Multi-sensor Fusion For Unmanned Surface Vehicle Target Awareness Technology

In recent years,with the rapid development of unmanned surface vehicle technology,the functions of unmanned surface vehicle are becoming more and more powerful,such as hydrological data collection,transportation of goods,coastal defense patrol,reconnaissance and surveillance.The realization of these functions is inseparable from the cooperation of multi-sensor,such as when to complete the task of monitoring surface targets,you need to use inertial navigation to obtain the position information of your own boat in the global coordinate system,and you also need to use navigation radar or lidar sensors to obtain the position information of the surface target relative to the unmanned surface vehicle,with the help of the above two kinds of information,the position information of the surface target under the global coordinate system can be obtained to complete the monitoring task.At present,almost all unmanned surface vehicle are equipped with multiple sensors,so the target sensing technology for multi-sensor fusion of unmanned surface vehicle is a research hotspot in the field of unmanned surface vehicle technology and it has strong engineering research value.The target sensing technology for multi-sensor fusion of unmanned surface vehicle studied in this paper mainly involves the detection and tracking of surface targets,recognition and classification,and information fusion.The technology is mainly composed of three modules,namely detection and tracking module,recognition and classification module and fusion module.The detection and tracking module is responsible for detecting and tracking the water surface target,obtaining the position information of the water surface target.The recognition and classification module uses the position information of the water surface target to identify and classify the water surface target.The fusion module is mainly responsible for integrating the detection and tracking information of the target with the recognition and classification information,and forwarding them to the control center.The key technologies involved in this paper mainly include three parts: water surface target detection and tracking,target recognition and classification,and target trajectory prediction.The main research contents are as follows:Aiming at water surface target detection and tracking,this paper proposes a target detection and tracking algorithm based on two-stage DBSCAN clustering,which mainly includes two stages of detection and tracking.In the detection stage,two-stage DBSCAN clustering is proposed.First,large targets are clustered,and then small targets are clustered.In each stage,the size of clustering parameters is set according to the size of clustering targets.In the tracking phase,the Hungarian algorithm mainly uses the distance between two consecutive images to achieve the tracking effect.This method solves the problem of noise,complex target environment and other factors that cause a large target to be imaged into several small targets,which affects the subsequent tracking.It can be applied to complex water environment.Aiming at target recognition and classification,this paper proposes a target recognition and classification algorithm based on Yolo v5 s network structure and Anchor Free.Due to the size of water surface targets in the image varies greatly,and there may be many targets to be recognized and classified.Therefore,the Yolo V5 s network structure is adopted,which can solve the problems of multi-scale and classification and recognition speed.In addition,on the premise of ensuring the accuracy,Yolo v5 s network structure is further optimized,and Anchor Free is introduced,which reduces the size of the classification model and improves the classification and recognition speed.The model can be easily deployed to the recognition and classification module without worrying about the limitations of memory and classification speed.Aiming at the target trajectory prediction,this paper uses the method of combining Kalman filter and velocity information to predict the target trajectory.If only Kalman filter is used,large prediction error will be produced in the front of prediction.If only velocity information is used,large prediction error will be produced because of the large change of target velocity direction.Therefore,the combination of the two methods has a good trajectory prediction effect.