Research On Tracking Algorithm Of Multi-target Data Association Fusion

With the rapid development of modern science and technology,the level of modern warfare has been continuously improved,which has caused a great threat to the detection and tracking of single-station radar systems.The multi-station radar system expands the range of target detection and ensures the combat capability of the entire system.For the multistation radar multi-target tracking problem in complex environments,it is necessary to consider how to select appropriate key technologies such as data fusion and data association algorithms to obtain the best tracking results.Under the centralized fusion structure,this paper mainly studies the multi-radar multi-target point trace fusion and data association problems.The main work of this paper is as follows:(1)The multi-radar measurement data fusion problem under the centralized fusion structure is studied.Aiming at the shortcomings of the centralized weighted fusion algorithm,based on the multi-radar centralized fusion structure,a multi-radar data fusion method based on convolutional neural network is proposed.The method utilizes the advantages of convolutional neural network in extracting spatial features,extracts the error features between each radar measurement,and then uses the powerful function approximation ability of the fully connected layer to estimate the fusion point trace of multiple radar point traces.Through the simulation experiments,the following conclusions are drawn:(1)Compared with the traditional single radar multi-target tracking,the multi-radar fusion algorithm has better tracking performance,and the data fusion algorithm based on the convolutional neural network proposed in this paper has better tracking performance in the same scene.It is better than the centralized weighted fusion algorithm,which verifies the effectiveness of the method in this paper;(2)When the radar with higher fusion accuracy is added to the fusion system,the fusion performance of the method in this paper is also improved to a certain extent,which shows the adaptability of the method in this paper.(2)The multi-object data association problem in the cluttered trace environment is studied.In view of the shortcomings of traditional data association algorithms,this paper regards the data association problem as a classification problem,and proposes a data association algorithm based on convolutional long short-term memory network.Radar measurement data has both spatial and time-series characteristics.Using traditional time-series neural network can only extract the time characteristics of the data,but cannot use the spatial characteristics of the data,and the effect of the prediction model is not ideal.However,convolutional long-short-term memory networks can not only deal with timing problems,but also take advantage of the spatial characteristics of data.Therefore,this paper uses the convolutional long and short-term memory network to construct a prediction method for the correlation of radar measurement data.This method uses the spatial and temporal characteristics of different targets to associate and match the radar measurement data,and can output one target and multiple radars each time.The measured association probability is used for subsequent target tracking filtering processing.The following conclusions are drawn through simulation experiments:(1)In the case of high clutter density,the nearest neighbor method has incorrect association.The data association algorithm based on convolutional long-term memory network proposed in this paper is more efficient than the joint probability data interconnection algorithm.The excellent performance verifies the effectiveness of the method in this paper;(2)In the scenario of high target density,the running time of the joint probability data interconnection algorithm increases significantly,while the time of the method in this paper is relatively stable and lower than the joint probability data interconnection algorithm.In addition,the root mean square error of the data association algorithm based on convolutional long short-term memory network proposed in this paper is lower than that of the joint probability data interconnection algorithm,which shows the effectiveness and real-time performance of the method in this paper.