Research On Target Tracking Based On Unmanned Surface Vehicles Marine Radar And Visual Image

With the development of artificial intelligence,computer technology and sensor technology,autonomous unmanned system has made great progress in recent years.The key technologies applied in marine autonomous unmanned system represented by unmanned surface vehicles,such as environment perception,data fusion,scene understanding and so on,are becoming a research hotspot.The ability to detect and track the targets in the complex and changeable navigation environment is the premise for the unmanned surface vehicle to complete its mission independently.However,the unmanned surface vehicle has the characteristics of small volume,fast speed and easy to be disturbed,which brings great challenges to the perception technology of the unmanned surface vehicle.This paper aims at joint perception of the marine radar and visual sensor carried by the unmanned surface vehicle.For the problem of cooperative target detection and tracking of marine radar and visual image,in this paper,a target detection and tracking system is designed based on the cooperation of marine radar and visual sensor by using image processing,adaptive Kalman filter,YOLOv3 target detection algorithm and visual target matching strategy.The effectiveness of the system is verified by the real ship experiment.The main research contents are as follows:First,aiming at the problem of target detection of unmanned surface vehicle marine radar,the image processing method is applied to analyze the marine radar image,analyze the characteristics of the target in the radar image,separate the target and the background,and then use the characteristics of the target to distinguish different targets,extract the target information and realize the target detection.For the problem that the target may be lost in the continuous marine radar image,an optimized adaptive Kalman filter method is utilized in this paper,which processes the variance of maneuvering acceleration,and introduces a strong tracking filter.From the simulation experiments and real ship experiments,it is concluded that the adaptive method used in this paper has better prediction effect than the original method,and it is more suitable to estimate the target position by using the predicted value when the target is lost,so as to realize the target detection and tracking in the smaller gate.Then,for the problem of water surface target detection and tracking of vision sensor,the ship target detection is realized by using YOLOv3 target detection algorithm.And the location loss in the YOLOv3 algorithm is optimized to the GIoU function to solve the problem that the location loss of the original YOLOv3 algorithm is the same,but the IoU is different.From the experimental results,the detection accuracy of the optimized loss function is improved.Aiming at the problem of visual image target tracking,the target moving feature data is obtained by target detection algorithm,and the target tracking is realized by Kalman filter algorithm.In the case of multiple objects in the visual image,a matching strategy of visual objects is adopted to ensure the correctness of object tracking,and the effectiveness of the strategy is proved by experiments.Finally,aiming at the problem of vision sensor’s limited field of vision,on the basis of adaptive Kalman filter,YOLOv3 target detection algorithm and visual target matching method for radar and visual image processing,in this paper,the strategy of using radar to find and locate the target is designed,so as to guide the visual system to track the target.The real ship experiment shows that the strategy can solve the problem of limited vision of vision sensor.In this paper,a water surface target detection and tracking system is designed based on the images of marine radar and vision sensor,which lays a foundation for the environment perception technology of unmanned surface vehicles.