Research On Autonomous Dynamic Obstacle Avoidance Technology Of Unmanned Surface Vehicles Based On AVOD

With the development of the national marine development strategy,unmanned surface vehicles are gradually being developed and applied due to their flexibility,safety and other advantages.In the complex ocean traffic environment,unmanned surface vehicles often face the risk of collision with dynamic obstacles.Improving the autonomous and dynamic obstacle avoidance capability of unmanned surface vehicles is the prerequisite for ensuring that unmanned surface vehicles can safely complete various tasks at sea.With the topic of "Research on Autonomous Dynamic Obstacle Avoidance Technology of Unmanned Surface Vehicles Based on AVOD ",this paper analyzes the autonomous dynamic obstacle avoidance requirements of unmanned surface vehicles,builds the software framework and hardware framework of autonomous dynamic obstacle avoidance system for unmanned surface vehicles,and proposes a multi-sensor information fusion sensing method for 3D target detection and a dynamic obstacle avoidance method combined with track prediction,and verify the effectiveness of the algorithm.The main work of the thesis includes:(1)Analyze the related problems in the process of dynamic obstacle avoidance for unmanned surface vehicles,determine the steps for the realization of the dynamic obstacle avoidance function,and select the sensors required for information collection and the key technologies for obstacle avoidance system according to the dynamic obstacle avoidance requirements.On this basis,the hardware framework and software framework of the autonomous dynamic obstacle avoidance system for unmanned surface vehicles are built.(2)In order to obtain the relevant information of dynamic obstacles from the unknown environment,a multi-sensor information fusion perception system based on AVOD is built,and the detection of dynamic obstacles is realized through the AVOD network,so as to obtain a directional three-dimensional bounding box surrounding the dynamic obstacles;The output of the 3D target detection network is used as the input of the 3D target tracking module,and the motion tracking of dynamic obstacles is realized through state prediction and data association,so as to obtain the motion information of the dynamic obstacles.(3)Build a dynamic obstacle trajectory prediction network based on RNN-LSTM,use the trajectory prediction network to predict the navigation trajectory of the dynamic obstacle,and then combine the trajectory prediction result of the dynamic obstacle with the perception result as the input of the speed obstacle method.On this basis,a set of feasible collision avoidance speeds is constructed,and the collision risk,optimal collision avoidance speed and safe return point during dynamic obstacle avoidance are calculated to realize safe collision avoidance during local path cruise.(4)Collect multi-sensor information and establish data sets required for target detection and track prediction,and train the target detection network and track prediction network.Propose corresponding performance evaluation indicators to evaluate the target detection performance,target tracking performance and track prediction performance of the obstacle avoidance system.Set a variety of dynamic collision avoidance scenes and initial motion parameters,and conduct real ship collision avoidance experiment in actual sea environments.The experimental results show that the autonomous dynamic collision avoidance system proposed in this paper can detect and track moving targets accurately 100 meters away during obstacle avoidance,and unmanned surface vehicles can successfully bypass the dynamic obstacles to reach the preset local target points in a variety of collision avoidance scenarios,which verifies the reliability and effectiveness of the autonomous dynamic obstacle avoidance technology proposed in this paper.