Research On Multi-target Tracking Algorithm Of Millimeter Wave Surveillance Radar In Traffic Scene

Millimeter wave radar is an important part of the current intelligent traffic monitoring system.The use of millimeter wave radar for multi-target vehicle tracking is a research hotspot in the field of intelligent transportation.For passing vehicles on the road,the millimeter-wave radar monitoring system uses target tracking and trajectory prediction to determine whether the vehicle has dangerous driving behavior and then issues an early warning signal.Therefore,radar monitoring and early warning can help reduce traffic accidents.At present,there are many multi-target tracking algorithms based on video images,but in actual traffic monitoring environments,these methods have poor adaptability and are easily affected by complex environments,and millimeter-wave radar has high detection accuracy and strong anti-interference ability,which can be well adapt to complex scenes.This paper studies and applies the clustering algorithm and data association algorithm in the multi-target tracking technology of millimeter-wave monitoring radar in traffic scenes,and designs a multi-target tracking system according to the actual traffic monitoring scenes.The specific work is summarized as follows:(1)A spindle-based density peak fuzzy clustering algorithm is proposed.The algorithm combines the search for the density peak and the iterative update of the spindle,and uses the secondary clustering to modify the results of the initial clustering,thereby improving the accuracy of the clustering.In addition,the algorithm finds the density peak and the cluster center accurately,which improves the efficiency of secondary cluster correction and shortens the running time of the algorithm.The test results of actual scenes show that,compared with the DBSCAN clustering,fuzzy C-means clustering and k-Means clustering algorithm,the clustering algorithm improves the accuracy of the classification results by more than 15%,and solves the problem of inaccurate clustering of adjacent vehicles.(2)A k-nearest neighbor joint probability data association algorithm is proposed.Based on the clustering results,the algorithm improves the traditional joint probabilistic data association algorithm,proposes new pre-aggregation matrix and aggregation matrix,and improves the accuracy of data association and the efficiency of algorithm running.Simulation comparison experiments and the test results of actual scenes show that,compared with the nearest neighbor data association and joint probabilistic data association algorithm,the association accuracy of the data association algorithm can reach more than 95%,which solves the problem of inaccurate multi-target tracking.(3)Designed a multi-target tracking system.In view of the problem of inaccurate tracking of adjacent vehicles in highway scenes,the system combines the above clustering and data association algorithms to design a multi-target tracking algorithm,which solves the problems of indistinguishable targets and inaccurate data association when tracking multitargets of adjacent vehicles.The test results of actual scenes show that the success rate of multi-target tracking is over 95%,and it is robust to multiple application scenes.