The Study Of Monocular Vision Based Surface Target Detection And Dynamic Obstacle Avoidance Methods For Unmanned Surface Vessels

At present,with the support of the country for the development of unmanned surface vessel(USV)and the continuous popularization of the USV in the fields of industry and military,the USV is gradually developing towards intelligence and autonomy.As the key technology of realize the autonomous navigation of the USV,the performance of path planning directly affects the safety of unmanned navigation.Considering the actual navigation environment of the USV,this thesis mainly deals with the target detection and autonomous obstacle avoidance in the USV system.In the aspect of autonomous obstacle avoidance,it is carried out in the order of research of “first global and then partial,first static and dynamic”.The main research contents and results are as follows:Firstly,Obstacle target detection is the premise of path planning.On the basis of analyzing the dynamic obstacles that may be encountered during the navigation of the USV,this paper focuses on the research of detection methods for surface ship targets.A fine-grained ship detection system is established,which is based on YOLO(You Only Look Once)deep convolutional neural network.Due to the need of network model training,a ship image dataset is established.Simulation results verify that the detection system can be effectively applied to ship fine-grained target detection,which has a high detection rate and can identify the detected ship.And the detection rate reaches 27 frames per second.At the same time,this paper completed the video detection experiment in the waters of the Erqi Yangtze River Bridge.Experiment results show that the algorithm can effectively resist the interference caused by visibility,waves and background.Besides,it accurately detect the position and category of ships in the Yangtze River waters.The video detection speed reaches 10 frames per sec,which basically meet the real-time requirements of maritime target detection.Secondly,according to the multi-objective problems such as energy consumption,safety and smoothness of the USV path planning,a global path planner is proposed on improved particle swarm optimization algorithm.The full name of the algorithm is adaptive hybrid particle swarm optimization algorithm(AHPSO).It combines the global and local particle swarm optimization algorithms.Besides,it sets the adaptive strategy which can adaptively adjust the coefficient of inertia weight and acceleration.Simulation results verify the effectiveness of the AHPSO algorithm applied to multi-objective path planning.The relative optimality of multiple targets such as energy consumption,path and rudder angle can be achieved.More importantly,improved method significantly speeds up the operation of the algorithm.Finally,considering the local dynamic obstacle avoidance mode in the USV system,a local path planner is designed for the local path planning of the USV.The ship’s close encounter area is proposed which is combined with the international collision avoidance rules.The enhanced artificial potential field algorithm(EAPF)is obtained which is improved on the traditional artificial potential field algorithm(APF).Simulation results show that EAPF algorithm has good performance in calculating obstacle avoidance path.Under the situation of encounter and cross encounter,the USV can realize the local collision avoidance operation with the dynamic ship.Besides,the obstacle avoidance effect is obtained under the condition of environmental force.