Research On Water Surface Target Detection And Obstacle Avoidance Technology Of Unmanned Vessels In Inner River Based On Machine Vision

Path planning affects the success of autonomous navigation of unmanned craft.Among the numerous global path planning algorithms for unmanned craft,the A~*algorithm,as a heuristic search method,is fast in search and can plan a collision-free path from the starting point to the endpoint when the way exists.Unlike other local path planning algorithms,the dynamic window method considers the mechanical properties of UVs,and the planned path is closer to the actual route of UVs.In path planning,only the distance information can be detected,while the data cannot obtain the type of information when radar can detect obstacles.As a result,the collision avoidance strategy cannot implement in a targeted way.However,obstacle target detection can precisely solve this problem.Therefore,this paper focuses on optimizing the A~*algorithm and dynamic window method and through the integration of the two algorithms to achieve global real-time obstacle avoidance;The obstacle detection of the crewless boat in the inner river is studied.Specific research contents are as follows:(1)To address the problem of sensors such as laser radar being unable to detect target class information,this paper proposes an inland waterway surface obstacle detection system based on the YOLOv5 algorithm.Firstly,we establish an image data set of inland waterway surface obstacles required for this paper.Secondly,we input the data set into a neural network for training and build the network model.Finally,we test the trained network model in two modes:image and video.The test results show that the inland waterway surface obstacle detection model based on the YOLOv5 algorithm established by this paper can accurately classify inland waterway obstacles,and the target position detection accuracy is ideal.(2)In terms of global path planning of unmanned craft,the search domain of the A~*algorithm is improved from the 8*8 field to the 5*5 area given the low search efficiency of the A~*algorithm;To solve the problem that A~*algorithm has many redundant nodes,a screening critical path node function introduces to remove the redundant nodes,which makes the path of unmanned craft smoother.The evaluation function of the A~*algorithm is optimized,and the weight of H(x)is changed by adding the ratio of the distance between the current point and the target point and the distance between the starting point and the target point.The simulation results show that with the increase in map size,the improved A~*algorithm has some advantages compared with the traditional algorithm in search time,path length,and turning Angle.(3)In the area of local path planning for unmanned surface vessels,the conventional dynamic window approach struggles to differentiate between the nearest obstacle distances for static and dynamic obstacles.As such,this paper proposes a solution that relies on separate obstacle minimum distance evaluation functions for dynamic and static obstacles.In addition to this,separate conflict radii are established for each respective obstacle type.The proposed method includes increasing the weight parameter of the nearest distance evaluation function for dynamic obstacles,therefore increasing its contribution to the overall score.Ultimately,this aims to improve the vessel’s ability to avoid dynamic obstacles,whilst guiding the vessel towards less hazardous static obstacles in environments that include mixed levels of both types of obstacle.The simulation results suggest that,in such mixed static and dynamic environments,the proposed enhanced dynamic window approach outperforms the traditional dynamic window approach.(4)In response to the limitation of path planning with A~*algorithm only for known obstacles,failing to handle unknown and dynamic obstacles,this paper proposes a fusion algorithm for real-time global path planning by combining local real-time path planning with global offline path planning,aiming to achieve global real-time obstacle avoidance for unmanned vessels.Simulations and practical ship experiments were conducted to evaluate the proposed fusion algorithm.The simulation results demonstrate that the fusion algorithm has the ability to achieve global real-time obstacle avoidance,maintains a safe distance with dynamic obstacles,and satisfies the requirement of global real-time obstacle avoidance.The practical ship experiment shows that the fusion algorithm can effectively avoid unknown obstacles,proving its feasibility.The research achievement of this paper provides important support for the practical application of unmanned vessels and has certain theoretical significance.