Abstract: In the road traffic system, millimeter-wave radar has become a popular sensor for target motion information acquisition due to its high resolution and strong anti-interference ability. In the case of loss of radar observation information, the traditional target tracking algorithms will have a large tracking error or cannot carry out target tracking. To address this problem, a radar target tracking algorithm based on long short-term memory (LSTM) network is proposed in this paper, which uses the memory function of the LSTM network to train and predict the radar observations when the radar observation information is available. When the radar observation data are lost, the LSTM network is used to provide the predicted observation values for the extended Kalman algorithm, so as to ensure that the extended Kalman algorithm can continue tracking the target and achieve the purpose of reducing the target tracking error. The LSTM network is trained by radar measured data, and the simulation verification and analysis are carried out for the linear and curved motion states.The simulation results show that the proposed target tracking algorithm can still track the target when the radar observation data are lost, and the error of the target tracking algorithm is effectively reduced.