Target Tracking Method For Unmanned Surface Vehicle Based On Sensor Fusion

Today,the Unmanned Surface Vehicle(USV)is considered as intelligent surface vehicle for many tasks,such as environment monitory,people search and rescue in open water.It is also widely used for military applications,for instance,coastal defense,reconnaissance and target attack,all have crucial values.As the eyes and ears of the USV,environment perception technology with highly intelligence is the key.Therefore,research on it are meaningful.In this thesis,based on the "WAM-V" USV platform,a deep learning long-term object tracking method and a high-precision target tracking method based on information fusion by Lidar and camera are studied.The research work can be mainly divided into several partsFirstly,it briefly summarizes some relevant research of the USV and USV environment perception technology.Secondly,through the building,training and testing processes of two deep learning networks,the CNN target classification and detection methods are discussed,and their superiorities are proved.Furthermore,the problem of long-term object tracking in the water surface scene is studied,combining a correlation filter tracker and an offline trained deep learning detecter.Besides,a lost object recovery mechanism is proposed.Moreover,a Lidar point cloud filtering and clustering method for surface environment is studied,and a calibration method between monocular camera and Lidar is proposed,which ultimately forms up a surface target tracking system based on information fusion.Finally,in the virtual and real environment,the method studied in this thesis is evaluated,and the results are satisfactory.