Vision-based Unmanned Surface Vehicles Positioning For Inland Waters

High-accuracy positioning technology,especially when the satellite is not available,is considered as one of the ten key engineering problems,and also the core issues to realize the intelligence of vehicle Unmanned Surface Vehicle(USV).In order to realize vision-based USV positioning for inland waters,this dissertation has combined the onboard vision and information fusion technology to realize the USV position estimation,including relative positioning and absolute positioning technology.Relative positioning is achieved by visual odometry(VO)method,which requires stable and reliable visual features.Therefore,it is necessary to extract the stable area of the image to obtain stable features;Absolute positioning is realized by building visual map first and then positioning,in which only high-precision position information is needed in the process of mapping,and visual sensor and ordinary GPS are needed for positioning.In detail,several works are described as follows.(1)In order to achieve the extraction of the stable area of the onboard visual image,and then obtain the stable and reliable visual features,this article takes the WaterShore-Line(WSL)as the research object and proposes some WSL detection algorithms.First of all,CHR algorithm(Canny,Hough and RANSAC combined algorithm)can be used for straight WSL detection,which has high computational complexity because of the Hough transform;Secondly,LR algorithm(LSD and RANSAC combined algorithm)is proposed which has linear complexity,but it is also powerless for nonstraight WSL detection;Finally,in view of the shortcomings of the above algorithm,WSL detection algorithm based on image sequence is proposed,which constructs onshore line segment pool by the coarse-to-fine strategy,and then generates non-water area and WSL.Field experiments prove the method meets the actual scenarios demand.(2)In order to realize the USV relative positioning for inland waters,Water-VO method based on the stable feature points is proposed which is realized by the visual odometry(VO).Firstly,the commonly VO method cannot be initialized in the water environment.After analyzing the reason,it is found that the feature points in the water area of the image have a short survival time,low accuracy and large re-projection errors,which are collectively referred to as the instability characteristics of water feature points;Secondly,the performance of multiple feature extraction and matching algorithms is compared through the number of feature points matching and the running time.On this basis,this article specially designed a water environment visual odometry method(water-VO),which mainly includes key frame selection,system initialization and relative motion estimation;Finally,Field experiment results show the method can not only solve the problem of initialization failure caused by unstable visual features in the water environment,but also can improve the accuracy of USV positioning and motion trajectory estimation by removing unstable features.(3)Aiming at the demand for high-precision absolute positioning,this article proposes positioning method through a boat-mounted camera and an ordinary GPS module.This method introduces the visual map technology and adopts a multi-level positioning strategy.Firstly,the visual map had been generated,which is composed of equidistant nodes that contain corresponding DGPS information and image-related information;Secondly,in the positioning stage,multi-level positioning method is applied by GPS information and image information;Finally,the performance of the algorithm is verified through actual experiments,and improvement strategies are proposed for corresponding problems.The results show that the method only requires a common camera and GPS module,which can minimize its positioning cost while ensuring positioning accuracy.(4)When GPS information is missing,this article first analyzes the USV positioning process and principle,which can replace the GPS information to obtain the coarse positioning results,supplement and improve the multi-level positioning strategy of USV,and form a complete positioning method based on pure vision;Then,in order to solve the problem of inconsistent positioning solutions in different cameras,the federated kalman filter framework is introduced to fuse the localization process into a continuous and stable solution;Finally,it is verified through experiments that this method can replace GPS positioning to a certain extent and is a reliable supplement in the positioning process of USV.The method proposed in this article constitutes a vision-based USV positioning for inland waters,which can effectively meet the high-precision positioning needs.The relevant results have been verified by field experiments,showing that this method effectiveness and robustness.This research work is of great significance to the "unmanned" and intelligentization of USV.