Research On Obstacle Perception And Path Planning Of Unmanned Surface Vehicles

The ability of obstacles perception and collision avoidance path planning for an unmanned surface vehicle(USV)is the key to realize intelligent autonomous navigation.Due to the limitations of environment perception of onboard sensors and uncertainty of dynamic obstacles in complex unknown waters,many unsolved problems need to be tackled in the field of USV obstacles perception and path planning.Among those problems,this dissertation mainly focuses on four issues: 1)accurate detection and recognition of obstacles in abnormal sight scenes,2)reliable positioning and tracking of obstacles outside the visual range,3)efficient mapping in the cluttered environment,and 4)dynamic local path planning under multiple constraints.This dissertation investigates the methods of the USV obstacles detection and recognition,positioning and tracking,obstacles modeling,and dynamic path planning respectively.The main contents and innovations are as follows:Aiming at solving the problem of detection and recognition of obstacles in abnormal sight scenarios,a detection and recognition method based on an improved Faster R-CNN model is proposed.The proposed method adds a multi-scale dilated convolution module to the Res Net50 feature extraction network.To improve the detection and recognition accuracy of small-scale obstacles such as barge ships and floating objects,the dilated convolution operation is applied.The operation can increase the receptive field without changing the spatial resolution of the feature map,and thus further optimizing the number of anchor frames and matching methods.The obstacles detection and recognition system based on the camera is established for USV.The experiments is carried out to verify the proposed method.The results show a good result of fine classification of obstacles in abnormal sight scenarios such as obstacle occlusion and frame tilt.For positioning and tracking obstacles outside the visual range,a AIS data processing method based on Fermat’s curve fitting is proposed.The prior knowledge of obstacle movement is incorporated to mark,correct and fill in abnormal or missing AIS data.Then,an adaptive noise reduction algorithm is designed to extract the inherent trend characteristics of the components.Fermat’s curve is applied to solve the intermediate tangent point between the straight and curved navigation trajectory,and the fitted trajectory is obtained by calculating its mirrored curve.A remote-controlled test boat is used to simulate obstacles in 6 navigation states.The test results show that the proposed method can achieve the positioning and tracking of the dynamic obstacles.To construct mapping model in the cluttered environment,an obstacle modeling method using satellite images based on an efficient convex hulls algorithm is proposed.To extract the obstacle features,fuzzy mean clustering and masked method are used to divide the image of large-scale sea and land.Meanwhile,block parallel computing is applied to enhance edge details,which is conducive to segmenting features of the offshore areas by a dual-threshold method.Then,mathematical morphology operations are applied to fill the internal holes of obstacles and make the contours move out appropriately.An efficient algorithm is designed to mask the obstacles into convex hulls by using the operations of marking,coding,clustering,the determination of the least squares moment axis and edge points.Based on the mapping model,a reference global path for USV is obtained using the particle swarm optimization from the port of Xiaocu to the port of Meizhou.To optimize the local path under the multiple constraints,a hierarchical obstacle avoidance and recovery path planning method is proposed for the waterjet-propelled USV.A hybrid A* algorithm with motion primitive constraints is designed to obtain an initial path with node optimization and continuous curvature.According to different types of dynamic obstacles,the International Regulations for Preventing Collisions at Sea(COLREGs)rules are integrated,and a local threat map based on the Apollonius circle is constructed to avoid these obstacles without a certain motion status.Combined with the characteristics of the waterjet-propelled USV,which can move both forwards and backwards,the Reeds-Shepp curve is used to calculate the autonomous recovery path.Taking the cluttered Meizhou Bay as the application,the simulation demonstrates that our method can effectively avoid various obstacles and thus help achieve autonomous recovery.This dissertation studies the obstacle perception and path planning technologies for USV,and proposes obstacle detection and recognition,positioning and tracking,mapping methods for the complex sea area and obstacle avoidance path optimization strategies under the multiple constraints.Related achievements can provide references for the intelligent and autonomous vehicles,and technical supports for the design and manufacture of marine intelligent ships.