Research On Key Technologies Of Multi-Target Tracking For Millimeter Wave Radar In Short-Range

With the rapid development of autonomous driving technology in recent years,millimeter-wave radar has become an indispensable vehicle-mounted sensor for autonomous driving due to its advantages of penetrating smog,high resolution,and low cost.In terms of algorithm empowerment,multi-target tracking technology based on millimeter-wave radar plays an important role in vehicle obstacle detection and avoidance,adaptive cruise,and other functions.However,in short-range scenarios with multiple targets,the effect of radar tracking algorithms is not ideal due to factors such as the point cloud shape,the amount of clutter,and the maneuverability of the target.To improve the multi-target tracking performance of millimeter-wave radar in short-range scenarios,this thesis takes the radar multi-target tracking algorithm as the main line and conducts the following research:(1)Research on radar point cloud clustering and track initiation.Firstly,discuss and experiment on point cloud data screening.Then,aiming at the problem that the traditional clustering algorithm is difficult to distinguish the long-distance target and the clutter due to that the number of target measurement points is many in the near but few in the far,a segmented DBSCAN clustering method based on multi-frame accumulation is proposed.And the improved algorithm is proved to be effective by experiments.finally,the commonly used track initiation algorithm such as intuitive method,logic method and Hough transform method are studied and analyzed,and Monte Carlo simulation experiments are carried out,choose logic method as the track initiation algorithm.(2)According to the characteristics of strong mobility of the target in the road environment,the problem of state filtering and data association in multi-target tracking is studied.First,analyze and simulate the state prediction model,compare the prediction effect of the Constant Velocity(CV)model and the Interacting Multiple Model(IMM)in the target multi-maneuvering state,and the advantages of the IMM are verified.Then research and simulate state filtering algorithms such as Kalman Filter(KF),Extended Kalman Filter(EKF)and Unscented Kalman Filter(UKF),and compare the filtering results,choose EKF as the filtering algorithm.Finally,the Nearest Neighbor Data Association(NNDA)and Probabilistic Data Association(PDA)algorithms are analyzed and simulated,and the Joint Probabilistic Data Association(JPDA)algorithm is studied.An improved double-gate JPDA algorithm is proposed to solve the problems of increased calculation amount and trajectory convergence in the multi-target track crossing scene,and the effectiveness of the improved algorithm is verified by comparison and simulation.(3)According to the characteristics of the road scene,combined with the previous simulation and improvement,a double-gate IMM-JPDA-EKF multi-target tracking algorithm based on multi-frame clustering is proposed and simulated.The algorithm uses multi-frame clustering and double tracking gates to limit the number of measurements in the wave gate to reduce computational overhead.The IMM model is used to predict the target state,and the EKF is used to update the trajectory,which improves the tracking effect on maneuvering targets.Finally,experiments were carried out in fixed scenes and actual road scenes to verify the effectiveness of the improved algorithm.