. Object recognition is performed using a pre-set database of water object models, and object movement paths (motion vectors) are predicted using a simplified debris drift model and target tracking algorithm. The relevant data is then sent to the electronic control decision subsystem.

[0039] Interception Decision and Execution: After receiving the data, the electronic control decision subsystem immediately switches to the interception and tracking mode. It integrates water flow information and ship kinematics model through the hull-environment collaborative modeling module to calculate the interception point. Then, it generates control commands through model predictive control algorithm to drive the ship to the interception point along the optimized path.

[0040] Waste collection: When the system determines that the distance between the ship and the waste is less than a set threshold, it triggers the waste collection device to start and complete the waste collection; after the collection is completed, the system automatically returns to the autonomous cruise mode and continues to patrol the waters.

[0041] Safety and Monitoring: Operators can view the system's operating mode, motor speed, identification results, predicted trajectory of waste movement, and interception points in real time through the remote monitoring unit. They can also take over control at any time through the SBUS remote control to ensure operational safety.

[0042] The present invention has the following beneficial effects: 1. Interception-based operations significantly improve efficiency and success rate: By using visual motion prediction and ship-environment collaborative modeling, the system can proactively intercept dynamic drifting debris, replacing the traditional tailing and tracking method. This greatly improves the success rate and operational efficiency of debris cleanup in flowing waters and reduces ineffective navigation energy consumption.

[0043] 2. True mechatronics and software integration: Through deep collaboration between YOLO vision prediction, STM32H723 electronic control decision-making and power execution, kinematic models and environmental information are integrated to achieve an intelligent closed loop from environmental perception and prediction to motion control, with fast decision response speed and high interception accuracy.

[0044] 3. Advanced system architecture and strong scalability: The system adopts a three-layer architecture of APP / Module / BSP and a message mechanism similar to ROS, which results in low coupling between system modules. New functions (such as new prediction algorithms and sensor data fusion modules) can be easily added in the form of modules, which greatly improves the maintainability and scalability of the system.

[0045] 4. High reliability: FreeRTOS ensures real-time concurrent execution of multiple tasks; the SBUS manual priority mechanism provides reliable safety redundancy; the layered design isolates underlying hardware failures without affecting upper-layer logic; and a robust resource management and exception handling mechanism ensures stable system operation.

Claims

1. A surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making, characterized in that: It includes a visual perception subsystem, an electronic control decision-making subsystem, a power execution subsystem, a waste collection module, and a remote monitoring unit mounted on the surface waste interception and cleaning vessel; The visual perception subsystem, based on the MaixCam embedded AI vision development kit, is equipped with a YOLOv5n or YOLOv8n target detection model specifically trained on a water surface debris dataset. It utilizes the Kendryte K510 AI chip built into the vision development kit for real-time accelerated inference of image data, accurately identifying debris categories and outputting the location coordinates and confidence level of the target debris in the image. Simultaneously, it calls a target tracking algorithm to analyze continuous image frames and predict the short-term motion vector of the target debris; the motion vector includes the direction and velocity of motion. The target debris identification results are sent to the electronic control decision subsystem via serial communication. The electronic control decision subsystem uses an STM32H723 microcontroller as the main control chip and runs a layered software architecture based on FreeRTOS. It has built-in garbage movement prediction module, hull-environment collaborative modeling module, interception point calculation module, and model prediction control module. It is used to receive and process data from the visual perception subsystem and environmental water flow information. Combined with the hull-environment collaborative modeling module and the interception point calculation module, it calculates the interception point where the hull and the target garbage will meet at a certain time in the future and generates forward-looking control commands. The electronic control decision-making subsystem has two working modes: autonomous cruise and interception and tracking. It switches between working modes based on visual signals. When no target garbage data is received, the cleaning vessel operates in autonomous cruise mode according to a preset path. When the target debris data and the motion vector of the target debris are received from the visual perception subsystem, the system switches to interception and tracking mode and drives the ship to move toward the interception point. The power execution subsystem includes two brushless DC motors, two electronic speed controllers (ESCs) that are paired with the brushless DC motors, and a propeller. The electronic control decision subsystem outputs two independent PWM signals to the ESCs to control the rotational speed and direction of the two brushless DC motors, thereby enabling the ship to propel, turn, and move differentially to perform cruise and interception missions. The remote monitoring unit is connected to the electronic control decision subsystem via a wireless module, receives status information from the electronic control decision subsystem, and sends working mode switching commands and autonomous cruise parameters to the electronic control decision subsystem.

2. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making as described in claim 1, characterized in that: The visual perception subsystem uses the MaixCam embedded AI vision development kit, with an OV9732 image sensor or a model of equivalent or higher performance, supporting image acquisition at a resolution of at least 1920x1080; the processing frame rate is no less than 25fps when running the target detection model; the target tracking algorithm is KCF, SORT algorithm, or optical flow method; the serial communication uses the standard UART protocol with a baud rate of no less than 115200; the target garbage identification result includes target category, bounding box coordinates, confidence score, and motion vector information.

3. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making according to claim 1, characterized in that: The working mode switching mechanism of the electronic control decision subsystem is as follows: In autonomous cruise mode, the main control chip generates a PWM control signal through a preset cruise algorithm; when the target garbage identification result is received from the visual perception subsystem via the serial port, the cruise task is interrupted. The interception point calculation module is invoked, and combined with the hull-environment collaborative modeling module, the future intersection and interception point between the cleaning vessel and the target waste is calculated. The PWM control quantities of the two brushless DC motors are then calculated through the model prediction control module to drive the hull to move towards the interception point.

4. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making as described in claim 1 or 3, characterized in that: The electronic control decision subsystem also integrates an SBUS protocol parsing module for receiving control signals from the remote controller. The remote controller's control signals have the highest priority. When a valid remote controller control signal is received, the cleaning vessel will directly generate a corresponding PWM signal based on the remote controller control signal and output it to the power execution subsystem.

5. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making according to claim 1, characterized in that: The software architecture of the electronic control decision subsystem is clearly divided into three layers: application layer, functional module layer, and board-level support package. In the application layer, multiple concurrent tasks are created using FreeRTOS, including cmd.c task for communication, boat.c task for motor control, and intercept.c task for intercept point calculation. The tasks interact with each other and transmit instructions through a message center designed in the style of ROS, so as to achieve decoupling and efficient communication.

6. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making according to claim 1, characterized in that: The garbage collection module is controlled by the electronic control decision subsystem. When the cleaning vessel is in the interception and tracking mode and the distance between the vessel and the target garbage is determined to be less than a set threshold, the electronic control decision subsystem will activate the garbage collection device to complete the garbage collection.

7. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making according to claim 1, characterized in that: The waste motion prediction module in the electronic control decision subsystem acquires the position sequence of waste in continuous images based on visual tracking, uses a state estimation algorithm to filter out noise, establishes a motion model, and predicts the position and speed of waste in the next few seconds.

8. The surface debris interception and cleaning vessel based on YOLO visual prediction and motion cooperative decision-making according to claim 1, characterized in that: The ship-environment collaborative modeling module in the electronic control decision subsystem includes a ship kinematics model and a water flow environment model. The ship kinematics model establishes the relationship between the ship's linear velocity, angular velocity and the rotational speed of the left and right wheels through geometric relationships, and describes the ship's speed, steering angle and motor PWM control quantity. The water flow environment model is used to estimate the flow velocity and direction of the water flow and correct the trajectory of the ship and the garbage.

Images