Research On Multi-Source Data Perception And Track Fusion For Inland River Ships

In order to solve the problems of the large number of inland ships,the complex navigation scenarios,and the difficulty in ship monitoring and perception,the research on the intelligent perception of ship situation was carried out.Taking multi-sensor data fusion as the starting point,a multi-source data acquisition and fusion system was designed and developed.The measured data has carried out in-depth research on ship target detection and tracking,track correlation,data fusion estimation and other methods,and carried out experimental verification in the Wuhan section of the Yangtze River and Fuzhou Mawei Port.The specific studies are as follows:First,a ship target detection and tracking algorithm is proposed,which can detect and track the targets of three sensors of AIS,radar and video,analyze and subcontract the AIS data,and perform AIS data de-noising,AIS trajectory prediction and navigation trajectory subcontracting.and other operations;the radar image data is preprocessed,the connected domain detection is performed after the shoreline is removed to obtain the ship target,and the Kalman filter and the Hungarian matching method are used for target tracking;the video adopts the ship target detection algorithm based on YOLOv5,and the image data is collected for training images.The detection model and tracking mainly use the method of improving Kalman filtering and Hungarian matching combined with image HOG features.The above algorithm is verified by collecting the actual data of ships sailing in the Yangtze River section.Secondly,a multi-source data track association algorithm is proposed.First,in order to make each sensor data in the same coordinate system,time-space matching is performed on the track data from different sensors,and the difference value on the time line is performed on each data to unify the data frequency,and study the spatial coordinate conversion between each data;secondly,the track correlation algorithm of the nearest neighbor method,the fuzzy track correlation algorithm,and the evidence theory track correlation algorithm are introduced.Based on this,a multi-source data estimation fusion algorithm is proposed.First,the data fusion structure and the characteristics of each sensor data are studied,and the distributed fusion framework and the redundancy of the three sensor data of AIS,radar and camera are determined.Secondly,the convex combination fusion and covariance cross are The fusion is introduced,and the simulation experiments are carried out to compare the two algorithms.Finally,a real ship experiment is designed to verify the effectiveness of the fusion algorithm by comparing the real data of ferry navigation collected by the high-precision differential base station with the algorithm fusion data.Finally,the multi-source data fusion system is constructed,the hardware system and software platform are integrated,the multi-source data fusion application of inland river ship track is studied,the real-time monitoring of restricted waters and the application of historical track analysis are realized,and the application of the system in practical scenarios is verified.In this paper,the multi-source data fusion method of inland river ship track is adopted to make the correct correlation rate of track reach 89% and the target tracking accuracy reach more than 70%.The degree of data fusion accuracy is as 10 °.The research in this paper can realize real-time monitoring of water areas,monitor the navigation status of passing ships,reduce the occurrence of safety accidents,and ensure the navigation safety of ships.The multi-source data fusion system built in inland rivers also has a good application prospect.