Pedestrian Trajectory Prediction Using LSTM Regression Model in Automotive Radar Systems

In this study, we propose a two-step method to enhance pedestrian detection and driver safety in automotive radar systems. First, we assumed information regarding the range, velocity, and angle of pedestrians behind stationary vehicles using a frequency-modulated continuous wave radar system, with the cell-averaging constant false alarm rate algorithm employed to effectively distinguish between the target signals and noise components. Additionally, a multiple signal classification algorithm was used for the high-resolution angle estimation of the target. Subsequently, a long short-term memory network is applied to predict the movement of a pedestrian, with the results indicating an average error of 0.09 m between the predicted and actual points along the trajectory.