Research On Multi-target Tracking And Real-time Navigation Planning Technology Of Unmanned Surface Vehicle Based On Edge Computing

Unmanned Surface Vehicle,as a new type of unmanned platform on the sea,has great application value in the vast ocean.And it is the future development trend that unmanned surface vehicles can assist workers to conduct unmanned operation in the ocean.As one of the key technologies of unmanned surface vehicle,target detection tracking and real-time navigation planning technology is not only detect obstacles and track to predict the next track,but also it can according to the tracking estimate autonomous reasoning,replanning the current navigation path,improving obstacle avoidance capability of unmanned surface vehicle.However,unmanned surface vehicle face various uncertainty tests when it navigate autonomously in a complex ocean environment,The computing resources of the traditional unmanned surface vehicle system may be difficult to meet the demand of the calculation,which may lead to cannot real-time replanning of navigation path.By the title of “Research on Multi-target Tracking and Real-time Navigation Planning Technology of Unmanned Surface Vehicle Based on Edge Computing”,this paper analyzed the functional requirements of the real-time navigation planning system of unmanned surface vehicle,aiming at the problems of insufficient terminal calculation and poor real-time performance,the research is based Multi-target tracking and real-time navigation planning method for unmanned surface vehicle based on edge computing.First,designing the framework of real-time navigation planning system of unmanned surface vehicle based on edge computing;and then proposing a multi-target detection and tracking method for unmanned surface vehicle combined with YOLOv4-UKF and a real-time navigation planning method for unmanned surface vehicle by integrated with D* Lite.The feasibility of the method is verified by simulation and real ship experiments,which has high value in academic research and engineering application.The main works of this paper are illustrated as follows:(1)According to the functional requirements of the unmanned surface vehicle real-time navigation planning system,the framework of the unmanned surface vehicle real-time navigation planning system based on the edge computing is designed.The whole system is divided into the unmanned surface vehicle terminal,the edge terminal and the cloud,the unmanned surface vehicle terminal provides data collection,the edge terminal runs the algorithm proposed in this article,and the cloud is responsible for the real-time update of the algorithm,and the selection and design of the hardware and software of the system.Introduces the method of calculation offloading and storage optimization for the multi-sensor data of unmanned surface vehicle,propose the Heuristic Particle Swarm Optimization algorithm to complete the data offload between the unmanned surface vehicle and the edge platform,and use the HDFS storage optimization method to meet the multi-source data of the unmanned surface vehicle storage requirements.(2)By analyzing the current target detection and tracking methods,and propose a multitarget detection and tracking method for unmanned surface vehicle combined with YOLOv4-UKF,which uses the YOLOv4 algorithm to complete the multi-target detection and recognition of the unmanned surface vehicle,and the Hungarian algorithm is used to correlate and match the detection frame with the tracking frame,and the unscented Kalman filtering method is used to predict the state position of each target to complete multi-target tracking.(3)The map of the surrounding environment of the unmanned surface vehicle is constructed based on the Costmap grid method,and the information of obstacles predicted by tracking is written into the map,and perform expansion processing on each dynamic obstacle.Combine the international maritime collision avoidance rules to classify the obstacle collision risk level,evaluate the collision hazard of unmanned boats in four encounter situations;the improved D* Lite algorithm is used to replan the collision danger zone path for the quasidangerous zone and the dangerous zone,and complete the real-time path planning of the unmanned surface vehicle.(4)Based on the yolov4 algorithm to detect and identify the obstacles in the video taken by the unmanned surface vehicle,the effectiveness of the algorithm is verified.Based on the EKF algorithm and the UKF algorithm,the simulation experiment of target prediction and tracking is carried out,and the superiority of the UKF algorithm is verified by comparing the experimental results.By using MATLAB the platform conducts simulation experiments on the real-time planning method of the unmanned surface vehicle under single dynamic and multidynamic obstacle scenarios to verify the feasibility of the algorithm.Build an unmanned surface vehicle experimental platform for the proposed multi-target tracking and real-time unmanned surface vehicle based on edge computing the voyage planning method conducts real boat experiments to verify the safety and effectiveness of the method.