Water Surface Target Detection Based On Computer Vision And Unmanned Ship Platform Development

China is a country with particularly abundant water resources,with a long coastline and numerous rivers and lakes.Water park,water cruises and boat trips have been deeply integrated into the daily life of the public.However,on the water surface,once the hull rolls over,hits the reef and other uncontrollable factors cause the ship to sink and other safety accidents,it will cause very serious consequences.Therefore,after the ship’s safety accident caused the sinking,a fast,effective and accurate search method for people who fell into the water is proposed,which has important research significance for reducing casualties and property losses.This thesis focuses on the detection of surface targets and the research and development of unmanned ship systems.A target detection algorithm based on Faster Region with CNN feature(Faster-RCNN)and Single Shot Multi Box Detector(SSD)models is proposed.At the same time,a set of unmanned ship system composed of remote monitoring system,local base station and unmanned ship is developed.The algorithm is realized by engineering method,and the fast search and location of water surface target is realized.Aiming at the problem of hull sinking and the difficulty of quickly searching and locating the implementation personnel after the ship sinks,a fast surface detection algorithm based on the fusion of Faster-RCNN and SSD models is proposed.The Faster-RCNN algorithm and the SSD algorithm are two of the most popular algorithms in the field of computer vision target detection at present,and the two algorithms are selected to train the two algorithms separately.Among them,the Singapore Ocean Data Set provides ten types of objects common on the water surface,such as ferries,buoys,ships,speed boats,canoes,sailboats,people on the surface of the water,airplanes,and other objects.After the model training is completed,the Kalman filter algorithm is used to fuse and correct the target detection frames predicted by the two models,realizing the problem of real-time fast search and positioning of the surface targets,and solving the slow speeds caused by the two single models.The problem of low accuracy,while improving the stability and anti-interference of the algorithm.Aiming at the current slow water surface search,most of which use fixed cameras,this paper researches and develops a set of unmanned ship systems consisting of remote terminals,local base stations and unmanned ships.The remote terminal system is designed based on the B/S architecture,and the local base station is designed based on the C/S architecture.The database is designed accordingly,including functions such as target detection,historical data playback,one-click return and fixed-point cruise.This system breaks through the limitation of the shore-based mode of the traditional unmanned ship system.Based on the traditional mode,a remote terminal is introduced to facilitate the staff to remotely view the current hull position and search status in real time..Finally,the unmanned ship system developed in this paper is tested in the field.The tests show that each module of the unmanned ship system works normally,the communication is stable,and the accuracy and speed of surface target detection are verified.