The autonomous decision-making layer needs to predict the driving behavior of surrounding road traffic participants when making behavioral decisions. The core of behavior prediction is to track road traffic participants between different frames, and how to realize road traffic participant data between different frames. Matching the association becomes a difficult point of trajectory tracking; the trajectory tracking effect will directly determine the accuracy of the road traffic participant's attitude information acquisition, and thus determine the driving behavior prediction result.   In this paper, the millimeter-wave radar sensor is used to detect road vehicles. The data has certain accuracy error. In the process of road vehicle trajectory tracking, firstly, the vehicle model is kinematically modeled to obtain the attitude change of the vehicle between different frames. The amount of millimeter-wave radar data is compensated for different frames. Secondly, the target data is kinematically modeled according to the velocity information in the millimeter-wave radar data, and the Hungarian algorithm is used to perform different inter-frame matching associations, using an extended Kalman filter pair. The target data pose information is corrected. Finally, according to the matching association result, the historical trajectory of the road vehicle is smoothed by the second exponential smoothing of the variable parameters, so as to obtain the optimal trajectory tracking effect.  The actual vehicle experiment shows that the proposed algorithm can effectively track the road traffic participation, and the track is smooth and the real-time performance is good. Keywords: trajectory tracking, matching association, attitude information, kinematics modeling, filter

CICTP