Research On Obstacle Avoidance Of Unmanned Boat Based On Visual SLAM

Unman Surface Vehicle(USV)is an important tool for water surface and underwater exploration.Its autonomous and intelligent research has always been one of the research hotspots of unmanned platforms.As an important index of intelligent and autonomous unmanned boat,autonomous navigation has high requirements for the perception and processing ability of environmental information.It is necessary to conduct in-depth research on the positioning and obstacle avoidance technology of unmanned boat.In this thesis,Simultaneous Localization And Mapping(SLAM)technology is introduced into the research of unmanned boats,so that unmanned boats can use sensors such as cameras to achieve simultaneous positioning,water surface map construction and obstacle avoidance.The full text mainly completes the following work :(1)Aiming at the problem that it is difficult to effectively extract feature points in the image due to the flow state of the water surface,this thesis designs a fast segmentation algorithm based on the H component in the HSV(Hue Saturation Value)color space to segment the shore and water areas in the image.At the same time,the saliency detection of the water surface object is carried out,and the water surface obstacle is marked by the minimum external matrix.Only the feature point detection algorithm is used for the shore area and the water surface marker,which effectively reduces the interference of the dynamic area of the water surface,reduces the time consumption of the SLAM system to extract the feature points,and improves the accuracy of the pose estimation.(2)To solve the positioning problem of the area without feature points in the navigation of the USV,the extended Kalman filter(EKF)method is used to fuse the positioning data of IMU,GPS and VO as the front end of the visual SLAM,and the ORB-SLAM2 algorithm is improved.At the same time,the extraction method of ORB feature points is optimized,and the real-time performance of collecting feature points is improved.(3)The obstacle avoidance problem of USV can be attributed to the path planning problem and the local path adjustment problem in the known environment,and the obstacle avoidance method is studied.In this thesis,aiming at the underactuated and large hysteresis of the USV,the angle between the heading and the current point to the target point is introduced,and the value of the angle is calculated in real time.According to the range of the value,it is determined whether it is necessary to abandon the three neighborhoods opposite to the heading,so as to realize the optimization of the traditional global planning A * algorithm,which effectively saves the wayfinding time.In addition,the local obstacle avoidance algorithm TEB is studied to achieve local obstacle avoidance.(4)Using NVIDIA Jeston nano as the calculation board,GD32 as the control board,and other modules such as power supply and drive to complete the design of the USV controller.Using PC as the host computer,the USV controller and the host computer use WIFI communication to build a USV hardware experimental platform system based on visual SLAM.The water surface mapping and obstacle avoidance experiments are completed in a river channel.The experimental results verify the effectiveness of the visual SLAM and obstacle avoidance algorithms studied in this thesis.