关键词: 轨迹预测；全局规划；简单循环神经网络；最大似然估计

Abstract: In order to improve the ability of obstacle detouring in intelligent driving field, a kind of real-time method is proposed to predict the trajectory of obstacles. The method integrates vision sensors and a lidar sensor to observe the historical motion trajectory of the obstacle to give out the pose and speed of the obstacle with time stamp, which mean that the obstacle is described from three dimensions: time, space and speed. Based on historical observation data, simple recurrent neural network with maximum likelihood estimation can predict the trajectory of obstacles in real time.

Key words: Trajectory prediction; Global planning; Simple recurrent neural network; Maximum likelihood estimation