Machine Vision-based Docking And Recovery System For Unmanned Marine Vehicles

With the innovation of industrial informatization and the sweep of intelligent tide in recent years,the development of Autonomous Underwater Vehicle(AUV)and Unmanned Surface Vehicle(USV)has received great attention and is moving towards the direction of intelligence.Such unmanned Marine vehicles are characterized by high degree of automation,good flexibility and long operation time.They can replace humans to perform many time-consuming,repetitive and dangerous tasks in specific or large areas of sea,such as regional search,Marine mapping,environmental assessment,regional security and so on.The rapid Lanuch and Recovery(LAR)system for AUV and USV,as an important link in the research of unmanned Marine vehicles,are related to the efficiency,cost and safety of the whole system.Therefore,it is of great value to study the autonomous docking and recovery technology of this kind of intelligent unmanned Marine vehicles.Based on the background of AUV and USV recovery,this paper takes the autonomous docking task as the specific research objective.The solution and system implementation of the autonomous docking task of the unmanned Marine vehicles based on machine vision are proposed.Combined with specific platforms,surface docking verification experiments and underwater autonomous docking verification experiments are carried out,which provides a reference for the research on autonomous docking of unmanned Marine vehicles.Firstly,aiming at the core goal of autonomous docking based on visual guidance,a general architecture of docking visual servo system is proposed based on the introduction of docking system overview.The principle of camera model and image preprocessing technology are introduced around the preparatory knowledge needed to realize the docking task.For the foggy scene faced by the surface docking task,an improved defogging method based on dark channel prior and transmittance compensation is proposed to realize the automatic restoration of foggy images.Secondly,by analyzing the different characteristics of surface and underwater docking scenes and the different requirements for object detection,different technologies solutions are adopted to detect the docking targets.Aiming at underwater docking,a real-time detection method combining Histogram of Oriented Gradient(HOG)and Support Vector Machine(SVM)are proposed for underwater docking station.Aiming at the surface docking with more complex background,an improved surface docking target detection method based on YOLO and color threshold segmentation is proposed,which greatly improves the accuracy of object detection.Then,on the basis of the docking target detection,combined with the different characteristics of the docking targets,the respective pose estimation methods are studied.The distorted image can be corrected before pose estimation by camera calibration.For underwater docking,the extraction method of underwater light spots and the consistency matching strategy of light spots are designed.Under the condition of 2D-3D point pairs matching correctly,the relative pose between the docking station and AUV is solved according to EPn P algorithm.According to the accurate object detection results and the monocular camera imaging model,a fast positioning method for the docking target on the water surface is proposed.Finally,relevant experiments are designed to verify the results of this paper.Visual servo docking control software systems are developed for different experimental platforms before the test.The underwater autonomous docking experiments based on machine vision are carried out in the towing pool,which verify the effectiveness and reliability of the system.The object detection,positioning accuracy and the integrated docking experiments are carried out in dynamic ocean environment to verify the effectiveness and robusness of the surface docking and recovery system.